U.S. patent application number 17/234246 was filed with the patent office on 2021-08-05 for cross-layer troubleshooting of application delivery.
The applicant listed for this patent is ThousandEyes LLC. Invention is credited to Ryan Braud, Mohit V. Lad, Michael Meisel, Ricardo V. Oliveira.
Application Number | 20210243099 17/234246 |
Document ID | / |
Family ID | 1000005523359 |
Filed Date | 2021-08-05 |
United States Patent
Application |
20210243099 |
Kind Code |
A1 |
Lad; Mohit V. ; et
al. |
August 5, 2021 |
CROSS-LAYER TROUBLESHOOTING OF APPLICATION DELIVERY
Abstract
Techniques for cross-layer troubleshooting of application
delivery are disclosed. In some embodiments, cross-layer
troubleshooting of application delivery includes collecting test
results from a plurality of distributed agents for a plurality of
application delivery layers; and generating a graphical
visualization of an application delivery state based on the test
results for the plurality of application delivery layers (e.g.,
different application delivery layers).
Inventors: |
Lad; Mohit V.; (San
Francisco, CA) ; Oliveira; Ricardo V.; (San
Francisco, CA) ; Meisel; Michael; (San Francisco,
CA) ; Braud; Ryan; (San Francisco, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
ThousandEyes LLC |
Wilmington |
DE |
US |
|
|
Family ID: |
1000005523359 |
Appl. No.: |
17/234246 |
Filed: |
April 19, 2021 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
16251667 |
Jan 18, 2019 |
10986009 |
|
|
17234246 |
|
|
|
|
13839214 |
Mar 15, 2013 |
10230603 |
|
|
16251667 |
|
|
|
|
61649473 |
May 21, 2012 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
H04L 43/08 20130101;
H04L 41/22 20130101; H04L 41/0631 20130101; H04L 41/042 20130101;
H04L 43/062 20130101; H04L 43/045 20130101; H04L 43/0852 20130101;
G06F 11/079 20130101 |
International
Class: |
H04L 12/26 20060101
H04L012/26; G06F 11/07 20060101 G06F011/07; H04L 12/24 20060101
H04L012/24 |
Claims
1. A method, comprising: receiving, by a device, a first set of
application performance test results indicative of application
delivery performance of a distributed application at a network
layer or a transport layer; receiving, by the device, a second set
of application performance test results indicative of application
delivery performance of the distributed application at an
application layer; correlating, by the device, the first set of
application performance test results and the second set of
application performance test results using one or more shared
criteria; and identifying, by the device, a cross-layer performance
issue associated with delivery of the distributed application based
on the correlated first set of application performance test results
and the second set of application performance test results.
2. The method as in claim 1, wherein distributed agents perform one
or more active is measurements of end-to-end network properties
associated with the distributed application to generate the first
set of application performance test results and the second set
application performance test results.
3. The method as in claim 2, wherein the distributed agents are
located in different geographies for performing different types of
tests.
4. The method as in claim 2, wherein the distributed agents are
located in different geographies for targeting different sites,
locations, or metrics.
5. The method as in claim 2, wherein the distributed agents do not
instrument a destination site associated with the distributed
application.
6. The method as in claim 1, wherein the one or more shared
criteria are selected from the group consisting of: a time domain
and a space domain.
7. The method as in claim 1, further comprising: causing, by the
device, a graphical visualization of the cross-layer performance
issue associated with delivery of the distributed application to be
displayed on a graphical user interface (GUI).
8. The method as in claim 7, wherein the graphical visualization
comprises a network topology that allows a user to drill down into
one or more of a plurality of locations at different layers
including the network layer, the transport layer, and the
application layer.
9. The method as in claim 7, wherein the graphical visualization
includes a timeline that comprises a navigation widget that enables
a user to view an aggregate behavior of one or more is distributed
agents over time at a particular layer.
10. An apparatus, comprising: one or more network interfaces to
communicate with a distributed application of a computer network; a
processor coupled to the network interfaces and configured to
execute one or more processes; and an apparatus memory configured
to store a process executable by the processor, the process when
executed operable to: receive a first set of application
performance test results indicative of application delivery
performance of the distributed application at a network layer or a
transport layer; receive a second set of application performance
test results indicative of application delivery performance of the
distributed application at an application layer; correlate the
first set of application performance test results and the second
set of application performance test results using one or more
shared criteria; and identify a cross-layer performance issue
associated with delivery of the distributed application based on
the correlated first set of application performance test results
and the second set of application performance test results.
11. The apparatus of claim 10, wherein distributed agents perform
one or more active measurements of end-to-end network properties
associated with the distributed application to generate the first
set of application performance test results and the second set
application performance test results.
12. The apparatus of claim 11, wherein the distributed agents are
located in different geographies for performing different types of
tests.
13. The apparatus of claim 11, wherein the distributed agents are
located in different geographies for targeting different sites,
locations, or metrics.
14. The apparatus of claim 11, wherein the distributed agents do
not instrument a destination site associated with the distributed
application.
15. The apparatus of claim 10, wherein the one or more shared
criteria are selected from the group consisting of: a time domain
and a space domain.
16. The apparatus of claim 10, the process when executed further
operable to: cause a graphical visualization of the cross-layer
performance issue associated with delivery of the distributed
application to be displayed on a graphical user interface
(GUI).
17. The apparatus of claim 16, wherein the graphical visualization
comprises a network topology that allows a user to drill down into
one or more of a plurality of locations at different layers
including the network layer, the transport layer, and the
application layer.
18. The apparatus of claim 16, wherein the graphical visualization
includes a timeline that comprises a navigation widget that enables
a user to view an aggregate behavior of one or more distributed
agents over time at a particular layer.
19. A tangible, non-transitory, computer-readable medium storing
program instructions that cause a device in communication with a
distributed application to execute a process comprising: receiving,
by a device, a first set of application performance test results
indicative of application delivery performance of a distributed
application at a network layer or a transport layer; receiving, by
the device, a second set of application performance test results
indicative of application delivery performance of the distributed
application at an application layer; correlating, by the device,
the first set of application performance test results and the
second set of application performance test results using one or
more shared criteria; and identifying, by the device, a cross-layer
performance issue associated with delivery of the distributed
application based on the correlated first set of application
performance test results and the second set of application
performance test results.
20. The tangible, non-transitory, computer-readable medium of claim
19, the process further comprising: causing, by the device, a
graphical visualization of the cross-layer performance issue
associated with delivery of the distributed application to be
displayed on a graphical user interface (GUI).
Description
CROSS REFERENCE TO OTHER APPLICATIONS
[0001] This application is a continuation of U.S. patent
application Ser. No. 16/251,667, entitled CROSS-LAYER
TROUBLESHOOTING OF APPLICATION DELIVERY, filed on Jan. 18, 2019,
and U.S. patent application Ser. No. 13/839,214, entitled
CROSS-LAYER TROUBLESHOOTING OF APPLICATION DELIVERY, filed Mar. 15,
2013, now U.S. Pat. No. 10,230,603, which claims priority to U.S.
Provisional Patent Application No. 61/649,473, entitled CROSS-LAYER
VISIBILITY OF APPLICATION DELIVERY, filed May 21, 2012, all of
which are incorporated herein by reference for all purposes.
BACKGROUND OF THE INVENTION
[0002] Cloud computing generally refers to the use of computing
resources (e.g., hardware and software) that are delivered as a
service over a network (e.g., typically, the Internet). Cloud
computing includes using remote services to provide a user's data,
software, and computation.
[0003] Distributed applications can generally be delivered using
cloud computing techniques. For example, distributed applications
can be provided using a cloud computing model, in which users are
provided access to application software and databases over a
network. The cloud providers generally manage the infrastructure
and platforms on which the applications run. Various types of
distributed applications can be provided as a software as a service
(SaaS).
BRIEF DESCRIPTION OF THE DRAWINGS
[0004] Various embodiments of the invention are disclosed in the
following detailed description and the accompanying drawings.
[0005] FIG. 1 illustrates a functional block diagram of a platform
for cross-layer troubleshooting of application delivery in
accordance with some embodiments.
[0006] FIG. 2 illustrates a network performance visualization in
accordance with some embodiments.
[0007] FIG. 3 illustrates a graph of a visualization of a data path
to a destination site in accordance with some embodiments.
[0008] FIG. 4A illustrates a graph of a full connectivity
visualization of a data path to a destination site in accordance
with some embodiments.
[0009] FIG. 4B illustrates a graph providing a simplified view of a
visualization of a data path to a destination site in accordance
with some embodiments.
