U.S. patent application number 11/757112 was filed with the patent office on 2008-04-10 for methods and apparatus for scheduling, serving, receiving media-on-demand for clients, servers arranged according to constraints on resources.
This patent application is currently assigned to Digital Fountain, Inc.. Invention is credited to Gavin Horn, Per Knudsgaard, Soren Lassen, Michael G. Luby, Jens Rasmussen.
Application Number | 20080086751 11/757112 |
Document ID | / |
Family ID | 26944093 |
Filed Date | 2008-04-10 |
United States Patent
Application |
20080086751 |
Kind Code |
A1 |
Horn; Gavin ; et
al. |
April 10, 2008 |
METHODS AND APPARATUS FOR SCHEDULING, SERVING, RECEIVING
MEDIA-ON-DEMAND FOR CLIENTS, SERVERS ARRANGED ACCORDING TO
CONSTRAINTS ON RESOURCES
Abstract
A media object is scheduled for transmission between a server
and a client. The media object is partitioned into segments of
blocks, wherein each block is a unit of media for which a client
will wait to receive an entire block before playing out the block,
and wherein each segment includes an integer number of blocks. One
or more channels on which to serve each segment are determined, and
a rate at which to serve each segment is determined. Additionally,
a schedule pair for each channel is determined. The schedule pair
includes a time at which the client may start receiving on the
channel and a time at which the client may stop receiving on the
channel.
Inventors: |
Horn; Gavin; (Emeryville,
CA) ; Luby; Michael G.; (Berkeley, CA) ;
Rasmussen; Jens; (Alameda, CA) ; Knudsgaard; Per;
(Oakland, CA) ; Lassen; Soren; (San Francisco,
CA) |
Correspondence
Address: |
TOWNSEND AND TOWNSEND AND CREW, LLP
TWO EMBARCADERO CENTER
EIGHTH FLOOR
SAN FRANCISCO
CA
94111-3834
US
|
Assignee: |
Digital Fountain, Inc.
Fremont
CA
94538
|
Family ID: |
26944093 |
Appl. No.: |
11/757112 |
Filed: |
June 1, 2007 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
09768843 |
Jan 23, 2001 |
7240358 |
|
|
11757112 |
Jun 1, 2007 |
|
|
|
60254514 |
Dec 8, 2000 |
|
|
|
Current U.S.
Class: |
725/87 ;
348/E7.071 |
Current CPC
Class: |
H04N 21/2387 20130101;
H04N 21/2401 20130101; H04N 21/8456 20130101; H04N 21/6405
20130101; H04N 7/17318 20130101; H04N 21/23106 20130101; H04N
21/6408 20130101; H04N 21/23406 20130101; H04N 21/4331 20130101;
H04L 12/1881 20130101 |
Class at
Publication: |
725/087 |
International
Class: |
H04N 7/173 20060101
H04N007/173 |
Claims
1. A method of scheduling a media object for transmission between a
server and a client, the method comprising: partitioning the media
object into segments of blocks, wherein each block is a unit of
media for which the client will wait to receive an entire block
before playing out the block, and wherein each segment includes an
integer number of blocks; determining one or more channels on which
to serve each segment, the channels capable of carrying data
between the server and the client; determining a rate at which to
serve each segment; and determining a schedule pair for each
channel, the schedule pair including a time at which the client may
start receiving on the channel and a time at which the client may
stop receiving on the channel.
Description
CROSS REFERENCES TO RELATED APPLICATIONS
[0001] The present invention claims is a continuation of U.S.
application Ser. No. 09/768,843 filed Jan. 23, 2001 entitled
"Method and Apparatus for Scheduling, Serving, Receiving
Media-on-Demand for Clients, Servers Arranged According to
Constraints on Resources," which claims priority from U.S.
Provisional Application No. 60/254,514 (Atty. Docket No.
019186-002900US), filed Dec. 8, 2000 and entitled "Method For
Media-On-Demand For Clients And Servers With Constrained
Resources," the entire disclosure of which is herein incorporated
by reference for all purposes.
FIELD OF THE INVENTION
[0002] The present invention relates generally to media-on-demand
systems, wherein a large number of clients might make requests of
the system for a particular piece of media content that is served
in real-time, or near real-time, to the clients. In particular, the
present invention relates to optimized media-on-demand systems
including systems using chain reaction codes.
BACKGROUND OF THE INVENTION
[0003] "Media-on-demand" ("MOD") is a term often used to refer to a
client-server system, where many clients make requests from a
choice of many possible media objects that are to be served by a
server (or array of servers) to the clients. In general, a design
goal of a MOD system is to allow a client to request a media object
or stream and have the content play at the client with no
interruptions. The on-demand implies that there is little to no
delay experienced by the client before the media object starts
playing out at the client. On-demand also implies that the media
object starts playing out from the beginning or possibly some
specified point in the content, as opposed to joining a
transmission in progress. One example is a video-on-demand system,
where large numbers of clients may each make requests among many
different digital videos.
[0004] A digital cable broadcasting system is a digital video
server, as it serves up many digital video streams to many clients
(end-user televisions, for example). Such a system might support
hundreds of independent video streams and millions of clients. The
system can be easily built, however customers are constrained to
receive only the pre-selected set of videos at the times selected
by the system operator, so the system is not on-demand.
[0005] A true media-on-demand system is much more difficult to
build because the system may have to serve as many different
streams as there are active clients, as each client demands
different content (media objects) or the same content at different
times. Potentially each client will require an independent stream
of the media object it requested, where a stream is the flow of
data required by that client in order to play out the content from
the beginning without interruptions.
[0006] Some work relating to MOD systems proposes using broadcast
or multicast mechanisms in order for the MOD system to be scalable
to a large number of clients. See for example, J. W. Wong,
"Broadcast delivery", Proceedings of the IEEE, 76(12):1566-1577,
(December 1988).
[0007] The different multicast or broadcast strategies that have
been proposed in the literature can be divided into two distinct
classes: (1) user-centered and (2) data-centered. See, e.g., S.
Viswanathan and T. Imielinski, "Metropolitan area video-on-demand
services using pyramid broadcasting", Multimedia Systems,
4(4):197-208 (August 1996) (hereinafter "Viswanathan"). In
user-centered strategies, the server bandwidth is allocated
according to client requests, i.e., the bandwidth assigned at the
server to serve a particular media object can vary over time
depending on how many clients are requesting that media object. In
data-centered strategies, the server bandwidth is allocated among
the different media objects, i.e., the bandwidth assigned at the
server to serve a particular media object is constant over
time.
[0008] A typical user-centered strategy is "batching", wherein all
clients that make a request during an interval for the same media
object are all serviced by one stream. There are a number of
batching schemes that consider the various possible scheduling
policies for assigning the available server bandwidth to a
particular media object. See, e.g., C. Aggarwal, J. Wolf, and P.
Yu, "On optimal batching policies for video-on-demand storage
servers", Proc. Intl. Conf. on Multimedia Computing and Systems,
pp. 253-258, Hiroshima, Japan, (June 1996) (hereinafter
"Aggarwal"); K. C. Almeroth and M. H. Ammar, "The use of multicast
delivery to provide a scalable and interactive video-on-demand
service", IEEE Journal on Selected Areas in Communication,
14(6):1110-1122, (August 1996); A. Dan, D. Sitaram, and P.
Shahabuddin, "Scheduling policies for an on-demand video server
with batching", Proc. ACM Multimedia, pp. 391-398 (October
1998).
[0009] Another user-centered approach is stream merging, where a
client receives data from multiple streams simultaneously, and the
extra streams are dropped once the client catches up to the next
existing stream. See, e.g., A. Bar-Noy and R. E. Ladner,
"Competitive on-line stream merging algorithms for
media-on-demand", Draft (July 2000); A. Bar-Noy and R. E. Ladner,
"Efficient algorithms for optimal stream merging for
media-on-demand", Draft (August 2000); D. Eager, M. Vernon, and J.
Zahorjan, "Minimizing bandwidth requirements for on-demand data
delivery", Proc. Intl. Workshop on Advances in Multimedia
Information Systems, pages 80-87, (Indian Wells, Calif., October
1999) (hereinafter "EagerMIS99"); and D. Eager, M. Vernon, and J.
Zahorjan, "Optimal and efficient merging schedules for
video-on-demand servers", Proc. ACM Multimedia, vol. 7, pages
199-203 (1999). For user-centered strategies, the server bandwidth
requirement for a particular media object can be expected to grow
with the frequency of user requests for that media object. This may
be acceptable for a small number of users, but may be infeasible if
the number of users grows very large for very popular content.
[0010] Data-centered strategies are scalable to potentially
millions of users as, unlike user-centered strategies, the server
bandwidth required to serve a single media object is independent of
the number of user requests or the frequency of user requests. A
simple data-centered strategy is to divide the available bandwidth
for a media object equally among C channels, and to retransmit the
media object over one of the channels at equally spaced time
intervals. For example, some pay-per-view digital satellite systems
show the same one and half to two hour movie on four different
channels, where a new transmission starts every half an hour. In
this case, the worst case startup latency is just less than half an
hour, where the startup latency, Ts, is defined to be the amount of
time that passes between when the client requests the media object
and the media object commences playing out on the client's media
object player. The startup latency, Ts, may include delays such as
processing the client's request at the server, propagation delay in
the network, and decoding and encoding delays at the client and
server respectively. Halving the startup latency for a media object
requires doubling the number of channels and therefore the server
bandwidth requirement also doubles.
[0011] Viswanathan describes "pyramid broadcasting", which is an
early data-centered protocol that greatly reduces startup latency
and only requires a server bandwidth that is logarithmic in the
length of the content instead of linear. A media object is
partitioned into segments, where the segment size may vary. Each
segment is transmitted repeatedly in a looping fashion at the same
rate as the other segments, where a different channel may be used
for each segment. Many other similar schemes have also been
proposed to reduce the client startup latency or the maximum client
temporary storage requirement for pyramid broadcasting, where the
temporary storage requirement is the storage the client needs to
save the data for the media object that has been downloaded at the
client and not played out. If the data is saved at the client, then
the storage needed is at least the size of the media object. See,
e.g., C. Aggarwal, J. Wolf, and P. Yu, "A permutation-based pyramid
broadcasting scheme for video-on-demand systems", Proc. IEEE Int'l
Conf. on Multimedia Systems, (Hiroshima, Japan, June 1996); K. Hua
and S. Sheu, "Skyscraper broadcasting: A new broadcasting system
for metropolitan video-on-demand systems", Proc. ACM SIGCOMM, pp.
89-100 (Cannes, France, 1997); and L. Gao, J. Kurose, and D.
Towsley, "Efficient schemes for broadcasting popular videos", Proc.
Inter. Workshop on Network and Operating System Support for Digital
Audio and Video, (July 1998). A characteristic of all these schemes
is that they require that data within each segment be downloaded in
order. A client therefore would have to wait for the beginning of
each segment to be transmitted before beginning to download or play
out the content from that segment.
[0012] U.S. Pat. No. 6,018,359 issued to Kermode (hereinafter
"Kermode") describes a generalized scheme wherein a client can
download multiple segments simultaneously. Kermode proposes
downloading the segments starting from any point and reordering the
data at the client, which allows the clients to download the
segments asynchronously. The Kermode scheme and the other
data-centered schemes with varying segment sizes described above,
impose a restriction that each segment be downloaded at a rate that
is greater than or equal to the play out rate of the content.
Another restriction is that every segment is served at the same
rate.
[0013] A second class of data-centered protocols fixes the size of
each segment to be equal, but varies the rate that each segment is
transmitted. See, e.g., L. Juhn and L. Tseng, "Harmonic
broadcasting for video-on-demand service", IEEE Trans. on
Broadcasting, 43:268-271 (September 1997) (hereinafter "Juhn97");
L. Juhn and L. Tseng, "Adaptive fast data broadcasting scheme for
video-on-demand service", IEEE Trans. on Broadcasting, 44:182-185
(June 1998); J.-F Paris, S. W. Carter, and D. D. E. Long,
"Efficient broadcasting protocols for video on demand",
International Symposium on Modeling, Analysis and Simulation of
Computer and Telecommunication Systems (MASCOTS), vol. 6, pages
127-132 (July 1998); J.-F Paris, S. W. Carter, and D. D. E. Long,
"A low bandwidth broadcasting protocol for video on demand", Proc.
International Conference on Computer Communications and Networks,
vol. 7, pages 690-697 (October 1998). Junh97 proposed an early
scheme of this type, referred to as "harmonic broadcasting".
Harmonic broadcasting and its variants require that the client be
able to download a media object at a rate that is at least as large
as the server bandwidth assigned to the media object, which grows
logarithmically with the content length.
[0014] Another restriction of all the above media-on-demand schemes
is that if any of the data is lost, then they either don't play
back in full fidelity, or else they have to wait for a complete
cycle of that segment before the missing data piece is re-sent.
Either scenario is sub-optimal. As described in U.S. Pat. No.
______ (U.S. patent application Ser. No. 09/246,015, filed Feb. 5,
1999 and entitled "Information Additive Code Generator And Decoder
For Communication Systems") (hereinafter "Luby I") and U.S. Pat.
No. ______ (U.S. patent application Ser. No. 09/399,201, filed Sep.
17, 1999 and entitled "Information Additive Group Code Generator
And Decoder For Communication Systems" (hereinafter "Luby II"),
chain reaction coding is a useful method of recovering from missing
data in many communications systems. Luby I and Luby II describe
the application of chain reaction codes for content download, and
not for an on-demand media streaming application. In some
implementations of chain reaction coding, the probability of a
decoder being able to decode a media object is low until the
decoder has collected enough data, where enough data is
approximately the size of the entire media object. Thus, it is
unlikely that a media object can be decoded in parts when the
encoding is applied to the entire media object as a whole.
SUMMARY OF THE INVENTION
[0015] According to one embodiment of the invention, a method of
scheduling a media object for transmission between a server and a
client is provided. The method comprises partitioning the media
object into segments of blocks, wherein each block is a unit of
media for which a client will wait to receive an entire block
before playing out the block, and wherein each segment includes an
integer number of blocks. The method also comprises determining one
or more channels on which to serve each segment, the channels
capable of carrying data between the server and the client, and
determining a rate at which to serve each segment. The method
further comprises determining a schedule pair for each channel, the
schedule pair including a time at which the client may start
receiving on the channel and a time at which the client may stop
receiving on the channel.
[0016] In another embodiment of the invention, a system for
scheduling a media object for transmission between a server and a
client is provided. The system includes a module for partitioning
the media object into segments of blocks, wherein each block is a
unit of media for which a client will wait to receive an entire
block before playing out the block, and wherein each segment
includes an integer number of blocks. The system additionally
includes a module for determining one or more channels on which to
serve each segment, the channels capable of carrying data between
the server and the client, and a module for determining a rate at
which to serve each segment. The system further includes a module
for determining a schedule pair for each channel, the schedule pair
including a time at which the client may start receiving on the
channel and a time at which the client may stop receiving on the
channel.
