U.S. patent application number 16/997610 was filed with the patent office on 2020-12-03 for nvm express over fabrics.
The applicant listed for this patent is Toshiba Memory Corporation. Invention is credited to Yaron Klein.
Application Number | 20200382595 16/997610 |
Document ID | / |
Family ID | 1000005030726 |
Filed Date | 2020-12-03 |
United States Patent
Application |
20200382595 |
Kind Code |
A1 |
Klein; Yaron |
December 3, 2020 |
NVM EXPRESS OVER FABRICS
Abstract
A network storage appliance comprises solid state disks, a
network interface adapter communicatively coupled to the solid
state disks and a host client, a non-volatile semiconductor memory
device communicatively coupled to the solid state disks and the
network interface adapter, and a CPU communicatively coupled to the
non-volatile semiconductor memory device and the network interface
adapter. The non-volatile semiconductor memory device can receive
data from the host client via the network interface adapter, store
the data temporarily, and transfer the data to one of the solid
state disks. The CPU can receive a write request from the host
client via the network interface adapter, determine whether or not
the non-volatile semiconductor memory device is available to store
the data, initiate the network interface adapter to perform the
transfer of the data from the host client to the non-volatile
semiconductor memory device if the non-volatile semiconductor
memory device is available to store the data, and initiate the
network interface adapter to broadcast a query to one or more
external storages if the non-volatile semiconductor memory device
is not available to store the data.
Inventors: |
Klein; Yaron; (Raanana,
IL) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Toshiba Memory Corporation |
Tokyo |
|
JP |
|
|
Family ID: |
1000005030726 |
Appl. No.: |
16/997610 |
Filed: |
August 19, 2020 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15668314 |
Aug 3, 2017 |
10785301 |
|
|
16997610 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06F 15/17331 20130101;
H04L 49/356 20130101; H04L 67/1097 20130101 |
International
Class: |
H04L 29/08 20060101
H04L029/08; G06F 15/173 20060101 G06F015/173; H04L 12/931 20060101
H04L012/931 |
Claims
1. A network storage appliance comprising: a plurality of solid
state disks; a network interface adapter communicatively coupled to
the plurality of solid state disks and a host client, wherein the
network interface adapter is configured to: retrieve data from the
host client, and transfer the data from the host client to the
plurality of solid state disks; a non-volatile semiconductor memory
device communicatively coupled to the plurality of solid state
disks and the network interface adapter, wherein the non-volatile
semiconductor memory device is configured to: receive data from the
host client via the network interface adapter, store the data
temporarily, and transfer the data to one of the plurality of solid
state disks; and a CPU communicatively coupled to the non-volatile
semiconductor memory device and the network interface adapter,
wherein the CPU is configured to: receive a write request from the
host client via the network interface adapter, determine whether or
not the non-volatile semiconductor memory device is available to
store the data, in response to determining that the non-volatile
semiconductor device is available to store the data, initiate the
network interface adapter to perform the transfer of the data from
the host client to the non-volatile semiconductor memory device via
the network interface adapter, and in response to determining that
the non-volatile semiconductor memory device is not available to
store the data, initiate the network interface adapter to broadcast
a query to one or more external storages via the network interface
adapter.
Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application is a continuation of application Ser. No.
15/668,314, filed Aug. 3, 2017, the subject matter of which is
hereby incorporated herein by reference in its entirety.
FIELD OF THE INVENTION
[0002] The present disclosure relates to non-volatile memory
express storage appliances and methods for implementing
non-volatile random-access memory in high performance storage
networks.
BACKGROUND
[0003] Low latency for write commands is an important feature in
High Performance Computing (HPC) and especially in storage
networks. An example of its importance is the growing area of
in-memory databases (IMDBs). The increase in computing resources,
such as CPUs and memory, enables loading the entire database into
memory and process it by the CPU. This enables the fast in-memory
processing of large data sets. However, incoming transactions must
be persistent, and thus must be written to persistent media. As
processing stalls until acknowledgement of such transactions, it
becomes a bottleneck to the entire process.
[0004] The introduction of NVMe.TM. ("nonvolatile memory express")
devices with fast PCIe (Peripheral Component Interface Express)
interfaces and efficient storage stacks have reduced the access
time to the persistent media. In addition, remote direct memory
access (RDMA) network interface controllers (NIC) have evolved to
provide very fast (single digit .mu.sec) data transfer. The
introduction of the NVMe.TM. over Fabric standard enables an
environment, where host clients can access NVMe.TM. disks via low
latency network fabrics. Each host client and storage appliance has
an RDMA NIC that enables low latency transfer between each element
in the fabric. However, there remains a long felt need to
efficiently manage and implement NVMe.TM. devices in high
performance storage networks.