[0010] FIG. 4C illustrates a graph providing a simplified view of a
visualization of a data path to a destination site with problem
nodes in accordance with some embodiments.
[0011] FIG. 4D illustrates a graph providing a selective
exploration expanding a data path from selected source/agent site
in accordance with some embodiments.
[0012] FIG. 5A is screen shot illustrating an HTTP availability
drop during a particular time interval in accordance with some
embodiments.
[0013] FIG. 5B is screen shot illustrating a network view that
shows packet loss from the agents shown in FIG. 5A in accordance
with some embodiments.
[0014] FIG. 5C is screen shot illustrating a visualization of a
problem zone shown in FIG. 5B in accordance with some
embodiments.
[0015] FIG. 6 illustrates a graph of a visualization of routing
paths in accordance with some embodiments.
[0016] FIG. 7 illustrates a graph of a visualization of routing
paths including a series of nodes indicating data loss in
accordance with some embodiments.
[0017] FIG. 8A and FIG. 8B illustrate views for showing Border
Gateway Protocol (BGP) and packet loss in accordance with some
embodiments.
[0018] FIG. 9 illustrates a visualization of DNS data in accordance
with some embodiments.
[0019] FIG. 10 illustrates Anycast debugging, to identify any
relevant DNS problems, in accordance with some embodiments.
[0020] FIG. 11 illustrates a path visualization that shows all
problem agents routing to an Ashburn, Va. instance in accordance
with some embodiments.
[0021] FIG. 12 illustrates a visualization of HTTP data in
accordance with some embodiments.
[0022] FIG. 13 illustrates a visualization of a web page load
performance in accordance with some embodiments.
[0023] FIG. 14 illustrates a bottleneck analysis by provider for
page loads in accordance with some embodiments.
[0024] FIG. 15 illustrates a visualization of transactions (e.g.,
web transactions) in accordance with some embodiments.
[0025] FIG. 16 shows a summary of transaction steps and identifying
bottlenecks in accordance with some embodiments.
[0026] FIG. 17 illustrates a flow diagram for cross-layer
troubleshooting of application delivery in accordance with some
embodiments.
[0027] FIG. 18 illustrates another flow diagram for cross-layer
troubleshooting of application delivery in accordance with some
embodiments.
DETAILED DESCRIPTION
[0028] The invention can be implemented in numerous ways, including
as a process; an apparatus; a system; a composition of matter; a
computer program product embodied on a computer readable storage
medium; and/or a processor, such as a processor configured to
execute instructions stored on and/or provided by a memory coupled
to the processor. In this specification, these implementations, or
any other form that the invention may take, may be referred to as
techniques. In general, the order of the steps of disclosed
processes may be altered within the scope of the invention. Unless
stated otherwise, a component such as a processor or a memory
described as being configured to perform a task may be implemented
as a general component that is temporarily configured to perform
the task at a given time or a specific component that is
manufactured to perform the task. As used herein, the term
`processor` refers to one or more devices, circuits, and/or
processing cores configured to process data, such as computer
program instructions.
[0029] A detailed description of one or more embodiments of the
invention is provided below along with accompanying figures that
illustrate the principles of the invention. The invention is
described in connection with such embodiments, but the invention is
not limited to any embodiment. The scope of the invention is
limited only by the claims and the invention encompasses numerous
alternatives, modifications and equivalents. Numerous specific
details are set forth in the following description in order to
provide a thorough understanding of the invention. These details
are provided for the purpose of example and the invention may be
practiced according to the claims without some or all of these
specific details. For the purpose of clarity, technical material
that is known in the technical fields related to the invention has
not been described in detail so that the invention is not
unnecessarily obscured.
[0030] Cloud computing generally refers to the use of computing
resources (e.g., hardware and software) that are delivered as a
service over a network (e.g., typically, the Internet). Cloud
computing includes using remote services to provide a user's data,
software, and computation.
[0031] Distributed applications can generally be delivered using
cloud computing techniques. For example, distributed applications
can be provided using a cloud computing model, in which users are
provided access to application software and databases over a
network. The cloud providers generally manage the infrastructure
and platforms on which the applications run. Various types of
distributed applications can be provided as a software as a service
(SaaS).
[0032] Users typically access cloud-based distributed applications
(e.g., distributed applications) through a web browser, a
light-weight desktop, and/or mobile application (e.g., mobile app)
while the enterprise software and user's data are typically stored
on servers at a remote location. Such cloud-based distributed
applications can allow enterprises to get their applications up and
running faster, with improved manageability and less maintenance,
and can enable enterprise IT to more rapidly adjust resources to
meet fluctuating and unpredictable business demand. Thus,
distributed applications can allow a business to reduce Information
Technology (IT) operational costs by outsourcing hardware and
software maintenance and support to the cloud provider.
[0033] However, a significant drawback of distributed applications
is that troubleshooting performance problems can be very
challenging and time consuming. For example, determining whether
performance problems are the result of the cloud provider of the
distributed application, the customer's own internal IT network, a
user's client device, and/or intermediate network providers between
the user's client device and the cloud provider can present
significant challenges.
[0034] What are needed are new techniques to visualize and
troubleshoot the performance of distributed applications.
[0035] Accordingly, techniques for cross-layer troubleshooting of
application delivery are disclosed. In some embodiments, various
techniques are provided for cross-layer visualization and
troubleshooting of application delivery, such as for performance
problems associated with distributed applications. For example, a
platform for identifying and/or determining performance problems
associated with a distributed application(s) can be provided. As
another example, the platform can generate reports that include
various cross-layer visualizations that facilitate identifying
and/or determining performance problems associated with a
distributed application(s). As yet another example, various
techniques described herein can be used to diagnose application
deliver problems from cloud service providers, such as SaaS and/or
other network delivered based applications (e.g., web sites, online
stores, cloud based software, and/or other such network based
applications and/or services) to determine the causes or sources of
the application delivery performance issues or problems.
[0036] In some embodiments, cross-layer troubleshooting of
application delivery includes collecting test results from a
plurality of distributed agents for a plurality of application
delivery layers; and generating a graphical visualization of an
application delivery state based on the test results for the
plurality of application delivery layers (e.g., different
application delivery layers).
[0037] For example, the graphical visualization of the application
delivery state can facilitate cross-layer troubleshooting of
problems (e.g., associated with application delivery of a
distributed application). As another example, the graphical
visualization of the application delivery state can facilitate
cross-layer visualization and troubleshooting of application
delivery (e.g., associated with application delivery of a
distributed application, which can identify and/or facilitate
diagnosing of causes of application delivery problems).
[0038] In some embodiments, cross-layer troubleshooting of
application delivery further includes outputting the graphical
visualization of the application delivery state based on the test
results for the plurality of application delivery layers.
[0039] In some embodiments, the graphical visualization of the
application delivery state based on the test results for the
plurality of application delivery layers facilitates cross-layer
visualization and troubleshooting of problems (e.g., associated
with application delivery of a distributed application).
[0040] In some embodiments, the graphical visualization of the
application delivery state based on the test results for the
plurality of application delivery layers facilitates cross-layer
troubleshooting of application delivery by providing for
correlation of the test results across a plurality of layers (e.g.,
application delivery layers).
[0041] In some embodiments, the graphical visualization of the
application delivery state based on the test results for the
plurality of layers facilitates cross-layer troubleshooting of
application delivery by providing for correlation of the test
results across a plurality of layers using a space domain and/or a
time domain.
[0042] In some embodiments, the plurality of distributed agents are
controlled by an agent controller. In some embodiments, the
graphical visualization of network performance is generated by a
platform for cross-layer visibility and troubleshooting of
distributed applications 100, such as shown in FIG. 1. In some
embodiments, the network performance test results from the
plurality of distributed agents for the plurality of layers are
stored in a storage tier (e.g., which can include a database or
another type of data store).
[0043] Overview of Techniques for Cross-Layer Visibility and
Troubleshooting of Distributed Applications
[0044] In some embodiments, cross-layer visibility and
troubleshooting of distributed applications includes using software
agents to collect information from different points in a network
across different layers in the application delivery (e.g., of a
distributed application), as further described herein with respect
to various embodiments. For example, such information can be
collected and aggregated by centralized collectors and aggregators
and presented to a user as a Software as a Service (SaaS). In some
embodiments, different layers are correlated using one or more of
the following: a monitored object, time, and location.
[0045] In some embodiments, various techniques described herein
allow users to drill down to identify problems and locations
whether at different layers (e.g., a network, transport, and/or
application layer) to correlate across application delivery layers
to determine whether, for example, network issues are affecting the
performance of a distributed application(s), as further described
herein with respect to various embodiments. For example, such
techniques can be applied to both internal network diagnostics
(e.g., for an organization's internal network) and/or external
network diagnostics (e.g., for a web site accessed across the
Internet, such as for a cloud-based distributed application).