[0017] In another aspect of the invention, a method of serving a
media object is provided. The method comprises receiving segments
of a media object, wherein each segment includes an integer number
of blocks, wherein each block is a unit of media for which the
client will wait to receive an entire block before playing out the
block, and wherein each block includes one or more input symbols.
The method also comprises, for each segment, receiving an
indication of one or more channels on which to serve the segment
and, for each segment, receiving a rate at which to serve the
segment. The method additionally comprises determining an order in
which to encode blocks. The method further comprises generating
output symbols for each block in the order, and transmitting the
output symbols on the corresponding one or more channels, wherein
each segment is served at the corresponding rate.
[0018] In yet another embodiment of the invention, an apparatus for
serving a media object is provided. The apparatus includes a block
encoder coupled to receive segments of a media object, wherein each
segment includes an integer number of blocks, wherein each block is
a unit of media for which a client will wait to receive an entire
block before playing out the block, and wherein each block includes
one or more input symbols. The block encoder includes an input to
receive an order in which to encode the blocks, and wherein the
block encoder is configured to generate, in the order, output
symbols for each block. The apparatus additionally includes a
transmitter coupled to receive the output symbols from the block
encoder, and coupled to receive, for each segment, an indication of
one or more channels on which to serve the segment and a rate at
which to serve the segment. The transmitter is configured to serve
the output symbols on the corresponding one or more channels at the
corresponding rate.
[0019] In yet another aspect of the invention a method of receiving
a media object that includes segments of blocks, wherein each
segment includes an integer number of blocks, and wherein each
block is a unit of media for which a client will wait to receive an
entire block before playing out the block is provided. The method
comprises receiving a media object description of the media object,
and joining and leaving each of a plurality of channels according
to the media object description to download the segments. The
method additionally comprises reassembling the blocks in each
segment, and playing the blocks out in an order after a startup
latency.
[0020] In still another embodiment of the invention a system for
receiving a media object that includes segments of blocks, wherein
each segment includes an integer number of blocks, and wherein each
block is a unit of media for which a client will wait to receive an
entire block before playing out the block, is provided. The system
includes a module for handling input of a media object description
of the media object, and a module for handling channel joins and
channel leaves for each of a plurality of channels according to the
media object description, wherein the channels are capable of use
for downloading the segments to the client. The system also
includes a module for reassembling the blocks in each segment, and
a module for playing the blocks out in an order after a startup
latency.
[0021] Advantages of the present invention include providing
improved scheduling of media objects that are to be served to
clients so that the clients may play out the media objects
uninterrupted after a startup latency. Advantages also include
improved serving of media objects to clients. Advantages further
include improved receiving of media objects served by a server.
[0022] A further understanding of the nature and the advantages of
the inventions disclosed herein may be realized by reference to the
remaining portions of the specification and the attached
drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
[0023] FIG. 1 is a block diagram of a MOD system with k servers and
m clients according to one embodiment of the present invention.
[0024] FIG. 2 is a block diagram of a MOD server with a single
server and a single client according to one embodiment of the
present invention.
[0025] FIG. 3 is a block diagram of a block encoder as may be used
in the MOD system shown in FIG. 2.
[0026] FIG. 4 is a block diagram of an encoder as may be used in
the block encoder shown in FIG. 3.
[0027] FIG. 5 is an illustration of how an output symbol may be
generated from a set of associated input symbols.
[0028] FIG. 6 is a block diagram of a receive module as may be used
in the MOD system shown in FIG. 2.
[0029] FIG. 7 is a simplified flow diagram of a process that may be
used by the client scheduler shown in FIG. 2, to determine the time
to join and leave channels, and when to reassemble all the blocks
received on a channel.
[0030] FIG. 8 is a simplified flow diagram of a process that may be
used by the client scheduler shown in FIG. 2, to determine when to
reassemble each block.
[0031] FIG. 9 is a plot of the amount of content that should be
downloaded, and the amount of content that has played out, from the
time the client request the content until it has completed playing
out.
[0032] FIG. 10 and FIG. 11 show plots of the maximum amount of
content that may be downloaded by the client for .alpha.=2 and
.alpha.=3/4 respectively.
[0033] FIG. 12 shows a plot of the amount of required server
bandwidth as the length of the movie varies from 1 minute to 4
hours for optimal server bandwidth scheduling.
[0034] FIG. 13 is a simplified flow diagram of a process that may
be used by the media object schedule generator shown in FIG. 2, to
determine how to schedule a media object using a variable rate
fixed segment size scheduler.
[0035] FIG. 14 is an illustration of how a variable rate fixed
segment size scheduler may schedule the first five segments in a
media object.
[0036] FIG. 15 is a simplified flow diagram of a process that may
be used by the media object schedule generator shown in FIG. 2, to
determine how to schedule a media object using a fixed rate
variable segment size scheduler.
[0037] FIG. 16 shows a plot of the amount of required server
bandwidth versus .alpha. for various numbers of concurrent channels
at the client for a fixed rate variable segment size scheduler.
[0038] FIG. 17 shows a plot of the amount of required server
bandwidth versus the number of concurrent channels at the client
for a fixed rate variable segment size scheduler.
[0039] FIG. 18 is an illustration of how a fixed rate variable
segment size scheduler may schedule the first ten segments in a
media object when the download bandwidth for a single segment is
split among later segments.
[0040] FIG. 19 is an illustration of how a fixed rate variable
segment size scheduler may schedule five segments in a media object
when the download bandwidth for two segments is combined for use in
later segments.
[0041] FIG. 20 is a simplified flow diagram of a process that may
be used by the media object schedule generator shown in FIG. 2, to
determine how to schedule a media object using a restricted server
channel scheduler.
DESCRIPTION OF THE SPECIFIC EMBODIMENTS
[0042] The present disclosure references the following
applications, the entire disclosures of which are herein
incorporated by reference for all purposes:
[0043] (1) U.S. Pat. No. ______ (U.S. patent application Ser. No.
09/246,015, filed Feb. 5, 1999 and entitled "Information Additive
Code Generator And Decoder For Communication Systems") ("Luby
I");
[0044] (2) U.S. Pat. No. ______ (U.S. patent application Ser. No.
09/399,201, filed Sep. 17, 1999 and entitled "Information Additive
Group Code Generator And Decoder For Communication Systems" ("Luby
II");
[0045] (3) U.S. Pat. No. ______ (U.S. patent application Ser. No.
09/587,542 (Atty. Docket No. 019186-001610US), filed Jun. 1, 2000
and entitled "Dynamic Layer Congestion Control for Multicast
Transport" (hereinafter "Dynamic Layering Application");
[0046] (4) U.S. Pat. No. ______ (U.S. patent application Ser. No.
09/668,452 (Atty. Docket No. 019186-002300US), filed Sep. 22, 2000
and entitled "On Demand Encoding With a Window" (hereinafter
"Windowing Application");
[0047] (5) U.S. Pat. No. ______ (U.S. patent application Ser. No.
09/691,735 (Atty. Docket No. 019186-002100US), filed Oct. 18, 2000
and entitled "Generating High Weight Encoding Symbols Using a
Basis" (hereinafter "High Weight Application"); and
[0048] (6) U.S. Provisional Patent Application No. 60/248,443
(Atty. Docket No. 019186-002200US), filed Nov. 13, 2000 and
entitled "Scheduling Multiple Files on a Single Server"
(hereinafter "Scheduling Multiple Files Application").
[0049] The above-referenced applications provide teachings of
systems and methods that may be employed in certain embodiments
according to the present invention. It is to be understood that
these systems and methods are not required of the present
invention, and many other variations, modifications, or
alternatives may also be used.
[0050] A MOD system may be constrained by available resources such
as the total network bandwidth available to the MOD system, the
client download rate or the total number of channels (or multicast
groups, where a multicast network is used) available to the client
or the server.
[0051] In a true MOD system, the only delay that a client will
experience after a request is made for a media object is the
propagation delay before the first data of the media object
arrives. In practical MOD systems, a small startup latency at the
client is permitted in order to make the system scalable, more
reliable and reduce the amount of required system resources.
[0052] In some cases, the client or the network between the server
and the client are more of a constraint on transmission of the
stream. For example, the maximum rate that the client can download
a media object, Rd, may be constrained by the client's link to the
network. What is crucial is the relationship between the download
rate Rd and the media play out rate Rp. If a personal computer user
is connected to the Internet over a 56 Kbps (kilobits per second)
modem line, i.e., Rd=56 Kbps, and requests a full-motion,
full-resolution 4.0 Mbps (megabits per second) DVD quality video,
i.e., Rp=4 Mbps, then that user might not be able to receive that
video stream, and start viewing it uninterrupted without a
substantial delay in which almost the entire video is downloaded.
This is true regardless of the speed of the server, or how much
bandwidth is available between the server and the network.
Similarly, the aggregate rate at which the server can deliver data
to the network may be limited by the server hardware or the
available bandwidth in the network. Increasing the server's output
may be extremely costly or infeasible, so using the bandwidth
available at the server efficiently may be an important system
constraint. The amount of bandwidth used by a particular media
object at the server is defined to be Rs, i.e., the aggregate rate
at which the object is being served.
[0053] A more realistic example is a play out rate Rp of 300 Kbps
and a download rate Rd of 400 Kbps. In this case, an MOD system
should be carefully designed to enable a client to download at a
slightly greater rate than the play out rate and still experience
uninterrupted play out of the media.
[0054] Another factor to consider is that even though the server
bandwidth, Rs, and the maximum client download rate, Rd, may be
fixed for a particular media object, the actual rate that a client
may download the media object can vary over time, depending, for
instance, on the current point in the play out of the content, or
the presence of congestion in the network.
[0055] For any MOD system, the number of channels available for use
may also be an important resource the MOD system should consider in
determining how to serve a set of media objects efficiently. In a
multicast network, the number of multicast groups available to the
server, g, could be limited due to server or network administrative
reasons such as the memory required to maintain a list of multicast
groups, or the number of joins and leave messages the server and
the network routers can handle. Similarly, the number of multicast
groups, r, that the client is actively joined to at once may be
limited due to constraints imposed by the system design, memory
limitations, or the processing power of the network interface card
(NIC) at the client.
Partitioning a Media Object into Blocks and Segments
[0056] A media object file may be divided into sequentially
numbered blocks, where the block index indicates the temporal
position of each block in playing out the media object. As used
herein, "media object" refers to data representing some media
content where the size of the data is defined and "media stream"
refers to data representing some media content where the size of
the data is open-ended. An example of the former is a 60 minute
video encoded at a rate of 4 Mbps and an example of the latter is a
live streaming video feed. Unless otherwise indicated, it should be
assumed herein that what is described with reference to media
objects can also be used with reference to media streams and vice
versa.
[0057] In the present invention, a block may be treated as the
fundamental unit of play out in the system, in that the client
cannot play out any of the content associated with a block until
the entire block is received at the client. In one embodiment, for
instance for video content, a block could be a single video frame
and the sound associated with that frame. In another embodiment, a
block could be a minute of compressed video. The size of each block
may be varied. One or more blocks are grouped together to form a
segment. In one embodiment, each segment includes only one block.
If each segment is transmitted on a separate channel, or set of
channels, then a client can join and leave channels such that the
client downloads the segments in some order. The client can also
reassemble and view the media object from the beginning, while
other segments are still arriving.
[0058] The client stores the packets for a block as they arrive and
waits for the entire block to arrive before playing it out. There
need not be a time that is global to all clients associated with
each block, at which the client starts downloading or playing out
the block. Each client can initiate the session at any point in
time, and time may then be measured for each client relative to
when it initiated the session. In embodiments of the MOD scheme,
each client experiences the same startup latency. Similarly, the
resources required for each client may be identical and
predictable, and may be based on the size and rate that each block
is being served. Thus, the global time at which a particular client
starts downloading or playing out a block may be particular to when
it initiated the download of the media object or stream, but the
experience of each client may be the same relative to when it
initiated the session.
[0059] The transmission of the segments from the server to the
client across the channel in the present invention will be
generally described in the context of a chain reaction coding
system. However, it is to be understood that the invention is not
limited to chain reaction coding systems. Other methods for
transmitting the segments across the channel will be described in
more detail below.
[0060] Luby I, Luby II, the High Weight Application, and the
Windowing Application describe various chain reaction coding
systems in detail. As described therein, a chain reaction encoder
generates output symbols from input symbols of the input file as
needed. Luby I, Luby II, the High Weight Application, and the
Windowing Application define and describe various concepts relating
to chain reaction coding, and those teachings are incorporated by
reference herein rather than restating them here in their entirety.
Generally speaking, Luby I, Luby II, the High Weight Application,
and the Windowing Application teach how to make and use a chain
reaction coding encoder and decoder to transport data, including a
process for generating an output symbol or a group of output
symbols from selected input symbols.
[0061] Luby I, Luby II, the High Weight Application, and the
Windowing Application generally describe chain reaction coding
systems for an entire input file. In the present invention, the
file, or media object, is broken into blocks, where each block may
be encoded and decoded independently using a chain reaction encoder
and decoder, respectively. In some implementations of chain
reaction coding, the size of each block determines the efficiency
of the chain reaction encoder and decoder. Generally, there is a
trade off between the overhead and the encoding/decoding speed. For
a fixed number of symbols, generally the encoding/decoding speed
(in Mbps) increases as the symbol size increases. However, the
amount of overhead, i.e., the number of extra output symbols that
the decoder should collect greater than the block size, is
proportionately smaller for larger blocks. To minimize the required
overhead the blocks should therefore be as large as possible.
Hence, for a fixed block size, increasing the symbol size improves
encoding/decoding speed at the cost of overhead performance. One
approach is to set the block size (which may be variable) to match
the smallest unit of media object that can be played out by the
client or transmitted by the server.
[0062] It should be understood that in different contexts in this
specification, references are made to variables that stand for
indefinite numbers and where different variables are used in
different contexts, those numbers need not be the same.
Basic System Overview
[0063] FIG. 1 shows an exemplary MOD system 100. As illustrated
there, MOD servers 102 serve media objects to MOD clients 104 over
a network 106. As shown, MOD system 100 includes k MOD servers and
m MOD clients, but the actual values for k and m are not
particularly relevant to this description. In the preferred
embodiment, network 106 is the Internet, but other networks or
subsets of the Internet can be used as well, such as intranets,
virtual private networks, LAN's, WAN's, wireless or satellite
networks. In fact, network 106 need not be a network in the
strictest sense of the term, but may be a collection of
point-to-point connections.
[0064] However configured, MOD servers 102 receive requests for
media objects from MOD clients 104. In FIG. 1, an administrative
server 125 may determine a schedule of which media objects are
available. The administrative server 125 may also maintain a list
of MOD clients 104 that are eligible to download a particular media
object. For instance, in a pay-per-view system, the administrative
server 125 may maintain a list of MOD clients 104 that have paid
and are currently eligible to download and/or view a particular
media object. The administrative server 125 also may control the
MOD server 102, and may determine a schedule of which MOD servers
will serve a particular media object, or portion of a media object.