SUMMARY OF INVENTION
[0005] The present disclosure relates to a network storage
appliance comprising solid state disks, a network interface adapter
communicatively coupled to the solid state disks and a host client,
a non-volatile semiconductor memory device communicatively coupled
to the solid state disks and the network interface adapter, and a
CPU communicatively coupled to the non-volatile semiconductor
memory device and the network interface adapter. The network
interface adapter can retrieve data from the host client and
transfer the data from the host client to the solid state disks.
The non-volatile semiconductor memory device can receive data from
the host client via the network interface adapter, store the data
temporarily, and transfer the data to one of the solid state disks.
The CPU can receive a write request from the host client via the
network interface adapter and determine whether or not the
non-volatile semiconductor memory device is available to store the
data. In the case of determining that the non-volatile
semiconductor memory device is available to store the data, the CPU
can initiate the network interface adapter to perform the transfer
of the data from the host client to the non-volatile semiconductor
memory device via the network interface adapter. In the case of
determining that the non-volatile semiconductor memory device is
not available to store the data, the CPU can initiate the network
interface adapter to broadcast a query to one or more external
storages via the network interface adapter. Further, the CPU can
send an acknowledgement status to the host client via the network
interface adapter that indicates that the data has been stored in
the non-volatile semiconductor memory device.
[0006] According to one implementation, the non-volatile
semiconductor memory device may comprise one of resistive
random-access memory, magneto resistive random-access memory, nano
random-access memory, and battery-backed dynamic random-access
memory.
[0007] In some implementations, the network storage appliance may
include a volatile memory device communicatively coupled to the
solid state disks and the network interface adapter. The volatile
memory device may receive data from the one or more external
storages via the network interface adapter in the case of
determining that the non-volatile semiconductor memory device is
not available to store the data. The one or more external storages
may correspond to one or more second network storage appliance. The
volatile memory device may comprise dynamic random-access memory
(DRAM).
[0008] In other implementations, the volatile semiconductor memory
device may store the data temporarily and transfer the data to one
of the solid state disks.
[0009] In certain implementations, the CPU may receive a response
to the query from a first external storage of the one or more
external storages via the network interface adapter. The response
to the query may indicate that the first external storage is
available to store the data. The CPU may send a second write
request to the first external storage via the network interface
adapter. The one or more external storages may be one or more of
many network storage appliances in a storage network.
[0010] According to one implementation, the second write request
may initiate the first external storage to begin the transfer of
the data from the host client to the first external storage.
[0011] In some implementations, the CPU may receive a signal from
the one or more external storages via the network interface
adapter. The signal may indicate that the data has been stored in
the one or more external storages.
[0012] In other implementations, the CPU may send a second
acknowledgment status to the host client via the network interface
adapter. The second acknowledgment status may indicate that the
data has been stored in the one or more external storages.
[0013] According to some implementations, the network interface
adapter functions according to a remote direct memory access (RDMA)
protocol. The network interface adapter may be connected to a
network communication fabric. According to one implementation, the
network interface adapter accesses the non-volatile semiconductor
memory device independently of the CPU.
[0014] In some implementations, the non-volatile semiconductor
memory device may comprise a reserved storage area for the host
client. The non-volatile semiconductor memory device may further
comprise a local storage area and a remote storage area.
[0015] In certain implementations, the non-volatile semiconductor
memory device may determine whether the reserved storage area is
available to store the data. In response to determining that the
reserved storage area is available to store the data, the
non-volatile semiconductor memory device may allocate space from
the reserved storage area in which to store data.
[0016] In other implementations, the non-volatile semiconductor
memory device may, in response to determining that the reserved
storage area is not available to store the data, determine whether
the local storage area is available to store the data. In response
to determining that the local storage area is available to store
the data, the non-volatile semiconductor memory device may allocate
space from the local storage area in which to store the data.
[0017] According to one implementation, the non-volatile
semiconductor memory device may determine whether the remote
storage area is available to store the data. In response to
determining that the remote storage area is available to store the
data, the non-volatile semiconductor memory device may allocate
space from the remote storage area in which to store the data.
[0018] In some implementations, the data is transferred from the
host client to the non-volatile semiconductor memory device of the
target network storage appliance within a time period greater than
0.1 .mu.sec and less than 1 .mu.sec.
[0019] A second aspect of the present disclosure relates to a
method of transferring data from a host client to a network storage
appliance. The method comprises sending, from a host client, a
first write request to a target network storage appliance via a
network interface adapter. The method also comprises sending, from
the target network storage appliance, a second write request to a
second network storage appliance via the network interface adapter.
The method further comprises transferring, from the host client,
data to a non-volatile semiconductor memory device of the second
network storage appliance via the network interface adapter that is
associated with the first write request. Further, the method
comprises sending, from the second network storage appliance via
the network interface adapter, a signal to the target network
storage appliance that indicates that the data has been stored in
the non-volatile semiconductor memory device of the second network
storage appliance. The method also comprises sending, from the
target network storage appliance via the network interface adapter,
an acknowledgement status to the host client that indicates that
the data has been stored. Further, the method comprises
transferring, from the non-volatile semiconductor memory device of
the second network storage appliance via the network interface
adapter, the data to a volatile semiconductor memory device of the
target network storage appliance. The method further comprises
transferring, from the volatile semiconductor memory device of the
target network storage appliance, the data to a solid state disk of
the target network storage appliance.