[0046] In some embodiments, various techniques described herein
allow for determination of a network topology to indicate, for
example, anycast routing issues used by DNS providers (e.g., such
as when there are multiple DNS servers sharing the same IP address)
as well as structure and/or activity of a network topology, as
further described herein with respect to various embodiments. For
example, such techniques can be applied to both internal network
diagnostics (e.g., for an organization's internal network) and/or
external network diagnostics (e.g., for a web site accessed across
the Internet, such as for a cloud-based distributed
application).
[0047] In some embodiments, various techniques described herein
allow for diagnosing SaaS applications, such as for cloud-based
distributed applications, using, for example, a lightweight agent,
as further described herein with respect to various
embodiments.
[0048] A Distributed Testing Framework
[0049] In some embodiments, cross-layer visibility and
troubleshooting of distributed applications includes a distributed
framework to distribute tests across different agents in the
Internet. For example, agents can be executed on hosted providers
using cloud computing distributed across multiple ISPs, which are
controlled by agent controllers to perform one or more tests as
further described herein, in which the test results can be
collected for correlation and analysis, as further described herein
with respect to various embodiments. In some embodiments, agents
are computing resources that are controlled, and, for example, can
be either virtual or dedicated servers. Agents can be distributed
across different geographies and networks, for example, distributed
agents can be distributed to mostly Tier-1 and Tier-2 networks to
avoid the noise of bad connectivity of last mile connections.
[0050] An example of a system architecture for providing
cross-layer visibility and troubleshooting of distributed
applications is shown in FIG. 1 as described below.
[0051] FIG. 1 illustrates a functional block diagram of a platform
for cross-layer troubleshooting of application delivery in
accordance with some embodiments. In particular, FIG. 1 illustrates
an environment in which a platform for cross-layer visibility and
troubleshooting of distributed applications 100 includes
distributed agents 116-120 (e.g., which can be distributed across
various geographies and/or devices for performing different types
of tests and/or targeting different sites, locations, and/or
metrics) that collect data based on configured tests, and the
distributed agents 116-120 send this data to a controller(s) 114
(e.g., agent controllers). The controller 114 stores the data in a
storage tier 112 (e.g., providing permanent storage) that can be
used by a web tier 104 to generate visualizations and reports to
users accessing the platform 100 using client devices (e.g.,
computers, laptops, smartphones, and/or various other computing
devices).
[0052] For example, a report can be output to a user to present the
collected and analyzed cross-layer application delivery information
of a distributed application. Example reports can include various
visualizations and/or diagnostic information as further described
herein with respect to various embodiments. For example, the report
can facilitate troubleshooting application delivery associated with
the distributed application to determine whether performance
problems are the result of the cloud provider of the distributed
application, the customer's own internal IT network, a user's
client device, and/or intermediate network providers between the
user's client device and the cloud provider. The report can also
include recommendations to the user to resolve any such determined
application delivery problems associated with the distributed
application. In some cases, the report can also be provided to a
third party, such as the SaaS provider of the distributed
application and/or a network provider, which can be provided as
information to indicate the source of such determined application
delivery problems associated with the distributed application.
[0053] In the example shown, the user of client device 106
(hereinafter referred to as "Bob") is employed as an IT manager of
a distributed application company ("SaaS Company"). The user of
client device 108 (hereinafter referred to as "Alice") is employed
as an IT manager of a national company ("ACME Company"). As will be
described in more detail below, Bob and Alice can each access the
services of platform 100 (e.g., platform for cross-layer visibility
and troubleshooting of distributed applications) via web tier 104
over a network, such as the Internet. The techniques described
herein can work with a variety of client devices 106-108 including,
but not limited to personal computers, tablet computers,
smartphones, and/or other computing devices.
[0054] In some embodiments, platform 100 generates various reports
based on results of the network performance tests to facilitate
cross-layer visibility and troubleshooting of application delivery
associated with a distributed application(s), as further described
herein. In some embodiments, platform 100 includes a data store,
such as storage tier 112 for storing results of the network
performance tests and/or the reports.
[0055] In some embodiments, a set of agent controllers 114 is
provided as shown to send various tests (e.g., such as the various
test described herein with respect to various embodiments) to the
distributed agents for execution by the distributed agents. For
example, agents can be executed on hosted providers using cloud
computing distributed across multiple ISPs, which are controlled by
agent controllers to perform one or more tests as further described
herein, in which the test results can be collected for correlation
and analysis, as further described herein with respect to various
embodiments.
[0056] In some embodiments, the tests are configured through a web
interface by a user. For example, typical parameters can include
the frequency of various tests, the target of the tests, and the
agents (e.g., or locations) where the tests are to be performed.
The test parameters can be sent from the controller (e.g., agent
controllers 114) to the distributed agents after an agent checks-in
(e.g., using a pull mechanism). After an agent executes a test, the
agent can export the test result(s) back to the controller. The
controller can then provide the results back to a data store (e.g.,
storage tier 112) for permanent storage (e.g., or temporary
storage). Besides periodic tests, a controller can also send
on-demand tests to an agent(s) through, for example, a Remote
Procedure Call (RPC) call for immediate or on-demand execution.
[0057] In various embodiments, platform 100 is a scalable, elastic
architecture and may comprise several distributed components,
including components provided by one or more third parties.
Further, when platform 100 is referred to as performing a task,
such as storing data or processing data, it is to be understood
that a sub-component or multiple sub-components of platform 100
(whether individually or in cooperation with third party
components) may cooperate to perform that task.
[0058] In some embodiments, tests include various types of tests to
facilitate cross-layer visibility and troubleshooting of
application delivery associated with a distributed application(s),
as further described herein. Example network tests include data
path measurement tests, routing path measurement tests, and
end-to-end network metrics tests. Example DNS tests include per
name server testing and Domain Name System Security Extensions
(DNSSEC) bottom-up validation tests. Example HTTP tests include
testing of steps of a Uniform Resource Locator (URL) fetch. Example
page load tests include testing of a load of an entire web page
using a web browser (e.g., a typical web browser). Example
transaction tests include performing a multi-step scripted
transaction from a web browser (e.g., a typical web browser). These
and various other tests are discussed in greater detail below.
[0059] Cross-Layer Correlation
[0060] In the next sections, various techniques for allowing users
(e.g., Bob and Alice) to easily navigate between different layers
of data are described in accordance with some embodiments. For
example, platform 100 can provide users with a natural flow of
action for root cause identification of problems as further
described herein with respect to various embodiments.
[0061] Slicing Data Across Time--Rounds and Timeline
[0062] In some embodiments, tests aggregate data over certain time
intervals (e.g., small time intervals), referred to as rounds. In
some embodiments, a round includes one sample of data from each
agent (e.g., each agent can be placed in a different geographical
location) for a given test. In some embodiments, data over each
round is aggregated for a set of agents (e.g., all the agents) and,
for example, can be shown on a timeline. In some embodiments, a
timeline is implemented as a navigation widget that enables users
to see the aggregate behavior of all of the agents over time (e.g.,
or just an individual agent) and to click on a specific point in
time for a detailed drill down (e.g., to view specific cross-layer
network test related data at a specified time or time
window/interval). For example, a user can be provided with several
metrics to choose from, and a timeline can be associated with one
or more metrics.
[0063] Compound Tests
[0064] In some embodiments, some tests are composed of multiple
subtests that are run at the same time. For example, the subtests
can each be different and associated with different layers of
visibility. Performing different active tests at the same time for
different layers allows for switching between subtests while
keeping the same test and round (e.g., and eventually the same
agent location). For instance, a test can include an HTTP subtest
and a network subtest at the same time. Then the user can navigate
from HTTP metrics, to end-to-end network metrics, all the way down
to individual router hop metrics (e.g., by inspecting the data
path). For example, this can be useful to identify/classify
problems that are network induced versus problems that are back-end
related.
[0065] Correlating Different Layers
[0066] In some embodiments, when switching between views (e.g., or
layers), the context is maintained, such as the following: (1) the
test, (2) the round, and (3) the location (e.g., the agent ID or
the agent location). More formally, each layer L0 has an associated
vector of features, referred to as the context vector, C.sub.L0,
that includes all dimensions that uniquely identify a view in layer
L0. When moving between layer L0 and L1, a function is applied to
C.sub.L0 to provide C.sub.L1, such as for example, the following:
[0067] C.sub.L1=F.sub.L0->L1(C.sub.L0). We can think of F as a
matrix of functions, and the indexes are the current layer, and the
next layer the function is mapping to.