A list of media objects currently available for download and any
other parameters required by the MOD client 104 may be published by
the administrative server 125 at the request processor 130. For
instance, the request processor 130 may publish the play out rate,
cost, length, and schedule of availability of each media object
currently being served by the MOD servers 102. The MOD server 102,
administrative server 125 and request processor 130 may reside on a
single machine, separate machines, or some combination thereof.
[0065] Only one administrative server 125 and request processor 130
are shown, but it should be understood that other configurations
are possible. For instance, in one embodiment each MOD server 102
may be associated with an administrative server 125 and/or request
processor 130. In another embodiment, a subset of MOD servers 102
may be associated with the same administrative server 125 and/or
request processor 130. For example, all the MOD servers 102 in one
location may be associated with a single administrative server 125.
Alternatively, all the MOD servers 102 serving the same media
object or set of media objects may be associated with the same
administrative server 125. In another embodiment, two or more
request processors 130 may be associated with a single
administrative server 125. For instance, one administrative server
125, may have one request processor 130 for each different type of
media object that is available to different classes of MOD clients
104. Similarly, one request processor may be associated with
multiple administrative servers 125.
[0066] The media objects may be stored in media object storage 108,
but may also be a media object stream 115 fed to MOD servers 102
from other sources as needed. Only one media object storage 108 and
media object stream 115 is shown, but it should be understood that
other configurations are possible. For instance, in one embodiment
each MOD server 102 may be connected to a different media object
storage 108 and/or media object stream 115. In another embodiment,
a subset of MOD servers 102 may be fed by the same media object
storage 108 and/or media object stream 115.
[0067] The MOD client 104, may obtain a list of media objects
available using a client browser 135. The client browser 135 may
obtain the list of media objects from the request processor 130 via
the network 106. Note that the channel used by the client browser
135 may be the same physical channel used by the MOD client 104.
Once the client browser 135 requests a particular media object from
the request processor 130, the client browser 135 may download a
media object description from the request processor 130 which
contains information necessary for the client to begin downloading
the media object. For example, the media object description may
include the length, play out rate, startup latency and download
rate of the media object as well as the location of the MOD server
or servers that are currently serving the media object For
instance, in a multicast network, the media object description may
contain a list of multicast groups and the relative time at which
the client should join each multicast group. The client browser 135
may send the media object description to the MOD client.
[0068] In one embodiment, the client browser 135 may be a web
browser, which receives the schedule using an http or a similar
request from the request processor 130, which may include a web
server. However, this is not the only way a client may hear an
announcement of an MOD session. In another embodiment, the request
processor 130 may continuously advertise, on a dedicated set of
channels, which MOD sessions are available and the relevant
information required by the client to join to them. The client
browser 135 may maintain a permanent list of available MOD sessions
or just listen to the channel to see what is available based on a
user request.
[0069] For the purposes of describing some of the system
constraints, the connection between a MOD server 102(i) and network
106 may be modeled as a plurality of server channels 110(i) and the
connection between a MOD client 104(i) and the network 106 may be
modeled as a plurality of client channels 112(i). The server
channels 110 and client channels 112 may be physical and/or logical
channels. In one embodiment, the MOD client 104 and MOD server 102
may be connected via a multicast network. A server may send all
packets destined for a particular logical channel to a particular
multicast group. The client may join and/or leave a particular
logical channel by joining and/or leaving the particular multicast
group. A join or a leave message propagates through the network
towards the server up to the first router that can service the
request. The multicast network may make sure that packets sent to
multicast group are received by all the joined clients. In another
embodiment, the MOD client 104 and MOD server 106 may be connected
via a unicast network. The server may send all the packets destined
for a particular logical channel to a particular unicast address
and a particular port of a server. The client may join a particular
logical channel by sending a request to receive any packets
received by a particular server on a particular port address. The
server may make sure that all packets received on a particular port
address are copied and sent to all clients that have sent request
messages to receive such packets, and thus the server may be an
active agent in making the bindings between the logical channel and
the physical channels. In another embodiment, the MOD server 102
may be connected to the MOD client 104 via a broadcast network,
such as a satellite network and the MOD client 104 may either
adjust its receiver to receive a particular subset of physical
channels or receive a single physical channel and filter out the
data except for a particular subset of logical channels, or a
combination thereof. In one direction the physical channel may be
the satellite uplink from the MOD server 102 to the network, while
the client channel may be a terrestrial link to a local switch in
the network. Similarly, the server channels 110 from the network
106 to the MOD server 102 may be terrestrial links, while the
server channels 110 from the MOD server 102 to the network 106 may
be satellite based. Unless otherwise indicated, where multicasting
is referred to herein as a mechanism for scaling the required
server and network bandwidths, it should be understood that
broadcasting could be used as well.
[0070] The logical channels are each shown as bi-directional.
Requests may flow from MOD clients 104 towards MOD servers 102,
media objects may flow from media object storage 108 to MOD servers
102, served media data may flow from MOD servers 102 to MOD clients
104 and media objects to be played out may flow from MOD clients
104 to their associated MOD play out devices 120. Examples of play
out devices may include televisions, music-playing devices,
computer processors (where the media is to be operated upon by
computers as opposed to being viewed by users), or software running
on a general purpose computer (e.g., media player plug-ins or
client software).
A Basic MOD System
[0071] In one embodiment, a single MOD server 102 may serve one or
more media objects to any number of clients. In another embodiment,
a single MOD server 102 may only serve a portion of a media object
to any number of clients, where other MOD servers 102 serve the
rest of the segments in the media object. For example, one MOD
server 102 may serve the even numbered segments in a media object,
and a second MOD server 102 may serve the odd numbered segments. As
another example, one MOD server 102 may serve the segments in the
first half of the media object, while a second MOD server 102 may
serve the segments in the second half. In another embodiment, more
than one MOD server 102 may serve the same segment in a media
object. For example, two or more MOD servers 102 at different
geographic locations in the network may be used to serve the entire
media object, or overlapping portions of the media object
concurrently. Additionally, MOD clients 104 can download from one
or more of the MOD servers 102 concurrently. Multiple MOD servers
102 at the same or different locations in the network make the MOD
system 100 more reliable in the event portions of the network, or
some of the MOD servers 102 fail. One skilled in the art will
recognize many other combinations, variations, and alternatives to
the above embodiments and examples.
[0072] In one embodiment, a single MOD client 104 may receive the
same or disjoint portions of a single media object from one or more
MOD servers 102. For example, a MOD client 104 may download a
segment of the media object from the nearest available MOD server
102, or from a MOD server among the MOD servers 102(1-k) currently
serving the segment with the lowest aggregate load of MOD clients
104. As another example, the MOD client 104 may decide from which
MOD server 102 to download a segment based on its current reception
rate of other segments from that MOD server 102. In another
embodiment, the MOD client 104 may add or remove MOD servers 102
from which it is downloading in order to receive a segment. For
example, a MOD client 104 may try to download a segment from a
second MOD server 102 concurrently if there is congestion in the
network. As another example, a MOD client 104 may download a
segment from a second MOD server 102 due to failure of, lost
communication with, etc., a first MOD server 102 from which it was
downloading the segment.
[0073] FIG. 2 is a block diagram of a basic MOD system 200
comprising a single MOD server 102 and a single MOD client 104,
connected by a plurality of channels 245. Here for simplicity the
server channels 110, network 106 and client channels 112 have been
combined into a single set of channels 245. As stated before, these
channels may be physical channels or logical channels. In the MOD
system 200, a media object residing in media object storage 201, or
forming a media object stream 205, may be provided to a media
object symbol generator 210. Media object symbol generator 210 may
generate a sequence of one or more input symbols (IS(0), IS(1),
IS(2), . . . ) from the media object file or stream, with each
input symbol having a value and a position (denoted in FIG. 2 as a
parenthesized integer). The output of media object symbol generator
210 is provided to a block encoder 215.
[0074] The media object schedule generator 214 receives as input
the administrative server description of the media object and
generates a media object description for the request processor 130.
In one embodiment, the functionality of the media object schedule
generator 214 may be performed by the administrative server 125.
One function of the media object schedule generator 214 may be to
partition the media object into a set of segments, where each
segment comprises one or more blocks and has an associated size,
rate and set of channels to be served on. If the media object is a
stream, then the media object schedule generator 214 may partition
the media object as the content arrives. The rate and channel to
serve each block or segment may vary over time. For instance, in a
multicast network, a dynamic layering scheme may be used to serve
each segment. For example, the Dynamic Layering Application
describes various dynamic layering schemes that may be used. It is
to be understood, however, that many other types of dynamic
layering schemes may also be used. Moreover, other methods in
addition to dynamic layering schemes may also be used to serve each
segment.
[0075] The size, rate and set of channels to serve each segment and
block of the media object may serve as the input to the media
object block scheduler 216. Each segment of the media object may be
logically divided into a plurality of disjoint blocks by the media
object block scheduler 216. In one embodiment, the blocks may be
approximately equal in size. The media object block scheduler 216
may determine how many output symbols from each block of each
segment are to be generated by the block encoder 215. Each block
may be identified by a unique block index F. When chain reaction
codes are used, if output symbols are generated on the fly, the
media object block scheduler 216 should ensure that there are
enough output symbols generated for each block so that each segment
can be served at the rate determined by the media object schedule
generator 214. Each output symbol may be generated as a function of
input symbols from a single block or subset of blocks. The block or
set of blocks from which input symbols are chosen to generate an
output symbol will be referred to as the blocks associated with
that output symbol. In a specific embodiment, each output symbol is
associated with only one block.
[0076] The block encoder 215 provides output symbols to a transmit
module 240. Transmit module 240 may also be provided the key of
each such output symbol and the set of blocks associated with each
such output symbol. Transmit module 240 transmits the output
symbols, and depending on the keying method used, transmit module
240 may also transmit some data about the keys of the transmitted
output symbols, or the associated blocks, over a plurality of
channels 245 to a receive module 250. Transmit module 240 may
include a buffer to store output symbols, thereby allowing transmit
module 240 to transmit output symbols in an order that is different
from the order in which transmit module 240 receives the output
symbols. Transmit module may transmit each segment on a different
channel, on multiple channels, or subsets of segments on different
channels.
[0077] In a specific embodiment, block encoder 215 is a chain
reaction encoder as described below. Each block is encoded and
decoded separately, as described in the Scheduling Multiple Files
Application. Since each block or segment is sent on a separate
channel, it is not necessary to collect enough symbols to decode
the entire media object before decoding and hence playing out the
content. For instance, in one embodiment, the blocks or segments
are received and decoded in order. It is to be understood, however,
that a chain reaction code need not be used, and that each block
need not be encoded and decoded separately as described in the
Scheduling Multiple Files Application.
[0078] The system can also be implemented with other encoders or
without any encoding at all. In one embodiment, the output symbols
are simply the original input symbols which are broadcast
repeatedly by transmit module 240 on each channel in a looping
manner. In another embodiment, the original symbols are encoded by
block encoder 215 using an erasure correcting code other than a
chain reaction code, such as a Reed-Solomon code or a Tornado code.
These codes have a fixed number of encoded symbols so if there is
sufficient storage available at the server, the block encoder 215
need only generate each output symbol once and store it, where the
encoded symbols may be generated prior to the start of
transmission, or as needed by the transmit module 240. The encoded
symbols may then be broadcast repeatedly by transmit module 240 on
each channel either in a looping manner, or by choosing a random
output symbol each time. For encoded data, it is important to
design the system so that a client receives a minimal number of
duplicate encoded packets. The system works well using a chain
reaction encoder because the client may concurrently download
encoded data from multiple servers without coordination and not
receive redundant data. An additional advantage of chain reaction
codes is that they offer a greater amount of protection and
flexibility in system design against loss compared to a code that
has been designed for a fixed loss rate.
[0079] Receive module 250 receives the output symbols from the
plurality of channels and determines the key I and block F
associated with each output symbol. In one embodiment, the receive
module may use information explicitly transmitted with each output
symbol to determine the key and the block for each output symbol.
In another embodiment, receive module 250 may use timing
information, or information about the channel an output symbol was
received on, in order to calculate the key I or the block F for
that output symbol. Receive module 250 provides the output symbols
and their associated block and key to a temporary storage buffer
252.
[0080] A client scheduler 262 receives the media object description
from the client browser 135. The client scheduler 262 determines
the time to join and leave channels, and when to have the block
decoder 255 reassemble any block in its entirety, where reassemble
may be defined to include decode, or just reorder in the case only
the original data is sent. In one embodiment, the client scheduler
262 uses feedback from the block decoder 255 and receive module 250
in order to determine these actions. For instance, in the presence
of loss, or if the client varies its reception rate by joining and
leaving channels over time in the presence of congestion, the
client scheduler 262 may wait for the receive module 250 to
indicate that enough output symbols from a particular channel, or
for a particular block, have been received before issuing a command
to the receive module 250 to leave the channel, or the block
decoder 255 to reassemble the block F. In another embodiment, the
client scheduler 262 uses timing information from the media object
description to determine which channels to join and leave and when
to reassemble a block. The block decoder 255 receives a signal from
the client scheduler 262 to reassemble a block. The block decoder
255 loads the block F to be reassembled from the temporary storage
buffer 252, and may use the keys together with the corresponding
output symbols, to recover the input symbols (again IS(0), IS(1),
IS(2), . . . ). Block decoder 255 provides the recovered input
symbols to a media object reassembler 265, which generates a copy
270 of the media object file 201 or media object stream 205, or
feeds the reassembled media object directly to a media object
player 275.
[0081] In one embodiment the MOD play out device 120 receives the
reassembled media object as a stream as in the media object player
275. In another embodiment, the media object is first stored as a
media object file 270, and later accessed by the MOD play out
device 120. The MOD play out device 120 may wait for the entire
media object to be reconstructed before beginning the play out, or
just a portion of it, for instance, the first block or segment.
A Basic Block Encoder
[0082] FIG. 3 shows a basic block encoder 215 for an MOD system
using chain reaction codes. The block encoder 215 comprises a
random number generator 335, counter 325, stream identifier 322,
key generator 320 and encoder 315. Key generator 320 generates a
key for each output symbol to be generated by encoder 315. Key
generator 320 may use a combination of the output of a counter 325,
a unique stream identifier 322, and/or the output of a random
number generator 335 to produce each key. The counter 325, random
number generator 335 and stream identifier may all have a block
index F as input. In one embodiment, each key may be generated
independently of the block for which encoder 315 is to generate an
output symbol, and the same key may be used on multiple different
blocks. However, it is to be understood that the same key need not
be used on multiple different blocks. The output of key generator
320 is provided to the encoder 315. Using each key I provided by
key generator 320 and block index F, the encoder 315 generates an
output symbol, with a value B(I,F). The value of an output symbol
is generated based on its key and on some function of one or more
input symbols in the block F provided by the media object symbol
generator 210. In one embodiment, a heavy weight basis is
associated with each block so that output symbols of large weight
may be generated more efficiently. The High Weight Application
provides further details on generating output symbols using a
basis. It is to be understood, however, that many other methods of
generating large weight symbols using a basis. Moreover, many other
methods may be used to generate large weight symbols. The input
symbols used in generating an output symbol are referred to herein
as the output symbol's "associated input symbols".
[0083] FIG. 4 is a block diagram of one embodiment of block encoder
315 shown in FIG. 3. The block diagram of FIG. 4 is explained
herein with references to FIG. 5, which is a diagram showing the
logical equivalent of some of the processing performed by the
encoder shown in FIG. 4.
[0084] Block encoder 315 is provided with input symbols from a fast
buffer 405 and input symbol buffer 400 and a key I and a block F
for each output symbol it is to generate. As shown, L(F) input
symbols in order of position are stored in fast buffer 205, where
L(F) is the number of symbols in block F and the first input symbol
in block F occurs at position J. Also, while the size L(F) of the
block is expressed here as a function of the block index F, it
should be understood that in some variations, L(F) is fully
determinable from the block F while in other variations L(F) is a
function of other values as well, or is independent of F. In one
embodiment L(F) is the same for all the blocks.
[0085] The input symbols may be read as needed from the input
symbol buffer 400 into the fast buffer 415. In a specific
embodiment, fast buffer 405 would be a storage medium with a faster
access time than input symbol buffer 400. For example, fast buffer
405 may be implemented in RAM while input symbol buffer 400 is
stored in disk storage. In this embodiment, as many as possible of
the input symbols should be in fast buffer 405 when the calculator
425 is invoked, balancing the time savings of invoking the
calculator 425 when many symbols are in fast buffer 405 with the
time it takes to move input symbols from the input symbol buffer
400 to fast buffer 405.
[0086] Key I (provided by key generator 320 shown in FIG. 3) and
block F (provided by media object block scheduler 216 shown in FIG.
2) are inputs to value function selector 410, weight selector 415
and associator 420. The number of input symbols L(F) in block F may
also be provided to these three components 410, 415 and 420. A
calculator 425 is coupled to receive outputs from value function
selector 410, weight selector 415, associator 420, input symbol
buffer 400 and the fast buffer 405, and has an output for output
symbol values. It should be understood that other equivalent
arrangements to the elements shown in FIG. 4 may be used, and that
this is but one example of an encoder according to the present
invention.
[0087] Using I, F, and possibly L(F), weight selector 415
determines the number W(I,F) of input symbols that are to be
"associates" of the output symbol having key I. In one variation,
W(I,F) is chosen based on I but is not based on F. Once the number
W(I,F) is determined, weight selector 415 supplies the number to
associator 420 and to calculator 425 if needed.
[0088] Using I, F, W(I,F) and possibly L(F), associator 420
determines a list AL(I,F) of the W(I,F) positions of input symbols
selected among the L(F) symbols in block F to be associated with
the current output symbol. It should be understood that W(I,F) need
not be separately or explicitly calculated if associator 420 can
generate AL(I,F) without knowing W(I,F) ahead of time. Once AL(I,F)
is generated, W(I,F) can be easily determined because it is the
number of associates in AL(I,F). It should also be understood that
not all the input symbols used by the calculator need be in fast
buffer 405, as some or all of the input symbols may be obtained
from input symbol buffer 400. Preferably, as many input symbols as
possible can be obtained from the fast buffer 405.
[0089] Once I, W(I,F) and AL(I,F) are known and available to
calculator 425, then calculator 425 accesses the W(I,F) input
symbols referenced by AL(I,F) in fast buffer 405, or in input
symbol buffer 400 if the needed input symbols are not present in
fast buffer 405, to calculate the value, B(I,F), for the current
output symbol. Calculator 425 calculates the value B(I,F) of the
output symbol being calculated based on a value function V(I,F), if
a variable value function is used. One property of a suitable value
function is that it would allow the value for an associate in
AL(I,F) to be determined from output symbol value B(I,F) and from
the values for the other W(I,F)-1 associates in AL(I,F). One
preferred value function used in this step is the XOR value
function, since it satisfies this property, is easily computed and
easily inverted. However, other suitable value functions may be
used instead. Luby II describes, for instance, a system in which a
group of output symbols are generated using a Reed-Solomon value
function. Moreover, Luby II describes other value functions that
may also be used, including methods based on polynomials over
finite fields, methods based on linear systems of equations,
methods based on Cauchy matrices over finite fields, and other MDS
codes (of which Reed-Solomon codes are examples).
[0090] If used, value function selector 410 determines a value
function V(I,F) from key I, and possibly F and L(F). In one
variation, the value function V(I,F) is the same value function V
for all I and F. In that variation, value function selector 410 is
not needed and calculator 425 can be configured with the value
function V. For example, the value function may be XOR for all I,
i.e., the output symbol value is an XOR (exclusive OR) of the
values of all of its associates.
[0091] Block encoder 315 then outputs B(I,F). In effect, block
encoder 315 performs the action illustrated in FIG. 5, namely, to
generate an output symbol value B(I,F) as some value function of
selected input symbols. In the example shown, the value function is
XOR, the weight W(I,F) of the output symbol is 3, the block starts
at position J=5, the size L(F) of the block is 5, the associated
input symbols (the associates) are at positions 5, 7, and 8 and
have respective values IS(5), IS(7) and IS(8). Thus, the output
symbol is calculated as B(I,F)=IS(5) XOR IS(7) XOR IS(8) for that
value of I and F.
[0092] In some embodiments, the number L(F) of input symbols in
block F is used by the encoder 315 to select the associates. The
value L(F) may also be used by the encoder 315 to allocate storage
for input symbols. Where the input is a streaming file, a windowing
encoder such as, for example, the windowing encoder described in
the Windowing Application may be used to encode the block. It is to
be understood, however, that the many other types of windowing
encoders may also be used. Moreover, other methods in addition to
windowing may additionally be used to encode the block.
[0093] As described above, the media transmitted from a MOD server
102 to a MOD client 104 is preferably encoded using chain reaction
coding. In one embodiment, the coding can be done ahead of time, in
which case the encoded output symbols for the media object could be
stored in a storage buffer in transmit module 240. In a preferred
embodiment, the encoding can be done by the MOD servers 102 on the
fly, with the block encoder 215 continuously generating new output
symbols at the appropriate rate for each block. The operation of
several variations of a chain reaction coder 214 are described in
detail in Luby I, Luby II, the High Weight Application, and the
Windowing Application and rather than describe those in great
detail here, those references are incorporated herein.
A Basic Receive Module
[0094] FIG. 6 shows a basic receive module 250 from FIG. 2. The
receive module 250 comprises a receiver 600, a key regenerator 610
and a block index regenerator 620. Receiver 600 processes the data
received on the channels 245 and provides the output symbols to a
temporary storage buffer 252. Data that receiver 600 receives about
the keys of these output symbols is provided to a key regenerator
610, and data that receiver 600 receives about the block index is
provided to a block index regenerator 620.
[0095] Key regenerator 610 regenerates the keys for the received
output symbols and provides these keys to temporary storage buffer
252. Block index regenerator 620 regenerates the block index for
the received output symbols and provides these block indices to
temporary storage buffer 252. In one embodiment, temporary storage
buffer 252 will store all the output symbols from the same block in
a contiguous portion of the buffer. For instance, for a media
object comprising n blocks, the temporary storage buffer will
contain n disjoint portions of contiguous buffer space, where
output symbols from a particular block are stored in buffer space
reserved exclusively for that block. In another embodiment, the
buffer space for a block may be reused once that block has been
processed by the block decoder 255 in order to conserve the amount
of temporary storage buffer 252 required.
A Basic Client Scheduler
[0096] FIGS. 7 and 8 are simplified flow diagrams of a method
according to an embodiment of the invention that may be implemented
to control the actions of the client such as, for example, the
client scheduler 262 shown in FIG. 2. These diagrams are merely for
illustrative purposes and are not intended to limit the scope of
the claims herein. FIGS. 7 and 8 are a high-level flow diagram of a
method for controlling the actions of the client so that the client
can download the media object and play it out uninterrupted. FIG. 7
is an embodiment of a process that the client scheduler 262 may use
to determine which channels to join and leave and when to perform
these actions. FIG. 8 is an embodiment of a process that the client
scheduler 262 may use to determine when to reassemble each block.
First, in a step 710, the media object description is processed.
The media object description may include the order to join each
channel, the size of the segment or segments and the rates the
segment or segments are served on each channel, and the time at
which to join and leave each channel. In one embodiment, the media
object description may be simply a list of channels to join, where
the list is organized so that each time a segment completes
downloading on a channel, the client scheduler 262 will leave that
channel and join a new set of channels specified on the list. For
example, if the client only receives data on a single channel at a
time, the media object description may just be the list of the
channels to join in order. In another embodiment, the media object
description includes the list of channels, the segment and the size
of the segment associated with each channel. The media object
description may also include a security key for each channel and/or
the entire media object, which the client needs in order to
reassemble the received blocks on each channel. Note that only a
subset of these may be sufficient for the client scheduler 262 to
determine which channels to join and when to join them.
[0097] Next, in a step 715, the client scheduler 262 determines the
rules of joining and leaving the channels according to the media
object description. These rules may specify, for example, when a
client may join and/or leave a channel, the number of channels to
join at any particular time and/or time period, the order in which
to join and/or leave channels, etc. In one embodiment, the client
may only leave a channel when the segment completes downloading on
that channel. In another embodiment, the client may join and leave
the same channel more than once before a segment completes
downloading on that channel. For example, the client may join one
or more channels to increase its reception rate if, for example, it
does not experience congestion, and leave one or more channels to
decrease its reception rate if, for example, it does experience
congestion.
[0098] As described above, the order in which to join and/or leave
channels may be defined according to the media object description.
Therefore, in certain embodiments in which the client joins and/or
leaves channels as a function of congestion, the client may, for
example, join and leave channels in a stack-like manner according
to congestion (e.g., losses) experienced during the download. For
instance, if the client is not experiencing congestion, it may join
the next channel in the specified order to which it has not yet
joined, or re-join a channel that it had last dropped. Similarly,
the client may drop channels from which it has completed
downloading, and may also drop the most recently joined channel if
the client does experience congestion. Thus, the client may join or
drop channels in response to a level of congestion, and the
reception rate at the client oscillates accordingly. An average
reception rate should be maintained in order to guarantee that the
client will be able to play out the media object without
interruption.
[0099] In step 720, the client scheduler 262 sends a message to
receive module 250 to join the initial set of channels determined
in 715. The client scheduler 262 then waits for a message from the
receive module 250 to indicate that all the data from one of the
channels C has been received. When all the data for a segment on
channel C has been received, in a step 740, the client scheduler
262 sends a message to receive module 250 to leave channel C and
schedules the reassembly of the blocks from channel C as shown in
FIG. 8. If there are additional channels to join at this time, the
client scheduler 262 sends a message to receive module 250 to join
the appropriate channels in steps 750 and 760. If there are no more
channels to join at this time, in step 770 the client scheduler
checks if there are any segments that have not completed
downloading, or if there are more channels to be joined at a later
time. If there are still segments to be downloaded, or channels to
be joined in the future, then the client scheduler returns to step
730, otherwise the media object has completed downloading (step
780).
[0100] In another embodiment, the client scheduler may use timing
information to determine when to join and leave channels and may
join and leave channels at times other than when the reception for
a channel has completed. For example, a new channel may be joined
if some fraction of a segment on another channel has completed
downloading, or if some amount of time has passed.
[0101] FIG. 8 is a simplified flow diagram of a method according to
an embodiment of the invention that may be implemented to
reassemble a segment or set of segments downloaded on a channel C.
The flow diagram in FIG. 8 is initiated by a call from step 740 in
FIG. 7 when the client scheduler 262 is ready to reassemble the
blocks received on some channel C. First, in a step 805, the client
scheduler 262 determines if there is a set of blocks available that
is ready to be reassembled. If there is a set of blocks available,
then in step 810 the client scheduler 262 determines and orders the
set of blocks to be reassembled. Then, in steps 820 and 830, the
client scheduler 262 selects the first block F to be reassembled,
and sends a message to the block decoder 255 to reassemble block F.
The client scheduler 262, then loops through the set of additional
blocks from which to reassemble until all the blocks have completed
reassembly in steps 840 and 845. Once all the blocks in the set
have been reassembled, the client scheduler 262 returns to step
805. If there is no new set of blocks ready to be reassembled,
then, in a step 850, the client scheduler 262 performs the same
check as in step 770 of FIG. 7, to determine if there are any
segments that have not completed downloading, or if there are more
channels to be joined at a later time. If there are still segments
to be downloaded, or channels to be joined in the future, then the
client scheduler returns to step 805 and waits for the next channel
to complete downloading and be ready to reassemble. Otherwise, the
media object has completed reassembly (step 860).
[0102] In one embodiment, the block decoder 255 may reassemble a
particular block in a segment before the entire segment has been
received, and/or try and reassemble some or all of the blocks in a
segment for which data is currently being received. In another
embodiment, the block decoder 255 may reassemble a number of blocks
at once, where the number of blocks may vary from client to client
depending upon the system resources available to the block decoder
255. In a specific embodiment, the client scheduler 262 will
schedule the reassembly of each block so that if the block decoder
255 and the media object player 275 are using the same set of
system resources, the block decoder 255 will reassemble each block
using the resources in a manner that will not effect the quality of
the play out of the media object, while still ensuring that each
block is reassembled before it is required to be played out. For
instance, if the media object player 275 uses the processor at
specific known intervals, then the client scheduler 262 can ensure
that a block is not reassembled during those intervals.
Scheduling a Media Object According to a Set of Constraints
[0103] Several scheduling methods for the media object schedule
generator 214 shown in FIG. 2 will now be described. In one
embodiment, the media object schedule generator 214 may first
partition each media object into a set of disjoint blocks. Then,
the blocks may be grouped together to form a set of segments, where
each segment comprises one or more blocks. In another embodiment,
the media object schedule generator 214 may first partition each
media object into a set of disjoint segments, where each segment is
then partitioned into one or more disjoint blocks. Each segment is
scheduled to be served on one or more channels. The size of the
blocks and segments, and the rate on each channel may vary.
[0104] As shown in FIG. 2, the input to the media object schedule
generator 214 is an administrative server description, and the
output comprises a media object description which is sent to the
client, and a set of instructions for the media object block
scheduler 216 on how to serve each block.
[0105] The administrative server description comprises two parts:
the media object parameters and a set of system constraints for
serving the media object. The media object parameters include the
size S Mbits (megabits) and the play out rate Rp Mbps of the media
object. (Note that megabits is used here instead of megabytes to
simplify the equations below. To convert from megabits to
megabytes, divide by 8.) In one embodiment, the media object
parameters include a list of positions at which it is possible to
partition the media object. For example, for video content, the
media object schedule generator 214 may partition the media object
into frames, or sets of frames. The list may include the position
in the media object at which each frame starts. As another example,
the media object schedule generator 214 may be required to
partition the media object into scenes. For instance, the client
may want to insert commercials between the play out of different
blocks, and the commercials are desired to occur only at the end of
a scene.