[0068] In some embodiments, a Graphical User Interface (GUI) is
provided that can allow a user (e.g., Bob or Alice) to navigate
between different layers, such as to jump to different application
delivery layers (e.g., path visualization, BGP metrics, BGP route
visualization, etc.) and web layers (e.g., basic HTTP, etc.). In
some embodiments, a set of layers are correlated, and reports
including visualizations that present results of test performed at
the different correlated layers can be accessed, such as using
hyperlinks in a GUI menu that can encode the context vectors as GET
parameters in the URLs. For example, users can jump to Path
Visualization, BGP Metrics, BGP Route Visualization, and Basic HTTP
views while maintaining the context as discussed above.
[0069] Table 1 (below) shows a transition table between different
layers, indicating from which layers (rows) is it possible to jump
to (columns) in accordance with some embodiments.
TABLE-US-00001 TABLE 1 FROM/ Net: End- Net: DNS: Server Basic Page
TO to-end Path Viz BGP Metrics HTTP Load Net: End- -- Yes Yes Yes*
Yes* -- to-end Net: Path Yes -- Yes Yes* Yes* -- Viz BGP Yes Yes -
Yes* Yes* -- DNS: Server Yes* Yes* Yes* -- -- -- Metrics Basic HTTP
Yes* Yes* Yes* -- -- Yes* Page Load -- -- -- -- Yes* -- Transitions
between layers, *indicates transition is only available for
compound tests.
[0070] Application Delivery Layers
[0071] The next sections describe the layers involved in
application delivery and how data is collected and analyzed at each
step in accordance with some embodiments.
[0072] Network Performance
[0073] Measuring Network Performance
[0074] In some embodiments, an approach for measuring response time
of network level metrics from a network server is provided using a
train of TCP SYN packets (synchronise packets in transmission
control protocol). For example, providing such a train of TCP SYN
packets appears to the network server to be like any normal
connection that the network server will normally respond to; and
while the network server may throttle such connections, even if
throttling occurs, that would still generally provide an accurate
reflection of a typical user connection response time as that
network server is throttling for other such user connections as
well. Other approaches for measuring response time of network level
metrics from a network server can include using ping/Internet
Control Message Protocol (ICMP) techniques, but some servers and/or
Internet Service Providers (ISPs) (e.g., gateways/firewalls) often
block such ICMP traffic.
[0075] In some embodiments, in order to measure end-to-end network
properties, active measurements from agents (e.g., distributed
agents) to destination servers are performed. For example, servers
can be identified by host name (or IP address) and TCP port number.
Periodically, a train of N TCP SYN packets is sent to each server
from each agent. A full TCP connection is not established, because
the client sends a TCP RST (TCP reset) right after receiving the
SYN ACK response from the server. These N points allow for
measuring, for example, the following: (1) packet loss (e.g., ratio
of sent packets that were ACKed), (2) average network delay (e.g.,
time between SYN and ACK), and (3) network jitter (e.g., average
delta between consecutive delays).
[0076] Visualizing Network Performance
[0077] FIG. 2 illustrates a network performance visualization in
accordance with some embodiments. In particular, FIG. 2 is a screen
shot 200 of a GUI presented via an interface (e.g., using platform
100) that provides a visualization of a network performance using
various techniques described herein. As shown at 202, users can
select the target (e.g., server) that they want to look at, such as
shown in this example, www.example.com, port 443 (SSL). As shown at
204, there is also an option to select the network metric, such as
shown, in this example; "loss" (or packet loss) is selected.
[0078] As shown at 206, a timeline is provided that shows the
time-series of the metric. In some embodiments, as described
further below, the timeline allows a user to select an instant in
or interval of time to drill down to (e.g., interactive timeline
that allows users to select a specified time or time interval for
drill down for further analysis, such as to look at a window of
time of high packet loss to try to determine root cause(s) of such
high packet loss).
[0079] As shown at 208, a world map depicts the locations of the
distributed agents as well as a visual indication of their status
according to the selected metric 204. In this example, the red dots
(e.g., shown as completely shaded circles or another graphical
indicator to differentiate these icons) are agents that are
experiencing heavy packet loss, and the green dots (e.g., shown as
completely partially shaded or hashed circles or another graphical
indicator to differentiate these icons) are agents without packet
loss.
[0080] A summary box 210 provides a concise report of the metrics.
A table 212 shows a more detailed breakdown of the metrics as
measured by each agent. For example, the table can be sorted by any
column to facilitate the analysis of the data. On the right end of
each row, a "Run Test" button 214 is provided that allows users to
run on-demand tests to the target (e.g., target server). For
example, this can be useful to verify if a problem reported by an
automated test is still occurring.
[0081] Data Paths
[0082] Measuring Data Paths
[0083] Traceroute is generally a computer network diagnostic tool
for displaying the route (path) and measuring transit delays of
packets across an Internet Protocol (IP) network. While a
traceroute diagnostic using ICMP packets can be used to measure
data paths, such an approach may not be effective on some networks
as many ISPs block ICMP packets. Accordingly, in some embodiments,
in order to collect data paths, Time To Live (TTL) limited TCP SYN
packets are sent to a specific destination server. Routers reply
with a TTL Time Exceeded message every time they receive a packet
with TTL=1 (e.g., set counter to 1, which is decremented to 0 in
IPv4 by routers each time they are processed by router; add one to
the TTL to keep extending the path an extra hop; repeat 3 times for
destination to map out path, as shown in the below pseudo code
sample). Thus, probes are sent with increasing TTL to collect the
source IP addresses of the ICMP packets to reconstruct the path
packets are taking. In some cases, special precaution can be taken
to avoid issues with load balancing. In the case of TCP, if the
same 5 tuple (e.g., source IP, destination IP, source port,
destination port, and protocol) is kept between probes, balancers
will send packets in that flow through the same interfaces.
[0084] As shown below in Algorithm 1, a sample pseudo code scheme
is provided in accordance with some embodiments for measuring data
paths using TCP SYN packets as discussed above.
TABLE-US-00002 MAX_ROUNDS=3; For vRound=1 to MAX_ROUNDS For vTTL=1
to 255 vReply = SendTCPPacket(vTTL, vDestination, vSourcePort,
vDestPort); If vReply!=null push(vHops [vRound], vReply); End If
End For //distinguishing nodes with losses from non-responding
nodes If vDestination not in vReply Mark last responding node as
lossy End If End For
Algorithm 1: Data Path Measurements.
[0085] Because the final hop is a TCP server, we should always
expect a TCP SYN ACK packet back from the server. If that does not
happen, then it either means that the server is not reachable at
layer 3 or that the application stopped at the server. Note that
some hops in the path might not send ICMP TTL Exceeded messages, so
the server SYN ACK is used as a reference. If there are
non-responsive hops after a certain point on the way to the server
and if the destination server replies with a TCP SYN ACK, then we
assume that those hops do not send ICMP TTL Exceeded (so there is
no packet loss). On the other hand, if a SYN ACK is not received
from the server, and we have non-responsive hops after hop X in the
path, then we assume X is one hop way from the point where packets
are being dropped--that is, the last known good hop in the data
path.
[0086] Below is an example illustrating hops in a data path and
whether or not a response is received on such hops on a path
between a start (e.g., a Source Node) and a Destination Node.
START.fwdarw.[IP-1--responsive].fwdarw.[no response].fwdarw. . . .
.fwdarw.[no response].fwdarw.[Destination Node]
[0087] In this example, if we do not receive a response from the
Destination Node, then in this case we would identify the hop at
IP-1, which did respond (the last known good hop) as the path
termination point.
[0088] Visualization of Data Paths
[0089] In some embodiments, various techniques for visualization of
data paths are provided. In some embodiments, data delivery is
represented in the network in a per-hop model in which each hop
represents an interface of a physical router. In some embodiments,
a graph of a data path visualization of a data path to a
destination site (e.g., a destination, such as port 443 (SSL) of
www.example.com, or some other destination) is provided, such as
shown in FIG. 3 as discussed below.
[0090] FIG. 3 illustrates a graph of a visualization of a data path
to a destination site in accordance with some embodiments. In
particular, FIG. 3 is a screen shot 300 of a GUI presented via an
interface (e.g., using platform 100) that provides a visualization
of a network performance using various techniques described herein.
As shown at 302, users can select the test name (e.g., test to be
performed by distributed agents). As shown at 204, there is also an
option to select the network metric, such as shown, in this
example; "loss" (or packet loss) is selected. As shown at 306,
various agent filters (e.g., configurations or settings) can be set
for the agents. Based on the test results received from the
distributed agents for the selected test(s) and agent filter input,
a graph 308 is presented to provide a data path visualization of a
data path to a destination site in accordance with some
embodiments.