[0106] The system constraints for serving the media object may
include basic constraints such as:
[0107] 1. the maximum total server bandwidth Rsmax Mbps;
[0108] 2. the maximum client download rate Rd Mbps;
[0109] 3. the maximum startup latency at the client Ts seconds;
[0110] 4. the maximum total number of channels concurrently
available to the client r;
[0111] 5. the maximum total number of channels concurrently
available to the server g; and
[0112] 6. the maximum temporary storage available at the client
Mmax Mbits.
[0113] Additional system constraints may include the minimum block
size Bmin, the maximum block size Bmax, and the maximum segment
size Smax. The minimum segment size is at least the minimum block
size. If the maximum block size Bmax is at least twice as large as
the minimum block size Bmin, then any segment can be partitioned
into disjoint blocks that are all within the minimum and maximum
block sizes. If the maximum block size is smaller than twice the
minimum block size, then some adjustment in the sizes of the
segments may be necessary in order to partition each segment into
an integer number of blocks.
[0114] One constraint that normally applies to scheduling methods
is that, unless the client has already downloaded the media object,
or has it otherwise available, the client should receive enough
data to reassemble a block in its entirety by the time the block is
to be played out in the sequence. Otherwise, the client will either
interrupt the play out and pause while the data is being received,
or move onto the next block or segment to be played out or
downloaded.
[0115] A few definitions are set out below for use in later
descriptions of some of the scheduling methods. One output of the
media object schedule generator 214 is a partition of the media
object into n blocks, where the size of block i is b(i) Mbits. The
size of the entire media object is therefore S = l = 0 n - 1
.times. b .function. ( i ) .times. Mbits . ( Equ . .times. 1 )
##EQU1##
[0116] Another output of the media object schedule generator 214 is
a partition of the media object into n' segments, S(0), S(1), . . .
, S(n'-1), where segment S(i) is N(i) Mbits in size, and comprises
an integer number 1(i) blocks. Segment S(i) is served on g(i)
independent logical channels. Define r(i)=r(i, 0)+r(i,1)+ . . .
+r(i, g(i)-1), to be the aggregate serving rate for segment S(i),
where r(i, C) is the rate that segment S(i) is served on the C-th
such channel. If a segment S(i) is only being served on a single
channel, then the download rate for S(i) is referred to here as
r(i) without ambiguity. Block j in segment S(i) is served on up to
g(i) channels. Define rb(j)=rb(j, 0)+rb(j, 1)+ . . . +rb(j,
g(i)-1), to be the aggregate serving rate for block j, where rb(j,
C) is the rate that block j is served on the C-th channel for
segment S(i). The aggregate rate of all the blocks being served for
segment S(i) is r(i). The aggregate serving rate, i.e., the amount
of bandwidth required at the server is Rs = i = - n - 1 .times. rb
.function. ( i ) = i = 0 n ' - 1 .times. r .function. ( i ) .times.
Mbps . ( Equ . .times. 2 ) ##EQU2##
[0117] In one embodiment, the aggregate rate rb(j) for each block j
in segment S(i) is the same for all the blocks in segment S(i). In
one embodiment, the aggregate rate rb(j,C) for block j in segment
S(i), served on the C-th channel for segment S(i) is equal to
r(i,C) rb(j)/r(i) for every block in every segment. The data for
the different blocks in a segment is transmitted on each channel
serving that segment as described in the Scheduling Multiple Files
Application. It is to be understood, however, that many other ways
for transmitting data may be used in addition to methods described
in the Scheduling Multiple Files Application.
[0118] Where the blocks in each segment are encoded using chain
reaction codes, there is no difficulty associated with partitioning
the content of a block among many channels, since a specific set,
or order, of output symbols is not required, and multiple
independent streams of output symbols are information additive. If
the original data is transmitted across the channel, or a fixed
rate FEC code is used to encode each block, then the data from the
block can be partitioned and served on different channels. The
fraction of the block that is served on each channel is
proportional to the amount of data that the client is expected to
download from that channel. In another embodiment, the same data
for a block is served on more than one channel.
[0119] For each channel C serving the media object, the media
object schedule generator 214 produces a pair of values (I(C),
E(C)), where I(C) is the schedule time that the client first joins
channel C and E(C) is the schedule time that the client leaves
channel C. Both times I(C) and E(C) are given relative to the time
that the client first receives data for the media object, and
E(C)-I(C) is the approximate time that the client will spend
downloading a segment on channel C. In one embodiment, a client
downloading the media object is not required to start receiving
data from channel C at time I(C) relative to the time the client
first receives data for the media object. Similarly, a client
downloading the media object is not required to stop receiving data
from channel C at time E(C) relative to the time the client first
receives data for the media object. A client is only required to
minimally fulfill the schedule pair for each channel, wherein
minimally fulfill means that the client receives data from the set
of channels associated with a segment for a length of time
sufficient to download and reassemble the entire segment before it
is to be played out.
[0120] For example, consider a segment S(i) of size N(i) Mbits
being served at rate r(i) Mbps, where the segment is served on a
single channel C. If the segment is scheduled to start playing out
at time t+E(C) at the client, the client experiences no loss,
N(i)=r(i) (E(C)-I(C)), and the segment can be reassembled
completely at the client once at least N(i) Mbps of data has been
received, then the client minimally fulfills the schedule pair if
it downloads the segment for time E(C)-I(C), wherein the time the
client starts downloading the segment S(i) from channel C is less
than or equal to time t+I(C). Alternatively, if the client
downloads the segment on channel C for two separate time intervals,
such that the total time spent downloading the segment is E(C)-I(C)
seconds, and the second time interval ends before time t+E(C), then
the client also minimally fulfills the schedule pair for segment
S(i).
[0121] As another example, consider a client that starts
downloading the segments in order, where the next segment starts
downloading once one of the current segments being downloaded
completes downloading. If the schedule pair is chosen so that
N(i).ltoreq.r(i) (E(C)-I(C)) for every segment S(i), the client
experiences no loss, and segment S(i) can be reassembled completely
at the client once at least N(i) Mbps of data has been received for
the client, then the client minimally fulfills the schedule pair
for each segment, without explicitly knowing any of the schedule
pairs.
[0122] The media object schedule generator 214 should partition and
serve the media object so that all the system constraints are
satisfied. The list of instructions generated by the media object
schedule generator 214 for the media object block scheduler 216
include how the media object is partitioned into blocks and
segments, and a list of channels on which to serve each segment.
The list may also include the rate to serve each block in each
segment, or the rate to serve each block in each segment for each
channel for that segment.
[0123] The media object description describes how the media object
is to be partitioned and downloaded on the different channels to
the client. The media object description may include a list of
block and segment sizes, a list of channels for the media object,
and the schedule pairs for each channel. In another embodiment, the
media object description includes a schedule of the order to join
the channels and how many channels to join at each time. In this
embodiment, the client leaves a channel once the necessary data for
that segment on that channel has been received. The client may also
join and leave a channel multiple times in response to congestion
in the network.
[0124] A "pseudo-segment" with index -1 may be played out during
the startup latency period. That pseudo-segment can be an empty
block, in which case the media object does not start playing out
until the end of the startup latency period, or the pseudo-segment
may be a pre-downloaded segment S(-1) that starts playing out when
the client requests the media object. In this way, there is no
apparent delay at the client. One possible implementation of this
would be for every client to pre-download segment S(-1) of the ten
(or some other number) most popular pieces of media object
currently available. When the client requests one of these media
objects, then the pre-downloaded segment S(-1) starts to play out
while the segment S(0) is being downloaded, resulting in an
apparent latency of zero.
[0125] Define rb(i,t) to be the aggregate rate at which a client
downloads information about block i at time t, and define T(i) to
be time that block i begins playing out, where again time is
measured relative to the client, where time zero is when the client
initiates the session and the download starts. Given a fixed
maximum client download rate Rd, the goal of the MOD system is to
achieve uninterrupted play out of the media object. This leads to
the following three constraints: 1 ) .times. .times. For .times.
.times. all .times. .times. i , .intg. 0 T .function. ( i ) .times.
rb .function. ( i , t ) .times. .times. d t .gtoreq. b .function. (
i ) ( Equ . .times. 3 ) 2 ) .times. .times. For .times. .times. all
.times. .times. t , rb .function. ( i , t ) .ltoreq. rb .function.
( i ) ( Equ . .times. 4 ) 3 ) .times. .times. For .times. .times.
all .times. .times. i , t , i = 0 n - 1 .times. rb .function. ( i ,
t ) .ltoreq. Rd ( Equ . .times. 5 ) ##EQU3##
[0126] The constraint in (Equ. 3) is due to the fact that the
client should have finished downloading block i by the time it
needs to play it out, i.e., by the time blocks 0, . . . , i-1 have
completed playing out.
[0127] Although the disclosure below can be described for general
parameters, for clarity of explanation consider the special case
where the media object has a play out rate Rp and the blocks are
all the minimum size Bmin. Each block will have a play out time of
Tf=Bmin/Rp seconds. Using this notation, T(i)=Ts+iTf. Define the
startup latency number m to be the number of blocks that the client
plays out in Ts seconds, i.e., m=Ts/Tf=BminTs/Rp. Note that m need
not be an integer.
[0128] For example, consider a media object comprising five blocks
of size Bmin, where the client starts to play out the content after
a startup latency of m=2 blocks. FIG. 9 shows the amount of content
that should be downloaded (solid line), and the amount of content
that has played out (dashed line), from the time the client request
the content until it has completed playing out. The stair-like
graph showing the amount of content downloaded is due to the
constraint in (Equ. 3) that a block should be downloaded completely
before it can be played out. As the size of each block decreases,
the height of the stairs decreases and the two curves approach one
another.
[0129] In some embodiments, the play out rate is constant over
time. In other embodiments, the play out rate may vary over time
and Rp is the average play out rate for a block. If the play out
rate varies substantially from block to block, then the scheduler
may compensate by either varying the rate each block is served,
and/or the size of each block, so that the block is downloaded in
time to have uninterrupted play out at the client.
[0130] In one embodiment, the client accounts for variations in the
play out rate by including an additional delay called a buffering
latency, defined to be the time between when the latest segment has
finished downloading and been reassembled and when the first block
in the segment actually starts to play out. The buffering latency
can be introduced at the client by increasing the initial startup
latency. In this way, variations in the play out rate result in
variations in the amount of buffering latency and do not affect the
quality of the play out.
[0131] The constraint in (Equ. 4) exists because the client cannot
download data for a block at a greater rate than what is being
transmitted by the server for that block. The design of a scheduler
where the client experiences loss is considered in more detail
below.
[0132] Finally, the constraint in (Equ. 5) exists because the
client can't download at a rate greater than the maximum allowed
client download rate. Define the ratio of the maximum client
download rate to the play out rate to be .alpha.=Rd/Rp. In order to
achieve uninterrupted play out of the media object with a startup
latency number m, comprising n blocks of size Bmin, each having a
play out time of Tf, the following two inequalities should be
satisfied. .alpha..gtoreq.1/m (Equ. 6) .alpha..gtoreq.n/(n-1+m)
(Equ. 7)
[0133] Note that (Equ. 6) and (Equ. 7) are a special case of (Equ.
3) and (Equ. 5), where all the blocks are chosen to be of size Bmin
and there is no loss between the server and the client. The (Equ.
6) is the more stringent constraint when .alpha..gtoreq.1, and
(Equ. 7) is the more stringent constraint when .alpha.<1.
[0134] The conditions in (Equ. 6) and (Equ. 7) come from the
following considerations. The time it takes to play out i blocks of
size Bmin plus the initial startup time is (iTf+Ts), or (i+m)Tf
seconds. During this time, the client downloads at most (i+m)TfRd
Mbits of content. In order to have uninterrupted play out, when the
client starts to play out block i, it should have already
downloaded blocks 0 thru i, i.e., the first i+1 blocks, or
(i+1)TfRp Mbits of content. Therefore, (i+m)TfRd.gtoreq.(i+1)TfRp,
for i=0, 1, . . . , n-1, for uninterrupted play out at the client.
Equivalently, .alpha..gtoreq.(i+1)/(i+m), for i=0, 1, . . . , n-1.
If this is true for i=0 and i=n-1, then it is true for all i.
Therefore, the two conditions in (Equ. 6) and (Equ. 7) should be
satisfied for uninterrupted play out at the client.
[0135] For example, consider the play out of the media object
comprising n=5 blocks shown in FIG. 9. For .alpha..gtoreq.1, (Equ.
6) applies, since the first block should be downloaded before play
out can commence, and after that each block can be downloaded in
time since the download rate is greater than the play out rate.
Similarly, if .alpha.<1, then (Equ. 7) applies, since the last
block should be downloaded in time to play it out. If the last
block can be downloaded in time to play out, then all the other
blocks should already have been downloaded and played out.
[0136] If the two necessary conditions are satisfied, then a
straightforward scheduling method that satisfies the constraints in
(Equ. 6) and (Equ. 7), is to have the server make each segment a
single block and send out each segment on a single channel at a
rate of Rd Mbps, which will allow a client to achieve the startup
time Ts by consecutively downloading each segment in order at a
rate Rd. For this straightforward scheduling method, the server
should have a server bandwidth of Rs=nRd Mbps, so the server
bandwidth grows linearly with the size of the media object.
Alternative methods, described below, exhibit logarithmic growth in
the server bandwidth requirement.
[0137] In FIG. 10 and FIG. 11, the dashed lines represent the
maximum amount of content that can be downloaded by the client for
.alpha.=2 and .alpha.=3/4 respectively. Using the same example as
in FIG. 9, the amount of content that should be downloaded from the
time the client requests the content until it has completed playing
out is shown by the solid line in these figures. In FIG. 10, the
dashed line is to the left of the solid line, so uninterrupted play
out is possible. Decreasing the startup latency moves the curves
closer together. In FIG. 11, the dashed line crosses the solid
line, so uninterrupted play out is not possible. This agrees with
(Equ. 7) which requires that .alpha..gtoreq.5/6 for uninterrupted
play out. Increasing the startup latency number to m=8/3 makes
uninterrupted play out achievable since now the two curves only
intersect as the content completes downloading.
[0138] The scheduling methods described below all check if the
system constraints satisfy (Equ. 6) and (Equ. 7) as a first step
before trying to schedule the media object.
Adjusting the Schedule to Account for Channel Loss and Decoding
Overhead
[0139] The media object schedule generator 214 uses an overhead
function to adjust the rate and/or the time that each segment is
scheduled to be downloaded by the client. Define the overhead
function Eps(N(i)) to be a function of the size of segment S(i).