[0091] In some embodiments, each path of the graph starts with an
agent node in the left side that is shown with icon indicators
(e.g., shown as colored circles, completely shaded circles, or
another graphical indicator to differentiate these icons) according
to each agent's results or status relative to the selected metric,
in this case, packet loss. For example, nodes with a red outline
can represent points in the network dropping packets. In
particular, such nodes can indicate the last responsive hops in the
path, such that the packet loss is indicated to most likely be
happening in the immediate hop. As another example of screen
visualization of a data path to a destination site, even if view of
visualization of data paths is selected to show zero hops using
agent filter 306, the data path visualization can still display
hops with packet loss to illustrate potential problematic hops
(e.g., regardless of the number of hops to be shown, all bad nodes
with losses can be presented in the visualization graph 308).
[0092] For example, the network delay between consecutive hops can
also be determined (e.g., approximated or estimated) using this
technique. As shown as agent filter 306, a control "Color links
with delay>.times.ms" is provided that can provide an icon
indicator (shown as colored circles such as red circles, completely
shaded circles, or another graphical indicator to differentiate
these icons) for links that have more than a selected number of
milliseconds (ms) (e.g., 100 ms as shown). The threshold of link
delay can be dynamically changed using agent filter 306, which then
automatically updates the data path visualization accordingly.
[0093] In some embodiments, a hop expansion control of agent filter
306 allows the topology to be expanded any number of hops from any
side (e.g., agent or server) of the topology. For example, this can
provide a useful agent filter control from the server side, because
that is typically a customer actionable zone. As shown in graph
308, the links with a number on top are aggregation links that are
expanded as the number of hops in the filter increases or just by
clicking in the number.
[0094] In some embodiments, another useful feature is the ability
to select nodes and links in the topology shown in graph 308. For
example, selected nodes and links can be inserted into a selection
box, and their position in the topology can be tracked over time by
clicking in different points in the timeline. In this manner,
routing changes can be tracked over time. In some embodiments,
double clicking in a specific node selects all nodes and links
belonging to routes going through that node, making it easier to
track entire routes over time.
[0095] In some embodiments, the metric selection, as shown at 204,
affects timeline 206 and the agent node status indicators (e.g.,
coloring of the agent icons or some other icon indicator) in the
graph 308. For example, the users can select from one of three
end-to-end metrics, such as packet loss, latency, and jitter. In
some embodiments, the values of these metrics are computed from a
train of n TCP SYN packets sent to the destination about the same
time the data path measurements take place to preserve time
correlation between the end-to-end effect and the per-hop effect.
As an example, if agent x is experiencing a very high end-to-end
latency to reach the destination, such very high end-to-end latency
to reach the destination can be displayed with a color red (e.g.,
coloring of the agent icons or some other icon indicator) in the
graph 308, and by looking at the breakdown per hop, where the
bottleneck is located can be determined.
[0096] For example, the visualization layout (e.g., using a version
of the Sugiyama's algorithm or another algorithm) can display a
graph hop by hop and minimize the link crossings between nodes,
making the graph more readable.
[0097] Case Studies
[0098] FIG. 4A illustrates a graph of a full connectivity
visualization of a data path to a destination site in accordance
with some embodiments. In particular, FIG. 4A is a screen shot 400
of a GUI presented via an interface (e.g., using platform 100) that
provides a visualization of a network performance using various
techniques described herein. As shown at 306, various agent filters
(e.g., configurations or settings) can be set for the agents. Based
on the test results received from the distributed agents for the
selected test(s) and agent filter input, a graph 408 is presented
to provide a data path visualization of a data path to a
destination site in accordance with some embodiments. In
particular, FIG. 4A shows a hop-by-hop data path view from the
agents (e.g., the agents as shown and discussed below with respect
to FIGS. 5A-5C) to a target site, such as example web site (e.g.,
www.example.com).
[0099] FIG. 4B illustrates a graph providing a simplified view of a
visualization of a data path to a destination site in accordance
with some embodiments. In particular, FIG. 4B is a screen shot 420
of a GUI presented via an interface (e.g., using platform 100) that
provides a visualization of a network performance using various
techniques described herein. As shown at 306, various agent filters
(e.g., configurations or settings) can be set for the agents. Based
on the test results received from the distributed agents for the
selected test(s) and agent filter input, a graph 428 is presented
to provide a data path visualization of a data path to a
destination site in accordance with some embodiments.
[0100] In particular, FIG. 4B is a simplified view of the FIG. 4A
view of the hop-by-hop data path from agents to a target site, such
as example web site (e.g., www.example.com), in which only show two
hops away from the web site are shown (e.g., each IP address
associated with the target site (e.g., example web site), which is
associated with these three IP addresses, and hiding/consolidating
all other hops except that all nodes with losses (e.g., nodes with
losses exceeding a threshold) are shown in this view. For example,
if a user selects "3 nodes with losses" as shown at component
(e.g., a sub-window) 426 of FIG. 4B, then FIG. 4C is shown in
accordance with some embodiments, as discussed below.
[0101] FIG. 4C illustrates a graph providing a simplified view of a
visualization of a data path to a destination site with problem
nodes in accordance with some embodiments. In particular, FIG. 4C
is a screen shot 440 of a GUI presented via an interface (e.g.,
using platform 100) that provides a visualization of a network
performance using various techniques described herein.
[0102] As shown, FIG. 14C illustrates a simplified view of a graph
448 with problem nodes selected in accordance with some
embodiments. In particular, FIG. 14C shows the result of a user
selecting "3 nodes with losses" as shown at component (e.g., a
sub-window) 426 of FIG. 4B. For example, this can allow the user to
drill down to see more information regarding selected problem
nodes. As shown in this example, each of the problem nodes is
associated with a common provider that would be listed in the
details window for each of these three problem nodes. If the user
selects link 446 shown as (8) hops between Amsterdam, Netherlands,
then the next hop/node is shown with losses as shown in FIG. 14D in
accordance with some embodiments, as discussed below.
[0103] FIG. 4D illustrates a graph providing a selective
exploration expanding a data path from selected source/agent site
in accordance with some embodiments. In particular, FIG. 4D is a
screen shot 460 of a GUI presented via an interface (e.g., using
platform 100) that provides a visualization of a network
performance using various techniques described herein.
[0104] As shown, FIG. 4D illustrates a graph 468 providing a
selective exploration expanding a data path from Amsterdam in
accordance with some embodiments. In particular, FIG. 4D
illustrates an expanded data path if a user selects the link 446
shown as (8) hops between Amsterdam, Netherlands and the next
hop/node shown with losses as shown in FIG. 4C, which thereby
allowing for an expanded view of the intermediate hops on the data
path from Amsterdam to that problem node as shown in FIG. 4C.
[0105] FIGS. 5A-5C illustrate examples of using cross-layer
visibility of application delivery to determine, at different
layers of a network, sources of one or more application delivery
problems or performance issues in accordance with some embodiments,
as discussed below.
[0106] FIG. 5A is screen shot illustrating an HTTP availability
drop during a particular time interval in accordance with some
embodiments. In particular, FIG. 5A is a screen shot 500 of a GUI
presented via an interface (e.g., using platform 100) that provides
a visualization of a network performance using various techniques
described herein.
[0107] In particular, FIG. 5A shows a drop in HTTP availability at
time interval 21:45-21:48 on April 28 as shown at 502. As shown at
504, a window is provided that shows a number of error by type,
including the following: DNS, Connect (2 errors as shown), SSL,
Send, Receive (1 error), and HTTP. At table 506, the details of
these errors are also shown in a table format, in which a first
column is a location of agent, date is a time of measurement,
server IP is the destination address (e.g., target for HTTP
availability testing), response code (e.g., HTTP response code,
which is 200 if there are no errors, or empty if no response code
received), and a number of redirects, error type (e.g., as
discussed above), and error details (e.g., providing a description
of the particular error, and a run test option can be provided to
further diagnose this problem and to check if the problem still
persists).
[0108] FIG. 5B is screen shot illustrating a network view that
shows packet loss from the agents shown in FIG. 5A in accordance
with some embodiments. In particular, FIG. 5B is a screen shot 520
of a GUI presented via an interface (e.g., using platform 100) that
provides a visualization of a network performance using various
techniques described herein.
[0109] Specifically, FIG. 5B illustrates a network view that shows
packet loss, from the same agents as shown in FIG. 5A, in
accordance with some embodiments. More specifically, FIG. 5B is a
network view of the same destination at the same time interval for
same destination, as shown in FIG. 5A. However, FIG. 5B provides a
different view, a network view as shown, which indicates that the
problem is a packet loss issue at the network layer. FIG. 5B
illustrates that the cause of the HTTP availability drop shown in
FIG. 5A is revealed by three agents that each have significant
packet loss, such as shown in the details of the packet losses of
these three agents in table 526, which are the same agents that
were shown in FIG. 5A with problems on the HTTP tests/measurements.
Accordingly, FIG. 5B reveals that the problem is not at the HTTP
layer, but rather is a problem at the network layer at this
particular time interval. Accordingly, this illustrates an example
of various techniques disclosed herein for cross-layer visibility
and troubleshooting of application delivery in accordance with some
embodiments.
[0110] FIG. 5C is screen shot illustrating a visualization of a
problem zone shown in FIG. 5B in accordance with some embodiments.
In particular, FIG. 5C is a screen shot 540 of a GUI presented via
an interface (e.g., using platform 100) that provides a
visualization of a network performance using various techniques
described herein.
[0111] Specifically, FIG. 5C illustrates a path visualization 546
highlighting a problem zone as shown at 544 (e.g., in which an
earlier time as shown at 542 indicates that they share a common
next hop, indicating the possible problem location or problem zone)
in accordance with some embodiments. More specifically, FIG. 5C
illustrates a hop-by-hop path visualization to determine the
problem zone causing the packet losses, as shown in FIG. 5B. For
example, the dashed circled nodes (e.g., some other icon indicator)
can indicate the last known good hops on these paths, which
indicates that the subsequent node(s), in this case it is a shared
common next hop, thereby revealing that this hop is the likely root
cause of the network packet loss on this data path. Accordingly,
this illustrates an example of various techniques disclosed herein
for cross-layer visibility and troubleshooting of application
delivery in accordance with some embodiments.
[0112] Routing Paths
[0113] Collecting Routing Information
[0114] The Border Gateway Protocol (BGP) is a standard protocol
used to exchange routing information between different Autonomous
Systems (AS) (e.g., which is the control plane between different
networks, such as between Verizon networks and Time Warner
networks). An AS is an independently managed domain (e.g., an
organization, such as Dell, which is associated with a particular
AS number, such as AS number 3614, which is a unique identifier for
Dell's network), typically having a one-to-one mapping to an
organization. BGP messages carry routing information for individual
prefixes (e.g., or group of IP addresses), and originate from the
traffic destination outwards (e.g., BGP message propagation follows
the opposite direction of traffic propagation).
[0115] In some embodiments, routing information is collected from
public BGP data repositories that archive routing information from
hundreds of routers across the Internet. For example, by looking at
the AS PATH attribute of each message sent from router R, the
routing path R was using at each time can be determined, and this
information can also be used to determine when a certain
destination IP address (e.g., or prefix) is/was not reachable from
R.
[0116] In some embodiments, three different metrics for BGP
visualization are provided: (1) reachability, (2) number of path
changes, and (3) number of BGP updates. From the point of view of a
router (or monitor), reachability refers to the fraction of time
that the router can reach a certain destination prefix. Path
changes refers to the number of times the attribute AS PATH changed
for a certain destination prefix. Updates refers to the plain count
of BGP update messages received during the time interval.
[0117] For example, the BGP route information can be collected from
RouteViews and/or RIPE-RIS (Routing Information Service), which
publish BGP route information. As described herein with respect to
various embodiments (see, e.g., FIGS. 8A-8B, which are described
below), the BGP route information can be correlated with various
other layers of network information to allow for visualization of
this data and analysis of this data in a manner that facilitates
cross-layer visibility and troubleshooting of application delivery
(e.g., to determine if a problem is caused by a BGP routing
information related problem, data path packet loss related problem,
HTTP related problem, DNS related problem, and/or some other
problem). As an example, by correlating cross-layer network data
and visualizing such cross-layer network data, users can make
better sense of such information and the correlation and
presentation of such information can more clearly reveal if a BGP
routing issue at a particular time interval may be a cause of
various network problems during that time interval (e.g., HTTP
measurement errors may actually be a result of this BGP routing
issue, such as when looking at data path loss errors in which there
are random drops at random locations, a user can then look to see
if problem is at BGP level in which there may be no announced route
to get to that network AS number, as that will cause the routers to
just drop such packets for which it has no routing
information).
[0118] Visualizing Routing Paths
[0119] FIG. 6 illustrates a graph of a visualization of routing
paths in accordance with some embodiments. In particular, FIG. 6 is
a screen shot 600 of a GUI presented via an interface (e.g., using
platform 10002) that provides a visualization of a network
performance using various techniques described herein.
[0120] As shown, FIG. 6 illustrates similar components as in
previous visualizations, including metric selector 204, timeline
206, graph 608, and summary components 210. In some embodiments, a
force-directed layout is used to represent the origin of the routes
in the center and the connected ASes (e.g., each node is an AS)
laid out in a radial way as shown in the graph 608. Routers are
shown as the leaf nodes of the graph 608 and are visually indicated
according to the selected metric (e.g., visual indicators, such as
colors and/or other visual indicators of such icons, can be used to
indicate a selected metric). For example, the yellow/orange nodes
(e.g., or some other visual indicator(s)) can correspond to routers
that had path changes, while the green nodes (e.g., or some other
visual indicator(s)) can correspond to routers that did not have
any path changes.
[0121] As also shown in FIG. 6, the timeline 206 shows increments
of 5 minutes (e.g., other time intervals/increments can be used or
configured to be used), which are referred to as rounds. Data is
visualized per round, and in each round, the paths are shown, using
dashed style, and the final paths of the rounds are shown in solid
style. The red dashed links (e.g., or some other visual
indicator(s)) in the graph 608 corresponds to paths that were used
at some point during the round, but were not in use at the end of
the round, which, for example, can be useful to detect path
exploration in BGP.
[0122] Path Simplification Algorithms
[0123] Currently if a node does not respond with a TTL expired, we
leave the node white in the visualization. If there's loss, we
might have a series of "white" nodes, typically followed by the
destination, such as shown in FIG. 7.
[0124] FIG. 7 illustrates a graph of a visualization of routing
paths including a series of nodes indicating data loss in
accordance with some embodiments. As shown in graph 708, some of
these white nodes 710 (e.g., or other visual indicators can be used
to indicate such nodes) never reply to various agent testing, such
as probes with ICMP Time Exceeded. These nodes can be referred to
as real stars. Other nodes usually reply but, due to loss,
occasionally the reply is not received. These nodes can be referred
to as spurious stars. Distinguishing real stars from spurious stars
is a challenging problem. For example, the challenge is that there
is no other indication of packet loss, as both real stars and
spurious stars look the same. However, a technique that can be used
to address this issue by using a set of paths from multiple sources
to a given destination is described below.
[0125] For example, for nodes in the middle of the path, losses
will create diamond shaped elements of equal length. For example,
two paths of equal length are shown below:
#1: 3-x-x-5-6 #2: 3-x-x-x-6 where the numbers represent different
nodes in path visualization and the "x" represents a node that did
not reply (e.g., referred to above as a white node). In this
example, node 5 replied with Time Exceeded on path #1, but not on
path #2. Diamond shapes can be detected in the graph and such can
be reduced (e.g., path simplification can be performed) as
follows:
A-path1-B
A-path2-B
[0126] for every diamond A-B in the graph, if B is not the final
destination, length(path1)=length(path2), and path1 and path2 are
mergeable, then the paths are merged. Mergeable paths are paths
that do not have conflicting hops. For example, 3-4-x-6 3-5-x-6 are
not mergeable, because 4!=5 but, as another example: 3-4-x-6
3-4-5-6 are mergeable as there are no conflicting hops. In this
case, the paths can be merged into path 3-4-5-6.
[0127] Inferring Packet Loss
[0128] In some embodiments, every time a diamond gets reduced, loss
on a certain link should be accounted for as discussed below. For
example, two paths are shown below:
#1: 3-4-5 #2: 3-x-5 in which the above two paths can be merged into
path 3-4-5. We know how many packets were routed through path #1
where 4 replied (e.g., N number of packets), and we also know how
many packets were routed through path #2 where there was no reply
after node 3 (e.g., L number of packets). Accordingly, the loss
percentage on link 3-4 can be computed as L/(N+L). More generally,
the loss of a node n can be determined by computing as the total
number of packets lost in next hops divided by the total number of
packets forwarded to next hops of n, as follows:
Loss(n)=Sum L(i)/Sum L(i)+Sum N(i)
For terminal nodes, such as nodes that do not have any next hops
but are not a destination, the loss is always 100%. In some cases,
it can also happen that a node is terminal for some routes, but not
for other routes. In such cases, the loss can be computed as if
there were a next hop on the terminal route where all N packets
sent to it were lost. For these next hops, N is equal to the
threshold of consecutive white nodes used in the terminal condition
to stop probing a path.