The media object schedule generator 214 designs the schedule so
that a client receiving the data on a channel for the schedule
time, and with no loss, will receive N'(i)=(1+Eps(N(i)))N(i) Mbits
of data for segment S(i). The extra Eps(N(i))N(i) Mbits of data is
the overhead that the media object schedule generator 214 includes
for segment S(i) in the designed schedule to allow for loss between
the server and the client and/or any overhead required to
reassemble the blocks in the segment. Eps(N(i)) is chosen so that
each segment can be reassembled at the client with high
probability.
[0140] In one embodiment of chain reaction codes, the number of
extra symbols required at the decoder to reassemble a segment with
high probability is proportional to the square root of the number
of input symbols in that segment. The overhead function can
therefore be chosen to be Eps(N(i))=c/N(i).sup.0.5, where c is a
constant. The larger the value of c, the higher the probability
that a chain reaction decoder collecting N(i)+cN(i).sup.0.5 Mbits
of data, will be able to reassemble segment S(i). In one embodiment
c is 13 or 14 depending on the value of N(i).
[0141] In another embodiment, consider a client that loses at most
a fraction p of the data it is meant to receive. For example, a
client may adjust its reception rate in response to congestion by
joining and leaving channels. A client experiencing up to certain
levels of congestion may be expected to be joined to a particular
channel at least 1-p of the schedule time for that channel. If the
overhead function is chosen to be Eps(N(i))=p/(1-p), then a client
with no loss will receive 1/(1-p) times the data of a client that
loses a fraction p. Any client that loses up to a fraction p of the
data it is meant to receive for segment S(i) will receive at least
N(i) Mbits of information about S(i) and be able to reassemble
segment S(i) completely assuming a perfect erasure code is used. In
this case Eps(N(i)) is independent of N(i).
[0142] In another embodiment, consider a scheme where a client that
has no loss would receive d' packets on channel C for segment S(i),
and consider a client that loses each packet independently with
probability p. The probability that the client with loss receives d
or more packets is equal to i = 0 d ' - d .times. ( d ' i ) .times.
p i .function. ( 1 - p ) d ' - i , ( Equ . .times. 8 ) ##EQU4##
i.e., the probability that d'-d or less packets are lost. For
example, if N(i) Mbits is equivalent to 100 packets and perfect
erasure codes are used for the channel, and each packet is lost
with probability 0.1, i.e., an expected 10% packet loss rate, then
in order for the client to receive at least d=100 packets for
segment S(i) with 99.99% probability, d'=126 packets. In general,
Eps(N(i)) can be chosen according to the loss rate and number of
packets in segment S(i), so that at least N(i) Mbits is received
with some probability.
[0143] In another embodiment, the overhead function Eps(N(i))=0,
when no loss is allowed for at the client and the original data is
transmitted across the channels.
[0144] In addition, the overhead function could be chosen to
account for both a packet loss fraction p, as well as the decoder
overhead required for chain reaction codes or other codes. The
overhead function could also be dependent on the schedule pair, or
the number of blocks in the segment. For example, for a client that
experiences loss in bursts, the schedule time for each segment
affects the variance of the loss rate seen by the client for that
segment.
[0145] Another consideration in choosing the overhead function is
whether the client is concerned only about the quantity of data it
receives, or which particular data it receives. If the particular
data received by the client is important, then the overhead
function Eps(N(i)) may have to account for duplicate packets
received at the client. For instance, if the original data is
transmitted, or if a fixed rate FEC code is used, then it is
important to organize the data so that a client will receive all
the data it requires to reassemble the block, with as little
duplication as possible. In this case, choosing the overhead
function becomes much more difficult. When chain reaction codes are
used, the client only cares about the quantity of reception for
each block. The particular data that is received is generally not
an issue if the data for each block on the channel is transmitted
according to the methods described in Luby I, Luby II, and the
Scheduling Multiple Files Application unless the number of blocks
in a segment is large.
[0146] It is to be understood that the above overhead functions are
merely for illustrative purposes and are not intended to limit the
scope of the claims herein. One skilled in the art would recognize
many variations, modifications, and alternatives.
[0147] When Eps(N(i)) is not zero, (Equ. 6) and (Equ. 7) are still
necessary but may no longer be sufficient conditions for
uninterrupted play out to be possible at the client. For example,
if the client loses up to a fraction p of the data it is mean to
receive for each segment, then it's actual reception rate may be as
small as Rd(1-p). In this instance, (Equ. 6) and (Equ. 7) can be
made necessary and sufficient conditions by substituting
.alpha.(1-p) for .alpha..
Optimal Server Bandwidth Scheduling (OSB)
[0148] Optimal server bandwidth scheduling (OSB) is one scheduling
method that can be used by the media object schedule generator 214,
when there is no restriction on the client download bandwidth Rd.
The media object schedule generator 214 first partitions the media
object into n blocks. Each segment comprises a single block, i.e.,
n'=n and N(i)=b(i), for i=0, . . . , n-1, and each segment is
served on one channel, so there is a schedule pair (I(i),E(i))
associated with each segment. Segment S(i) is served at a rate of
N(i)(1+Eps(N(i)))/T(i) Mbps, where T(i) is the time that block i
begins playing out. If the play out rate is Rp Mbps, then T
.function. ( i ) = Ts + j = 0 i - 1 .times. N .function. ( j ) Rp (
Equ . .times. 9 ) ##EQU5## The schedule pairs are I(i)=0 and
E(i)=T(i), for i=0, . . . , n-1. The client initially joins all n
channels and receives data from each channel concurrently, i.e.,
g=r=n. After a segment has been received completely, the client
drops the channel associated with that segment.
[0149] For example, the media object schedule generator 214 can
choose to partition the media object into blocks that are the
minimum possible size, i.e., N(i)=b(i)=Bmin, for i=0, . . . , n-1.
Each segment will have a play out time of Tf=Bmin/Rp seconds, so
the startup latency number m=Ts/Tf. If Eps(N(i))=0, then segment
S(i) is served at a rate of N(i)/T(i)=Bmin/(Ts+iBmin/Rp)=Rp/(i+m)
Mbps, for i=0, 1, . . . , n-1. The required server bandwidth is
therefore Rs = Rp .times. i = 0 n - 1 .times. 1 ( i + m ) ( Equ .
.times. 10 ) ##EQU6##
[0150] The harmonic sum H .function. ( n ) = i = 1 n .times. 1 / i
##EQU7## can be approximated as ln(n), so for large m,
Rs.apprxeq.Rpln((n+m-1)/(m-1)) Mbps, i.e., the server bandwidth
requirement grows logarithmically in n.
[0151] For example, consider a two hour movie with a play out rate
Rp=4 Mbps and a startup latency of 1 minute. If the minimum block
size Bmin=40 Mbits, then Tf=Bmin/Rp=10, n=720 and m=6. The required
server bandwidth is 19.52 Mbps. FIG. 12 shows the amount of
required server bandwidth as the length of the movie varies from 1
minute to 4 hours.
[0152] In another embodiment, the media object schedule generator
214 groups all the blocks in the media object into a single segment
and transmits the media object on a single channel, i.e., g=r=1. In
this embodiment, each block in the segment is served at a different
rate on the channel, where block i is served at a rate of
rb(i)=N(i)(1+Eps(N(i)))/T(i) Mbps. The schedule pair for the single
channel is I=0 and E=T(n-1). The client may filter out any output
symbols from a block that has completed downloading. Once the first
block has been received completely, the client starts to play out
the media object. Thus, the startup latency is equal to the time it
takes to download the first block.
[0153] In another embodiment, for instance when the number of
channels concurrently available to the client of the server is
restricted, the n blocks of the media object may be grouped into
n'=min{r, g} segments, where each segment is transmitted on a
single channel. In this embodiment, each block in the segment is
served at a different rate on the channel, where block i is served
at a rate of rb(i)=N(i)(1+Eps(N(i)))/T(i) Mbps. The schedule pairs
for segment S(i) are I(i)=0 and E(i)=T(j), where j is the highest
block index in segment S(i), for i=0, . . . , n'-1, i.e., the index
of the last block to start playing out in segment S(i).
Alternatively, each block in segment S(i) may be served at the same
rate of max{rb(j)=N(j) (i+Eps(N(j)))/T(j)|S(i) contains block j}.
Here, the schedule pairs for segment S(i) are I(i)=0 and E(i)=T(j),
where j is the smallest block index in segment S(i), for i=0, . . .
, n'-1, i.e., the index of the first block to start playing out in
segment S(i). The number of blocks in each segment may be chosen so
that each segment is transmitted at approximately the same rate.
Alternatively, the number of blocks in each segment may be chosen
so that each segment contains approximately the same number of
blocks, or is approximately the same size.
Restricted Client Download Bandwidth Scheduling Schemes
[0154] The schemes that are described below are different
scheduling methods that can be used by the media object schedule
generator 214 when there is a restriction on the maximum client
download rate Rd. A maximum client download rate is a realistic
concern for most MOD systems since the client's maximum download
rate depends on the type of connection it has to the network. For
example, the MOD system may want to serve a media object with a 300
Kbps play out rate to clients connected over DSL with a maximum
Rd=400 Kbps, so Rd=4Rp/3. As another example, the MOD system may
want to serve an MPEG-1 media stream with a play out rate of 1.4
Mbps to clients connected over a T1 line with a maximum download
rate of 1.5 Mbps.
Variable Rate Fixed Segment Size Scheduling (VRFS)
[0155] Variable rate fixed segment size (VRFS) scheduling is a
scheduling method that can be used by the media object schedule
generator 214 when there is a restriction on the maximum client
download rate Rd. In one embodiment, the VRFS scheduler partitions
the media object into blocks and segments in a pre-processing step.
In another embodiment, the VRFS scheduler determines the next
segment size and blocks in the segment for the media object when
the VRFS scheduler is ready to schedule the next segment, or when
enough data for a new segment becomes available.
[0156] In one embodiment, the VRFS scheduler first partitions the
media object into blocks. For example, each block could be chosen
to be the minimum block size Bmin, or the blocks could be chosen
according to specific breaks in the play out. One or more blocks
are grouped together to form each segment. In another embodiment,
the VRFS scheduler first partitions the media object into segments.
Each segment is then partitioned into one or more blocks. In a
preferred embodiment, the blocks within each segment are all chosen
to be the same size. For VRFS scheduling, each segment is served on
one channel, so there is a schedule pair (I(i),E(i)) associated
with each segment.
[0157] The VRFS scheduler uses a greedy algorithm to determine when
to schedule a segment to be downloaded, where the segments are
scheduled to be downloaded in order When a segment S(i) completes
downloading at time E(i), the bandwidth used to download segment
S(i) is made available and the VRFS scheduler schedules as many new
segments to be downloaded as possible up to the maximum client
download bandwidth. Each segment is scheduled to start downloading
at the client as early as possible.
[0158] FIG. 13 is a simplified flow diagram of a method, according
to an embodiment of the invention, that may be implemented by a
media object schedule generator 214, using a VRFS scheduler, to
determine the rate and schedule pair for each segment in a media
object, so that the client can download the media object and play
it out uninterrupted. This diagram is merely for illustrative
purposes and is not intended to limit the scope of the claims
herein. First, in a step 1410, the schedule pair (I(-1), E(-1)) for
the pseudo-segment S(-1) is initialized to zero since the client
does not download this segment. The pseudo-segment S(-1) plays out
for Ts seconds so N(-1)=TsRp Mbits. Next in a step 1420 Ns(-1) and
Ns(0) are initialized to 0 and N(-1) respectively. In general, Ns
.function. ( i ) = j = - 1 i - 1 .times. N .function. ( i ) .times.
Mbits , ##EQU8## i.e., Ns(i) is the aggregate size of the segments
S(-1) to S(i-1). In another embodiment, Ns(i) can be
pre-calculated, for i=-1, 0, 1, . . . , n-1, if the size of all the
segments is known beforehand. If the play out rate is Rp Mbps, then
segment S(i) begins playing out Ns(i)/Rp seconds after the client
requests the content.
[0159] In step 1430, j, i and Rc are all initialized, where j
represent the segment S(j) in the schedule that is currently
playing out, i represents the next segment S(i) to be scheduled to
be downloaded while segment S(j) is playing out, and Rc represents
the download bandwidth scheduled for use at the client when segment
S(j) is playing out. At this point, no segment has been scheduled
for downloading, so i=0, and Rc=0. Initially, the pseudo-segment
S(-1) is playing out so j=-1.
[0160] In step 1440, the VRFS scheduler checks if there are any
more segments to be scheduled for downloading, i.e., if i<n, and
if the next segment to be scheduled to be downloaded is less than
the segment that is currently playing out, i.e., if i>j. In step
1450, the VRFS scheduler attempts to schedule the next segment S(i)
to be downloaded. The segment S(i) is designed to start downloading
when segment S(j) starts playing out and complete downloading when
segment S(i) is due to start playing out, i.e., I(i)=Ns(j)/Rp and
E(i)=Ns(i)/Rp. The serving rate r(i) for segment S(i) is chosen so
that N'(i) Mbits of data can be downloaded by the client in the
schedule time of E(i)-I(i) seconds.
[0161] Next in step 1460, the VRFS scheduler checks whether the
current available bandwidth at the client when segment S(j) is
playing out, which is Rd-Mbps, is large enough to start downloading
segment S(i) at rate r(i). If r(i).ltoreq.Rd-Rc, then the segment
S(i) can be downloaded when segment S(j) is playing out. In steps
1470 and 1475, Rc is updated to include the bandwidth required to
download segment S(i), Ns(i+1) is calculated, and the scheduler
increments i by 1, so the next segment can be scheduled for
downloading. If r(i)>Rd-Rc, then in steps 1480 and 1485, the
scheduler increments j by 1 to indicate that segment S(i) can not
be scheduled to be downloaded until at least the next segment
completes downloading and starts playing out. The bandwidth r(j)
that becomes available when segment S(j) completes downloading is
also subtracted from Rc, and the VRFS scheduler tries to reschedule
segment S(i) if i>j in steps 1440 and 1450.
[0162] In step 1490, the VRFS scheduler checks if all the segments
have been scheduled for downloading. If yes, then the VRFS
scheduler completes successfully in step 1494. If the VRFS
scheduler reaches a state where the next segment to be downloaded
is supposed to be the segment that is currently being played out,
i.e., i=j, then in step 1492, the scheduler declares that there is
no valid schedule using this scheduler.
[0163] For example, the media object schedule generator 214 can
choose to partition the media object into blocks that are the
minimum possible size, i.e., b(i)=Bmin, for i=0, . . . , n-1. Each
segment is one block and has a play out time of Tf=Bmin/Rp seconds,
so the startup latency number m=Ts/Tf. FIG. 14 shows how the VRFS
schedules the first 5 segments of content when Rd=1.25 Rp, m=2 and
Eps(N(i))=0. The x-axis shows time in units of blocks and the
y-axis shows the bandwidth in units of the play out bandwidth. The
height of each segment is equal to the rate that the segment is
downloaded by the client. The VRFS scheduler initially schedules
the first three segments at rates Rp/2 Mbps, Rp/3 Mbps and Rp/4
Mbps respectively. At this point, the client is downloading at an
aggregate rate of 1.08 Rp Mbps. Segment S(3) requires 0.2 Rp Mbps
to download so there is not enough bandwidth available to start
downloading segment S(3). When segment S(0) completes downloading,
the VRFS scheduler can schedule the client to start downloading
segment S(3) and segment S(4) at rates Rp/3 Mbps and Rp/4 Mbps
respectively. The VRFS scheduler first drops the bandwidth
associated with segment S(0), so the aggregate download rate when
segment S(0) is playing out is 1.17 Rp Mbps. The process continues
until all the segments have been scheduled to be downloaded.