[0129] In some embodiments, the graph visualization of such nodes
marks loss using a visual indicator(s) (e.g., as a red circle on a
node, the thickness of which is proportional to the loss
percentage, or some other visual indicator(s) can be used). In some
embodiments, a slider to control the amount of loss visible is
provided (e.g., "Mark nodes with more than x % loss").
[0130] Diamonds Involving the Destination
[0131] In some embodiments, the destination is treated as a special
case. The destination will reply to any TCP SYN packet it receives
with a TCP SYNACK, which means that it will respond to any packet
with TTL equal to or greater than the true number of hops to the
destination. As a result, an algorithm that can be used for the
destination case is similar to the path merging algorithm, but in
this case paths are not required to be of the same length. For
example, for the following paths:
A-path1-C
A-path2-C
[0132] where C is the destination, but path1 and path2 are of
different lengths. Then, if path1 and path2 are mergeable, the path
can still be reduced, such as follows:
A-x-x-x-x-C
A-B-C
[0133] the result of the path merging is A-B-C.
[0134] Conditions to Reduce a Diamond
[0135] In some embodiments, the condition to filter cases with real
stars is as follows:
if we have
A-x-C
A-B-C
[0136] the diamond is reduced to A-B-C ONLY IF the following
rules/conditions are satisfied: (1) there is a dominant path inside
the diamond that has a threshold percentage of routes going through
it (e.g., at least 75%); in this case A-B-C needs to be a dominant
path for the merging to occur; AND (2) there is a minimum of four
routes entering the diamond (e.g., this can be required to avoid
false positives closer to the agents were routes are less dense);
as three routes per agent are collected (e.g., in which this last
condition also forces the diamond to have routes from at least two
different agents).
[0137] Special Cases
[0138] In some embodiments, there are some special cases where the
above discussed two rules/conditions (e.g., default rules) are
altered slightly. For example, given a mergeable diamond with
source node A and destination node B:
(1) If B is actually the destination and rule 1 does not apply
(e.g., none of the paths in the diamond is dominant), but rule 2
still applies (e.g., at least 4 routes are entering the diamond),
then paths can be simplified/reduced with only stars into a single
hop notated with "?" indicating the uncertainty in the number of
hops. For example:
A-x-B
A-x-x-B
A-x-C-B
[0139] can be reduced to:
A-?-B
A-x-C-B
[0140] The "?" link is used in this example, because there is not
enough information is available to determine the true number of
hops (e.g., or true number of distinct paths) between A and B with
sufficient certainty. (2) If there are more than two distinct paths
inside a diamond, do not perform a merge unless there is a dominant
path that originates from the same agent as the mergeable path. For
example: agent1-A-B-x-F agent2-A-B-C-F (dominant) agent3-A-B-D-F
could not be reduced, even though A-B-C-F is dominant; neither
A-B-C-F nor A-B-D-F appears in a path originating from agent1,
which of the two (if any) agent1 is connected to cannot be
determined. However, if the following path: agent1-A-B-C-F were
also present in the above example, the path with the star could be
merged into A-B-C-F according to the default rules as discussed
above.
[0141] Information Over Time Versus Over Space
[0142] As discussed above, information collected from multiple
sources can be used to infer the reality of a path from a single
source. Similar techniques can be used across the time dimension.
For example, by looking at the data on the preceding and subsequent
intervals, the reality of the path can be inferred at the current
interval. Although the space-based approach discussed above can be
more convenient in some cases, such as if the necessary information
is readily available at the time of visualization, in other cases,
a time-based approach can also be used, such as for cases where not
enough data in one time slice is available to get an accurate
representation of a path.
[0143] Case Studies
[0144] FIG. 8A and FIG. 8B illustrate views for showing Border
Gateway Protocol (BGP) and packet loss in accordance with some
embodiments. FIGS. 8A and 8B are screen shots 800 and 810,
respectively, of a GUI presented via an interface (e.g., using
platform 100) that provides a visualization of a network
performance using various techniques described herein.
[0145] In particular, FIG. 8A illustrates a data path view in which
a data path from source site (e.g., Chicago node) to the lowest
listed destination does not have an IP address and is shown as
having packet loss (e.g., using a visual indicator for that
destination node, such as a color coded indicator or other
graphical user interface related visual indicator(s)). FIG. 5B
illustrates a BGP view, in which a window 812 shows an IP address
destination for which there are no routes to reach that
destination, which corresponds to the path from Chicago for which
there is packet loss on the path as shown in the data path view of
FIG. 8A. Accordingly, FIGS. 5A and 5B illustrate the user of
cross-layer visibility and troubleshooting into application
delivery in which a root cause of a packet loss seen on the Chicago
node in FIG. 8A can be illustrated by navigating to a BGP view for
that same time interval to determine that the Chicago node was
unreachable due to BGP route information related problems. In
particular, FIG. 8B provides a visual representation that explains
that the packet loss seen on the Chicago node in FIG. 8A is a
result of the unavailability to a certain BGP prefix.
[0146] DNS
[0147] Collecting and Visualizing DNS Information
[0148] FIG. 9 illustrates a visualization of DNS data in accordance
with some embodiments. In particular, FIG. 9 is a screen shot 900
of a GUI presented via an interface (e.g., using platform 100) that
provides a visualization of a network performance using various
techniques described herein.
[0149] In some embodiments, a DNS test (e.g., that can be performed
by one or more distributed agents and controlled by agent
controllers, such as agent controllers 114) includes several
authoritative name servers that serve a specific domain name. For
example, DNS queries from agents can be sent to the specified DNS
servers targeting the specified domain name (e.g., one or more
domain names that an entity wants to test for, such as
www.domain-name.com). For example, the following cases can be
marked as errors:
[0150] The absence of a reply--the name server does not reply
within a certain timeout to the query.
[0151] An empty reply--the name server replied with an NXDOMAIN, or
an empty resource record.
[0152] This technique can be used for identifying and resolving
application level problems (e.g., application layer response time),
as DNS tests can be correlated to identify which DNS name server(s)
may be the root of the problem. For example, if a problem is
identified with a particular DNS name server, then a user can drill
down to the network layer view to determine if it is a network
problem to access that DNS name server from a particular location
(e.g., a location at which one or more agents are present for such
testing).
[0153] As shown in FIG. 9, various visualization elements are
provided that are similar to previously discussed views. For
example, metric selector 204 can be used to choose between
Availability and Resolution Time. Timeline 206 can be used to
select an instant in time to drill down to. World map 208 shows a
geographical representation of the problem. Summary window 210
shows a small summary of the metrics/number of errors. Table 212
shows the breakdown of the DNS tests per location.
[0154] For example, when no server is selected, the view can show
aggregated metrics (e.g., average for the case of Availability and
Minimum value for the case of Resolution Time). A minimum can be
selected as this is the most likely the value a DNS resolver would
get if it had to query multiple authoritative servers for a
domain.
[0155] DNSSEC
[0156] In some embodiments, a similar visualization for DNSSEC is
provided, such as for the DNS extension that adds authentication
and data integrity to DNS. In order to test DNSSEC delegation
chains, a test from each agent is provided in which the test can
start at the bottom of the chain and verify signatures all the way
to the top of the chain. An error is triggered if there is any step
in the resolution chain that fails, either because a resource
record is missing or because a signature does not match.
[0157] Case Studies
[0158] FIG. 10 illustrates Anycast troubleshooting, to identify any
relevant DNS problems, in accordance with some embodiments. In
particular, FIG. 10 is a screen shot 1000 of a GUI presented via an
interface (e.g., using platform 100) that provides a visualization
of a network performance using various techniques described
herein.
[0159] In particular, FIG. 10 illustrates a DNS view, illustrating
DNS resolutions to the DNS servers server1.com (e.g., showing four
errors in this example) and server2.com (e.g., showing 7 errors in
this example) in table 1010, which are geographically represented
by location in red on the world map view 1008, and the time
interval is 6:00-6:00:47 on Mar. 8, 2012.
[0160] FIG. 11 illustrates a path visualization that shows all
problem agents routing to an Ashburn, Va. instance in accordance
with some embodiments. In particular, FIG. 11 is a screen shot 1100
of a GUI presented via an interface (e.g., using platform 100) that
provides a visualization of a network performance using various
techniques described herein.