[0164] As another example, consider again the two hour movie with a
play out rate Rp=4 Mbps, a startup latency of 1 minute, and a
segment size Bmin=40 Mbits, for all the segments. If Rd=2 Rp and
Eps(N(i))=0, then the VRFS scheduler schedules the first 35
segments to start downloading before block 0 starts to play out.
The play out begins when segment S(0) completes downloading. At
this point, the VRFS scheduler drops the bandwidth associated with
segment S(0), since it has completed downloading, and schedules
segments S(35) through S(40) for downloading. The last segment
S(719) is only scheduled to start downloading once segment S(137)
starts playing out. The required server bandwidth is Rs=21.95 Mbps,
which is about 2.43 Mbps greater than that required by OSB.
[0165] For the first 137 blocks played out, i.e., until the client
starts downloading the last segment in the content, the VRFS
scheduler uses about 99.9% of the available download bandwidth,
which is very efficient. When Rd is reduced to 1.5 Rp in the above
example, then the client downloads at about 99.8% of the maximum
download rate until the last segment has started downloading. The
required server bandwidth is 27.04 Mbps. In these examples, the
VRFS scheduler makes very efficient use of the available client
download bandwidth.
[0166] As the startup latency and/or the maximum client download
bandwidth are reduced, the VRFS scheduler may become less
efficient. In fact, since the VRFS scheduler is not guaranteed to
download at the maximum download rate Rd, it is possible to satisfy
the conditions in (Equ. 6) and (Equ. 7), and not actually be able
to achieve uninterrupted play out. For example, for equal sized
segments Bmin, m=2, n=5 and Rd=5 Rp/6, (Equ. 6) and (Equ. 7) are
satisfied, but the VRFS scheduler cannot schedule the five segments
to be downloaded to achieve uninterrupted play out at the
client.
[0167] In another embodiment, the VRFS scheduler can schedule the
segments to be downloaded in an arbitrary order, as long as each
segment completes downloading before it is scheduled to play out.
When a segment completes downloading, as many new segments as
possible are scheduled to start downloading at the client, until
all the segments have been scheduled to be downloaded.
[0168] In another embodiment, the segments can be served on more
than one channel. As before, once each segment completes
downloading, the VRFS scheduler tries to schedule as many segments
as possible for the client to download up to the maximum client
download bandwidth. If there is not enough bandwidth available to
completely download segment S(i) when segment S(j) is playing out,
then segment S(i) may be downloaded on more than one channel.
Alternatively, the server can avoid serving segment S(i) on
multiple channels and instead increase the size of segment S(i-1)
so that segment S(i-1) requires all the remaining download
bandwidth at the client when segment S(j) is playing out.
Fixed Rate Variable Segment Size Scheduling (FRVS)
[0169] Fixed Rate Variable Segment Size (FRVS) scheduling is a
scheduling method that can be used by the media object schedule
generator 214 when there is a restriction on the maximum number of
concurrent channels at the client r, and when there is a
restriction on the maximum client download rate Rd.
[0170] In one embodiment, the FRVS scheduler first partitions the
media object into n' segments, where each segment is further
partitioned into one or more blocks. In a preferred embodiment, the
blocks within each segment are all chosen to be the same size. Each
segment is served on one channel, so there is a schedule pair
(I(i),E(i)) associated with each segment. Each segment is served at
the same rate of b Mbps. If the maximum client download bandwidth
is Rd Mbps, then since r segments are downloaded concurrently,
b=Rd/r Mbps and the required server bandwidth is Rs=(n'Rd)/r
Mbps.
[0171] The FRVS scheduler schedules the segments to be downloaded
in order where the client is downloading from at most r channels at
a time. The client starts to play out the media object once the
segment S(0) has been received completely. Once segment S(i)
completes downloading, the client starts downloading segment
S(i+r), while i<n'-r. As each of the last r segments complete
downloading, there are no more segments to be downloaded so the
client's reception rate decreases. The FRVS scheduler chooses the
size of segment S(i) based on the constraint that segment S(i)
should finish downloading and be reassembled before segment S(i-1)
finishes playing out.
[0172] FIG. 15 is a simplified flow diagram of a method, according
to an embodiment of the invention, that may be implemented by a
media object schedule generator 214, using a FRVS scheduler, to
determine the size of each segment in a media object, so that the
client can download the media object and play it out uninterrupted.
This diagram is merely for illustrative purposes and is not
intended to limit the scope of the claims herein. First, in a step
1610, the N(-2), N(-3), . . . , N(-r) are initialized to zero.
These are the sizes of the r-1 of the r pseudo-segments pre-pended
by the FRVS scheduler to the sequence of segments. The other
segment is pseudo-segment S(-1) and plays out for Ts seconds so
N(-1)=TsRp Mbits. Segments S(-1), S(-2), . . . , S(-r) do not
appear in the schedule produced by the media object schedule
generator 214.
[0173] Next in step 1620, Ns(-1), Ns(-2), . . . , Ns(-r) are
initialized to zero and Ns(0) is initialized to N(-1). In general,
Ns .function. ( i ) = j = - r i - 1 .times. N .function. ( i )
.times. Mbits , ##EQU9## i.e., Ns(i) is the size of the first i
segments plus the size of the pseudo-segment S(-1) and the
pre-pended pseudo-segments of size 0. If the play out rate is Rp
Mbps, then segment S(i) begins playing out Ns(i)/Rp seconds after
the client requests the content. In step 1630, i is initialized,
where i represents the next segment to be scheduled to be
downloaded. At this point, no segment has been scheduled for
downloading, so i=0.
[0174] In step 1640, the FRVS scheduler checks if the size of the
media object S is greater than the cumulative size of all the
segments that have been scheduled so far excluding the r pre-pended
segments, i.e., S>Ns(i)-Ns(0). If no, then in step 1650, the
FRVS scheduler determines the size of the next segment S(i) to be
downloaded. First, the FRVS scheduler calculates the amount of data
N'(i) that a client with no loss would receive if it downloads
segment S(i) at rate b=Rd/r during the play out time of the
previous r segments. Then, the size of segment N(i) is chosen by
solving the equation N'(i)=N(i)(1+Eps(N(i))) for N(i). In step
1655, the FRVS checks if the segment N(i) is at least as large as
the minimum segment size. If it is not, then in step 1696, the
scheduler declares that there is no valid schedule using this
scheduler. If it is, then in step 1660, the schedule pair (I(i),
E(i)) is chosen. The segment S(i) is designed to start downloading
when segment S(i-r) starts playing out and complete downloading
when segment S(i) is due to start playing out, i.e.,
I(i)=Ns(i-r)/Rp and E(i)=Ns(i)/Rp. In steps 1670 and 1680, Ns(i+1)
is calculated, and the scheduler increments i by 1, so the next
segment can be scheduled for downloading.
[0175] When the cumulative size of the segments scheduled is
greater than or equal to the file size, the FRVS scheduler goes to
step 1690. In step 1690, i is decremented by 1 and in step 1692,
the FRVS scheduler adjusts the size of segment S(i) so that the sum
of the segments sizes N(0)+N(1)+N(2)+ . . . +N(n'-1)=S Mbits and
adjusts E(i) so that a client with no loss would download N'(i)
Mbits in time E(i)-I(i). Finally, in step 1694 the scheduler
completes.
[0176] If the last segment is smaller than the minimum block size,
then there are a number of ways for the FRVS scheduler to increase
the size. In one embodiment, the size of the last segment can be
increased by decreasing the download rate of the second last
segment to decrease it in size, and therefore increase the size of
the last segment by the same amount. In another embodiment, if the
last segment is small enough, the download rate of the second last
segment can be increased so that there is no need for another
segment.
[0177] For example, consider a 2 hour movie with a play out rate
Rp=4 Mbps, a startup latency of 1 minute, and a minimum block size
of 40 Mbits. If Rd=2 Rp, Eps(N(i))=0, and r=5, then n'=17 and the
segment sizes in Mbits are TABLE-US-00001 N(0) = 96 N(1) = 134.4
N(2) = 188.2 N(3) = 263.4 N(4) = 368.8 N(5) = 420.3 N(6) = 550.0
N(7) = 716.3 N(8) = 927.5 N(9) = 1193.2 N(10) = 1522.9 N(11) =
1964.0 N(12) = 2529.6 N(13) = 3254.9 N(14) = 4184.9 N(15) = 5382.9
N(16) = 5101.6
The required server bandwidth is Rs=27.2 Mbps, which is 7.68 Mbps
greater than OSB. When r is set to 12, there are 36 segments and
the server bandwidth is Rs=24 Mbps. The server bandwidth can be
further reduced by serving the last segment S(n'-1) at a lower
rate. For instance, for the above example where r=5, if the rate of
the last segment is reduced so that the last segment only completes
downloading when it is due to be played out, then the server
bandwidth is reduced by almost 1 Mbps. In another embodiment, the
rate b Mbps that each segment is served or the startup latency can
be reduced so that when the loop exits in step 1660,
Ns(i)-Ns(0)=S.
[0178] FIG. 16 shows the server bandwidth requirement for FRVS with
r=1, 2 and 5, in the above example when .alpha. is varied from 1 to
2. FIG. 17 shows the server bandwidth requirement for FRVS for the
above example when r varies from 1 to 12. If there is an additional
system constraint of a minimum block size of Bmin=40 Mbits, then
for the above example r should be less than or equal to 12, in
order to ensure that the first segment is at least as large as the
minimum block size. The server bandwidth generally decreases as r
increases although the additional server bandwidth savings also
decreases as r decreases.
[0179] If there is a restriction on the minimum segment size, then
it is still possible to design an FRVS scheme such that the client
can download up to r channels simultaneously as follows. Divide the
available bandwidth into r channels where each channel is assigned
an equal share of the client download bandwidth. As before, the
segment sizes that are already chosen determine the size of the
next segment. Each new segment is scheduled to be downloaded at an
aggregate rate of cRd/r, where c is an integer between 1 and r.
Generally, c should be chosen to be as small as possible but large
enough to ensure that the segment is at least as large as the
minimum segment size.
[0180] FIG. 18 shows an example of how the scheduler allocates the
available bandwidth to each segment when r=5, for the first 10
segments of the media object. In this example, the segments
increase in size, so the time to download future segments is
greater than for previous segments (since the segments also take
longer to play out). Therefore, future segments will require less
bandwidth to achieve the minimum segment size. The y-axis
represents the client download bandwidth and the x-axis represents
time. The length of each segment represents how long it is designed
to be downloaded by the client and the height represents the
bandwidth at which the client downloads the segment. For instance,
segment S(0) is downloaded from time 0 to Ts using 3 Rd/5 Mbps of
the bandwidth.
[0181] In this example, the scheduler will initially try to make
the segment S(0) to be N'(0)=RdTs/r Mbits in size. However, when
the FRVS solves N(0)=N'(0)/(1+Eps(N(0))), N(0) is less than the
minimum required block size. So, the scheduler tries to download
the first segment at rate cRd/r Mbps, incrementing c by 1 until
N'(0)=cRdTs/r Mbits allows N(0) to be greater than the minimum
segment size. In the figure, c=3. Similarly for segment S(1), c=2.
When segment S(0) completes downloading, there is enough time to
download segment S(2) at only 2 Rd/r Mbps and segment S(3) at Rd/r
Mbps. If segment S(3) is too small if downloaded at Rd/r Mbps, then
the FRVS scheduler could either make segment S(2) larger by
downloading it at 3 Rd/r Mbps, or assign S(3) a second channel when
S(1) completes downloading.
[0182] FIG. 19 shows an example of how the scheduler allocates the
available bandwidth to each segment for r=2. In this example, the
segments decrease in size, so the time to download future segments
is less than that for previous segments. Therefore, future segments
will require more bandwidth to achieve the minimum block size. In
FIG. 19, segment S(i+2) would not be large enough if it is only
downloaded at a rate of Rd/2 Mbps. The bandwidth available after
segment S(i+1) completes downloading is therefore also assigned to
segment S(i+2) so that S(i+2) achieves the minimum segment size.
Segments S(i+3) is downloaded at Rd Mbps.
[0183] In one embodiment, a segment that is served at a rate of
cRd/r Mbps is served on a single channel. In another embodiment, a
segment that is served at a rate of cRd/r Mbps is served on c
channels where each channel is served at a rate of Rd/r Mbps.
[0184] Since each segment should be downloaded completely before it
can be played out, the maximum storage requirement at the client is
lower bounded by the size of the largest segment. Thus, when the
client storage requirement is limited, it may be preferable to
increase the server bandwidth and the total number of segments in
the media object, and place an upper limit on the segment size. The
server bandwidth can be further divided so that a segment finishes
downloading as late as possible before it is scheduled to play
out.
[0185] In one embodiment, if the media object is first partitioned
into blocks where the block sizes are fixed, then the segment sizes
can be adjusted so that a segment contains an integer number of
blocks by decreasing or increasing the segment size. This can be
done by changing the rate that each segment is served. The size and
rate of segments scheduled later in the play out may be adjusted
accordingly.
Restricted Server Channel Scheduling (RSC)
[0186] Restricted server channel scheduling (RSC) is a scheduling
method that can be used by the media object schedule generator 214
when there is a restriction on the maximum number of concurrent
channels at the server g, the maximum client download rate Rd, and
the maximum number of concurrent channels at the client r. The RSC
scheduler partitions the media object into at most g segments,
where each segment is served on one or more channels such that the
maximum number of channels is g.
[0187] FIG. 20 is a simplified flow diagram of a method, according
to an embodiment of the invention, that may be implemented by a
media object schedule generator 214, using a RSC scheduler, to
determine the size of each segment in a media object, so that the
client can download the media object and play it out uninterrupted.
This diagram is merely for illustrative purposes and is not
intended to limit the scope of the claims herein. In this
embodiment, the RSC scheduler finds the schedule produced by the
FRVS scheduler with as many concurrent channels at the client as
possible, and with at most g channels at the server. For the FRVS
scheduler, when the number of concurrent channels at the client
increases, the number of concurrent channels at the server usually
increases as well, and the required server bandwidth usually
decreases.
[0188] First, in steps 2110 and 2120, the RSC scheduler initializes
r' to one and runs the FRVS scheduler on the media object with r'
concurrent channels at the client. Then, in step 2125, the RSC
scheduler checks if the FRVS found a valid schedule in step 2120,
i.e., if the FRVS scheduler in FIG. 15 completed in step 1694. If
no valid schedule was found, then in step 2196, the scheduler
declares that there is no valid schedule using this scheduler. If
it is, then in step 2130, the RSC scheduler checks if the number of
server channels g' used by the schedule created by the FRVS
scheduler is less than or equal to g, if r'<r, and if the
current schedule produced by the FRVS scheduler is a valid
schedule. If Yes, then in steps 2140 and 2150, the RSC scheduler
increments r' and runs the FRVS scheduler again with r' concurrent
channels at the client. Steps 2130, 2140 and 2150 repeat until the
FRVS scheduler creates a schedule for which g'>g, r'=r, or for
which the FRVS scheduler fails to produce a valid schedule. In step
2160, if r'=r, g'.ltoreq.g, and the last schedule produced by the
FRVS scheduler was a valid schedule, then the RSC scheduler can use
the last schedule found by the FRVS scheduler to serve the media
object so the scheduler finishes in step 2170. Otherwise, in step
2190, r' is decremented by 1. The RSC scheduler then uses the FRVS
schedule with r' concurrent channels at the client found in step
2192 and finishes in step 2194.
[0189] In one embodiment, after step 2192, the RSC scheduler
increases the number of segments in the FRVS schedule to be equal
to g' by splitting some of the segments into two or more segments
and adjusting the bandwidth to serve each segment accordingly. In
another embodiment, the RSC scheduler does not decrease r' in step
2190, and instead decreases the number of segments in the FRVS
schedule by combining segments and the bandwidth used to download
the segments. In another embodiment, the RSC scheduler can store
each schedule Sched(i) generated by the FRVS scheduler, for i=1, 2,
. . . r', and use the schedule Sched(i) that requires the least
server bandwidth Rs.
[0190] The RSC scheduler can also be used without a restriction on
the maximum number of concurrent channels at the client r by not
comparing r' to r in steps 2130 and 2160.
Client Temporary Storage Requirement
[0191] An additional constraint for the media object schedule
generator 214 to consider is the maximum client temporary storage
requirement when the media object is not being stored at the
client. When the client starts to play out block i, it has already
played out the first i blocks, so the client only needs to store
the data that has already been downloaded, but not yet played out.
Similarly, in the time that block i is played out, whatever
information that is downloaded about blocks i+1 to n-1 should be
stored at the client in temporary storage. Define Tf(i)=T(i+1)-T(i)
to be the play out time for block i and D(j,i) to be the fraction
of the time that block j is scheduled to be downloaded by the
client when block i is playing out. Note that since block j has
already been downloaded by the time block i is playing out for
j.ltoreq.i,D(j,i)=0, for j.ltoreq.i.
[0192] The additional storage required at the client for blocks i+1
to n-1 while block i is playing out is defined to be AS .function.
( i ) = Tf .function. ( i ) j = i - 1 n - 1 .times. rb .function. (
j ) D .function. ( j , i ) .times. .times. Mbits . ( Equ . .times.
11 ) ##EQU10## During the same time interval, block i plays out, so
b(i) Mbits of temporary storage is freed for use by other blocks.
Define M(i) to be the total client temporary storage requirement as
block i begins to play out, i.e., the amount of data that has been
downloaded but not yet played out. The storage requirement after
block i is played out can be calculated as the sum of the storage
requirement after block i-1 played out, and the difference between
the amount of data downloaded and the amount of data played out
when block i is playing out, i.e., M(i)=M(i-1)+AS(i)-b(i), for
0<i<n-1. The initial conditions is M(-1)=AS(-1), where AS(-1)
is defined as the storage requirement when the play out commences,
i.e., AS .function. ( - 1 ) = Ts j = 0 n - 1 .times. rb .function.
( j ) D .function. ( j , - 1 ) .times. .times. Mbits . ( Equ .
.times. 12 ) ##EQU11##
[0193] The maximum client temporary storage requirement is defined
to be Mmax=max{M(i)}. The temporary storage requirement at the
client is increasing when the amount of data downloaded during the
play out of block i is greater than the size of block i. If the
client reception rate is non-increasing over time, then Mmax occurs
at the first index i for which the average client reception rate is
less than the play out rate.
[0194] Since each segment should be downloaded completely before it
can be played out, the client maximum temporary storage requirement
is at least as large as the size of the largest segment. Thus, it
may be preferable to increase the server bandwidth and the total
number of segments the media content is broken into, and then place
an upper limit on the size of any one segment when the client
storage is limited.
[0195] One method of reducing the client temporary storage
requirement is to download the content as late as possible. This is
in conflict with the goal of reducing the server bandwidth Rs,
where generally a block is scheduled to be downloaded as slowly as
possible. Another method of reducing the client temporary storage
requirement, is to change the schedule at a point before the index
i, for which the storage requirement is a maximum.
[0196] In one embodiment, the segment or block with the highest
index that is downloaded before the index i for which Mmax=M(i)
will be scheduled to be downloaded at a later time, in order to
decrease the maximum client temporary storage requirement. In
another embodiment, the segment or block which uses the greatest
amount of temporary storage at the index i for which Mmax=M(i) will
be scheduled to be downloaded at a later time, in order to decrease
the maximum client temporary storage requirement. In a further
embodiment, a subset of segments are downloaded later to reduce the
maximum storage requirement, where these segments are selected
based on factors such as when these segments are actually required
to be played out by the media object player, and how much of the
temporary storage requirement these segments are using at time T(i)
for which Mmax=M(i).
[0197] For an OSB scheduler for which the media object schedule
generator 214 partitions the media object into segments of the
minimum block size Bmin, for i=0, . . . , n-1, when the client
starts to play out block i, the time already spent downloading the
last n-i blocks is equal to (i+m) Tf seconds, i.e., the time to
play out the first i blocks plus the startup time, Ts=mTf. The
maximum storage requirement at the client, occur for the smallest i
such that M(i+1)<M(i), or when i.apprxeq.(n+m-1)/(e)-m.
Therefore, the maximum client storage requirement is
Mmax.apprxeq.(nTfRp)/e. Since the size of the media content is
nTfRp Mbits, the client should store a maximum of approximately 37%
of the entire media content. To reduce the storage requirement, the
client can download segments S((n+m-1)/e-m) to S(n-1) at a later
point in the schedule, and at a higher rate.
[0198] For example, consider a two hour movie with a play out rate
Rp=4 Mbps and a startup latency of 1 minute. If the minimum block
size Bmin=40 Mbits, then Tf=Bmin/Rp=10, n=720 and m=6. If
Eps(N(i))=0, then the maximum temporary storage requirement for OSB
is 9.1 Gbits, which is about 32% of the media object, which in this
example is slightly less than the approximation above of 37%.
[0199] For a VRFS, FRVS and RSC scheduler, the maximum storage
requirement at the client can be upper bounded as follows. Consider
the first point in the schedule T(i) for which the designed client
reception rate is less than the play out rate for the media object,
and at which all the segments have already been scheduled for
downloading, i.e., I(i).ltoreq.T(i) for i=0, 1, . . . , n'-1. The
maximum amount of data that a client could have downloaded but not
yet played out at this point is (Rd-Rp)T(i), so
Mmax.ltoreq.(Rd-Rp)T(i) Mbits.
[0200] For example, consider again the two hour movie with a play
out rate Rp=4 Mbps, a startup latency of 1 minute, and a segment
size Bmin=40 Mbits, for all the segments. If Rd=2 Rp and
Eps(N(i))=0, then the maximum temporary storage requirement for
VRFS is 6.2 Gbits, which is about 22% of the media object. Note
that the maximum temporary storage requirement for VRFS is 2.9
Gbits less than for OSB, since in this example VRFS schedules all
but the first 35 segments for downloading at a later point in the
schedule and a higher rate than the OSB scheduler.
[0201] The designed client reception rate is less than the play out
rate for the first time as block 172 begins to play out. The last
segment is scheduled for downloading when block 137 starts playing
out, so T(172)=Ts+172Tf=1780 seconds. The general upper bound above
yields Mmax.ltoreq.(8-4)1780=7.1 Gbits which is about 15% greater
than the actual temporary storage requirement.
Maximum Server Bandwidth Requirements and Transitioning Between
Serving Media Objects
[0202] When the maximum server bandwidth Rsmax is limited, then the
server will have a maximum number of media objects that can be
served simultaneously. In order to add a media object, the server
may have to stop serving one or more of the media objects currently
being served. The transition of dropping one or more media objects
and starting to serve another should be done so that users can view
each media object for as long as possible.
[0203] In one embodiment, for the media objects being dropped, the
server stops serving the segments in order, i.e., the segments from
the beginning stop being served first, and for the media object
being added, the server starts serving the segments in order i.e.,
the segments from the beginning are served first.
[0204] Consider a media object that a client may start downloading
and playing out on demand starting at time t. Since the schedule is
designed so that segment S(i) starts downloading at the client I(i)
seconds after the client requests the content, the server should
start serving segment S(i) by time t+I(i) for i=0, 1, . . . , n'-1.
The bandwidth at the server required to serve segment S(i) is
therefore only needed at time t+I(i).
[0205] For each media object that is scheduled to stop being
served, enough segments from these media objects have to stop being
served by time t+I(i), so that there is j = 0 i - 1 .times. r
.function. ( j ) .times. Mbps ##EQU12## of bandwidth available at
the server to serve segments S(0) thru S(i), for i=0, 1, . . . ,
n'-1.
[0206] In one embodiment, the next segment S(j) that will stop
being served is the segment S(j) among all the media objects that
are no longer to be served with the smallest E(j). Define Te(j) to
be the time that segment S(j) stops being served at the server.
Define T0(j)=Te(j)-E(j) to be the latest startup time at which a
client can request the media object and still complete downloading
segment S(j). A media object that is no longer being served will be
no longer available to a client for downloading at the smallest
time T0(j), for j=0, 1, . . . , n'-1 for that media object.
[0207] For example, consider a pair of 2 hour movies with the same
play out rate Rp=4 Mbps, the same startup latency of 1 minute, and
the same minimum block size of 40 Mbits. If Rd=2 Rp, Eps(N(i))=0,
r=5, and the FRVS scheduler is used to schedule both movies, then
both movies are partitioned into n'=17 segments, where the segment
S(i) is the same size and scheduled to be served at the same rate
for each movie. The required server bandwidth for each movie is
Rs=27.2 Mbps. If the maximum server bandwidth Rsmax=27.2 Mbps, then
only one of these movies can be served at a time from this server.
The schedule pairs for the two movies are TABLE-US-00002 (I(0),
E(0)) = (0, 60) (I(1), E(1)) = (0, 84) (I(2), E(2)) = (0, 118)
(I(3), E(3)) = (0, 165) (I(4), E(4)) = (0, 231) (I(5), E(5)) = (60,
323) (I(6), E(6)) = (84, 428) (I(7), E(7)) = (118, 565) (I(8),
E(8)) = (165, 744) (I(9), E(9)) = (231, 976) (I(10), E(10)) = (323,
1274) (I(11), E(11)) = (428, 1655) (I(12), E(12)) = (565, 2146)
(I(13), E(13)) = (744, 2779) (I(14), E(14)) = (976, 3592) (I(15),
E(15)) = (1274, 4639) (I(16), E(16)) = (1655, 4844)
[0208] Consider the case where the second movie is scheduled to
start at t=7:00 pm. If the server stops serving all the segments
for the first movie, and starts serving all the segments for the
second movie at t=7:00 pm, then any client that has not downloaded
the entire first movie at this point will be unable to finish
watching the movie, i.e., any client that is not watching the last
segment S(16) of the first movie at t=7:00 pm. Since segment S(16)
is N'(16)=N(16)=5101.6 Mbits and takes N'(16)/Rp=5101.6/4=1275.4
seconds to play out, which is approximately 21 minutes, any client
that requests the movie after approximately 5:20 pm will not be
able to watch the whole movie. Therefore, the server should stop
serving the movie to new clients after approximately 5:20 pm.
[0209] Alternatively, if the server starts serving the segment S(i)
of the second movie I(i) seconds after 7:00 pm, and stops serving
the segment S(i) of the first movie at this time, then the segments
of the first movie stop serving in order. Similarly, the segments
from the second movie start serving in order. A segment in the
second movie only starts being served, when it is first possible
for any client to request that segment to be downloaded. Since r=5,
at 7:00 pm the server stops serving the segments S(0) thru S(4) of
the first movie. At 7:01 pm, the server stops serving the segment
S(5) of the first movie, and so on until segment S(16) of the first
movie stops being served at about 7:28 pm. For this example, the
time Te(i)=t+I(i) for each segment and the time
T0(i)=Te(i)-E(i)=t+I(i)-E(i) for i=0, 1, . . . , 16. The minimum
T0(i) is for i=15 where T0(15) is approximately 6:04 pm. Therefore
in this example, the first movie can be served for an extra 44
minutes without affecting the delivery of the second movie.
Incremental Scheduling of a Media Object
[0210] All of the above scheduling methods, with the exception of
OSB, do not require the scheduler to know the total length of the
media object in order to schedule the media object since the
scheduler can be run incrementally as the rest of the media object
arrives to be served. Because of this, all of these schedulers can
be used to perform on demand scheduling for a media stream. The
stream is partitioned into blocks and segments as the content
becomes available to the server. The media object description at
the client can be updated by information sent on the channels that
it has already joined. Alternatively, the client can use the client
browser to update the media object description, or can implement
the same functionality as the media object schedule generator 214
to determine what to do next.
[0211] It is to be understood that, in the above description, many
of the steps and/or procedures may be performed in a concurrent
manner (i.e., pipelined) in order to increase throughput. For
example, with an MOD server, all the output symbols in a segment
need not first be generated before transmitting any of the output
symbols in the segment. Rather, output symbols of one block may be
transmitted concurrently with the generation of output symbols in
another block. Also, completed output symbols of a block may be
transmitted concurrently with the generation of other output
symbols of the same block. As another example, with an MOD client,
all of the blocks in a segment need not be first reassembled before
playing out any of the blocks in the segment. Rather, one block may
start playing out concurrently with the reassembly of another
block. Similarly, the downloading of one segment may occur
concurrently with the reassembling and/or playing out of blocks in
a second segment. In yet another example, the reassembling of
blocks in one segment may occur concurrently with the playing out
of blocks in a second segment. One skilled in the art will
recognize many other steps in the above description that may
similarly be concurrently performed.
[0212] Additionally, it is to be understood that the various
functional blocks in FIGS. 1-4, 5-6, may be implemented by a
combination of hardware and/or software, and that in specific
implementations some or all of the functionality of some of the
blocks may be combined. Similarly, it is also to be understood that
the various methods discussed herein may be implemented by a
combination of hardware and/or software.
[0213] The above description is illustrative and not restrictive.
Many variations of the invention will become apparent to those of
skill in the art upon review of this disclosure. The scope of the
invention should, therefore, be determined not with reference to
the above description, but instead should be determined with
reference to the appended claims along with their full scope of
equivalents.
* * * * *