[0161] In particular, FIG. 11 illustrates a path visualization to
server2.com in path visualization graph 1102. In this example, DNS
Anycast is used, which can announce its IP from several different
locations (e.g., five different locations). This shows that a
commonality among the agents that have DNS errors is that all of
these agents are performing their DNS lookup with this particular
Anycast server in Ashburn Va., which can be inferred after looking
at the location of the hops immediately preceding the server in
path visualization graph 1102 as shown.
[0162] HTTP
[0163] FIG. 12 illustrates a visualization of HTTP data in
accordance with some embodiments. In particular, FIG. 12 is a
screen shot 1200 of a GUI presented via an interface (e.g., using
platform 100) that provides a visualization of a network
performance using various techniques described herein. The various
other components as shown in FIG. 12 are similar to components
shown in previously discussed views, including timeline 206, world
map 208, summary 210, and table 212 components of the screen shot
1200.
[0164] In some embodiments, for HTTP tests, the URL to be tested is
fetched from each agent location, the time to complete each step is
measured, and whether there was an error at any particular step is
determined and recorded. For example, an HTTP fetch can include the
following steps: DNS resolution, TCP connection establishment, SSL
certificate exchange, HTTP GET request, and/or HTTP response. In
some cases, the HTTP tests can capture the following metrics:
Availability: an indication of errors, the aggregate availability
is the fraction of agents that complete the test without any error;
Response Time: or Time to First Byte (TTFB) is the time it takes
for the client to start receiving HTTP data from the server; this
metric is useful to understand if delays are coming from network or
from backend; typically slow response times are caused by slow
networks; and/or Fetch Time: this is the time it takes to actually
receive all the HTTP data.
[0165] Web Page
[0166] FIG. 13 illustrates a visualization of a web page load
performance in accordance with some embodiments. In particular,
FIG. 13 is a screen shot 1300 of a GUI presented via an interface
(e.g., using platform 100) that provides a visualization of a
network performance using various techniques described herein.
[0167] In some embodiments, in order to measure the performance of
an entire web page, real web browsers can be used by agents (e.g.,
web testing agents that can be distributed and controlled by agent
controllers, such as agent controllers 114) to load the web page(s)
from different locations and extract the performance for each of
the components of the web page(s). For example, the components can
be grouped per domain and per provider to rank Response Time and
Throughput, and to better identify bottlenecks. Example metrics can
include the following: DOM time: the time it takes for the document
object model (DOM) to build; after the DOM is created, the user can
start interacting with the page; and/or Page load time: the time it
takes for the entire page to load, including style sheets and
images.
[0168] In some embodiments, when a user clicks in a location in
world map 1308, the waterfall at 1312 is updated. The waterfall
view 1312 provides a breakdown of the individual performance of
each component of the page, for example, including time
measurements for each of the following: Blocked: the time the
browser is waiting for a TCP connection to be available to process
the request; DNS: the time it takes to do a DNS resolution for the
name of the resource; Connect: the time it takes to establish a TCP
connection to the server; SSL/TSL: the time it takes to do the
SSL/TSL handshake; Wait: the time between sending the HTTP request
and getting the first byte of the reply back from the server; and
Receive: the time it takes from the first byte of the answer until
the last byte is received.
[0169] Case Studies
[0170] FIG. 14 illustrates a bottleneck analysis by provider for
page loads in accordance with some embodiments. In particular, FIG.
14 is a screen shot 1400 of a GUI presented via an interface (e.g.,
using platform 100) that provides a visualization of a network
performance using various techniques described herein.
[0171] In particular, FIG. 14 illustrates a page load view that
indicates detailed views of page load by provider (e.g., each
object can provide from different providers, such as Dell, Adobe,
and Akamai for www.example.com), for a particular period of time
and an average response time and throughput, using browser
measurements (e.g., using web testing agents, as similarly
discussed above).
[0172] Web Transactions
[0173] FIG. 15 illustrates a visualization of transactions (e.g.,
web transactions) in accordance with some embodiments. In
particular, FIG. 15 is a screen shot 1500 of a GUI presented via an
interface (e.g., using platform 100) that provides a visualization
of a network performance using various techniques described
herein.
[0174] In some embodiments, in addition to tests on individual web
pages, the platform for cross-layer visibility for application
delivery also supports transaction tests, such as for web-based
transactions. For example, a transaction can refer to a series of
scripted steps that are automatically executed in a browser. As an
example, a transaction can include going to a homepage of a web
site and adding an item into a shopping cart. As another example, a
transaction can include going to a home page of a web site, logging
into the web site with user credentials, and downloading a first
home login page associated with that user's credentials.
[0175] Accordingly, in some embodiments, transactions are used to
refer to extensions of single page loads, and, as shown in FIG. 15,
the visualization of transactions can similarly be provided as for
single page loads (e.g., simple, single web page load
transactions). In some embodiments, on each transaction, users are
allowed to define an initial step and a final step. For example,
these steps can be used to define the boundaries of interest in the
transaction and to define the boundaries for computing the
transaction time for the measured transaction.
[0176] As shown in FIG. 15, in waterfall 1512, all the steps of the
transaction are presented as vertical colored lines or using
another visual indicator(s) (e.g., the x-offset corresponds to
elapsed time since step 0). The individual components that are
outside of the interval defined between the initial and final step
can be faded out to visually emphasize the relevant steps. In the
summary component 1510, a breakdown of the time spent on each step
and the time each page loaded during a transaction took to load is
provided. For example, metrics for transactions can include:
Completion: the fraction of transaction steps that were completed
during the transaction; and/or Transaction time: the time between
the initial step and the final step of the transaction.
[0177] Case Studies
[0178] FIG. 16 shows a summary of transaction steps and identifying
bottlenecks in accordance with some embodiments. For example, FIG.
16 illustrates web transaction screen shot 1600 of a summary
including a time per step and time per page for an example web
transaction test results. In particular, as shown, step 6 for page
load #1 was measured as using a relatively significant/longer
amount of time. Thus, this indicates that step 6 for page load #1
may be the cause of the bottleneck for these web transactions.
[0179] Diagnosing Performance Issues Using Browser Extension
[0180] Performance problems can be challenging and time consuming
to diagnose when they cannot be reproduced. When users experience
performance problems and report to IT or to the service provider,
the providers often use data from a nearby city from an external
monitoring system to cross check. However, this is problematic,
because it does not collect data from the end user reporting the
problem.
[0181] Accordingly, in some embodiments, a new browser extension is
provided (e.g., for a web browser), which is a lightweight
extension that can be installed by any users that desire to report
their experience to a service provider or Information Technology
(IT) department. For example, when the user clicks on a small
"capture performance" icon on the browser of the user's client
device (e.g., laptop, computer, smart phone, or other computing
device that can execute a web browser), the add-on starts recording
errors and timings for each object received by the browser, and
also perform active measurements such as traceroutes at the same
instant of time. The extension then exports the data collected to a
collector of a service provider or IT department (e.g., cloud based
network performance service provider or IT), which channels the
data to the relevant customer of the service provider/IT based on
the domain that was being tested. For example, data on a test to
store.apple.com can be made available on the Apple users (e.g., in
the service provider's data for the customer) if they exist.
[0182] FIG. 17 illustrates a flow diagram for cross-layer
troubleshooting of application delivery in accordance with some
embodiments. In some embodiments, process 1700 is performed using
platform 100 as shown in FIG. 1. At 1702, collecting test results
from a plurality of distributed agents for a plurality of
application delivery layers is performed. At 1704, generating a
graphical visualization of an application delivery state for the
plurality of application delivery layers based on the test results
is performed. At 1706, the graphic visualization is output.
[0183] FIG. 18 illustrates another flow diagram for cross-layer
troubleshooting of application delivery in accordance with some
embodiments. In some embodiments, process 1800 is performed using
platform 100 as shown in FIG. 1. At 1802, sending tests to
distributed agents (e.g., application delivery tests) to perform
for different application delivery layers (e.g., to perform various
active measurements across different application delivery layers)
is performed. At 1804, collecting test results from the distributed
agents for the different application delivery layers is performed.
At 1806, correlating the collected test results for the different
application delivery layers is performed. At 1808, generating a
graphical visualization based on the correlated test results for
the different application delivery layers (e.g., generating a
graphical visualization of application delivery results for the
different layers and enabling layer correlation) is performed.
[0184] Although the foregoing embodiments have been described in
some detail for purposes of clarity of understanding, the invention
is not limited to the details provided. There are many alternative
ways of implementing the invention. The disclosed embodiments are
illustrative and not restrictive.
* * * * *
References