[0020] According to some implementations, the non-volatile
semiconductor memory device may comprise one of resistive
random-access memory, magneto resistive random-access memory, nano
random-access memory, and battery-backed dynamic random-access
memory.
[0021] In some implementations, The volatile memory device may
comprise dynamic random-access memory (DRAM).
[0022] In other implementations, the network interface adapter
functions according to a remote direct memory access (RDMA)
protocol. The network interface adapter may be connected to a
network communication fabric. According to one implementation, the
network interface adapter accesses the non-volatile semiconductor
memory device independently of the CPU.
[0023] According to some implementations, the non-volatile
semiconductor memory device may comprise a reserved storage area
for the host client. The non-volatile semiconductor memory device
may further comprise a local storage area and a remote storage
area.
[0024] In certain implementations, the method further comprises
determining whether the reserved storage area is available to store
the data and, in response to determining that the reserved area is
available to store the data, allocating space from the reserved
storage area in which to store the data.
[0025] In other implementations, the method comprises, in response
to determining that the reserved storage area is not available to
store the data, determining whether the local storage area is
available to store the data and, in response to determining that
the local storage area is available to store the data, allocating
space from the local storage area in which to store the data.
[0026] According to some implementations, the method further
comprises determining whether the remote storage area is available
to store the data and, in response to determining that the remote
storage area is available to store the data, allocating space from
the remote storage area in which to store the data.
[0027] In certain implementations, the method further comprises
transferring the data from the host client to the non-volatile
semiconductor memory device of the target network storage appliance
within a time period greater than 0.1 .mu.sec and less than 1
.mu.sec.
[0028] In certain implementations, the second network storage
appliance may be one of many network storage appliances in a
storage network.
BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The foregoing and other objects and advantages will be
apparent upon consideration of the following detailed description,
taken in conjunction with the accompanying drawings, in which like
reference characters refer to like parts throughout, and in
which:
[0030] FIG. 1 shows a schematic representation of an NVMe.TM.
storage appliance with non-volatile memory, configured according to
one or more embodiments of the present disclosure;
[0031] FIG. 2 is a flow diagram of method steps for a write command
in NVMe.TM. over Fabric with non-volatile memory, according to an
embodiment of the present disclosure;
[0032] FIG. 3 shows a schematic representation of an NVMe.TM.
storage appliance and a second NVMe.TM. storage appliance,
configured according to one or more embodiments of the present
disclosure;
[0033] FIG. 4 is a flow diagram of method steps for a write command
in NVMe.TM. over Fabric with a second NVMe.TM. storage appliance,
according to an embodiment of the present disclosure;
[0034] FIG. 5 shows a schematic representation of an NVMe.TM.
storage appliance with a non-volatile memory tier and NVMe.TM.
tier, configured according to one or more embodiments of the
present disclosure;
[0035] FIG. 6 shows a schematic representation of a non-volatile
semiconductor memory device in an NVMe.TM. storage appliance,
configured according to one or more embodiments of the present
disclosure; and
[0036] FIG. 7 is a flow diagram of method steps for space
allocation during a write command in NVMe.TM. over Fabric,
according to an embodiment of the present disclosure;
DETAILED DESCRIPTION
[0037] FIG. 1 shows a schematic representation of an NVMe.TM.
system 100 comprising a host client 102 and storage appliance 114
communicating over Fabric 112. The host client 102 is a computing
system that comprises an application 104, a CPU ("central
processing unit") 106, and DRAM ("dynamic random access memory")
108. The storage appliance 114 is a computing system that comprises
a CPU 126, non-volatile memory 308, and multiple NVMe.TM. disks
116. The host client 102 and storage appliance 114 interface using
RDMA network adapters 110 and 130 which are a type of NIC. DRAM 108
maintains data only for as long as power is applied. Fabric 112 is
a network topology in which nodes pass data to each other through
interconnecting switches and network connections such as electrical
cables, optical fibers, or wireless links.
[0038] Non-volatile memory 308 is random access memory that retains
its information when power is turned off. Non-volatile memory 308
may be battery-backed DRAM or one of many emerging technologies.
Battery-backed DRAM is similar to DRAM 108 but supported by a
battery such that the DRAM does not lose power and retains the
data. An example of an emerging non-volatile memory technology is
resistive random-access memory (RRAM or ReRAM) that works by
changing the resistance across a dielectric solid-material. Another
example of an emerging non-volatile memory technology is magneto
resistive random-access memory (MRAM) which uses magnetic elements.
In addition, there are future technologies such as Nano-RAM which
is based on carbon nanotube technology. Non-volatile memory 308 may
reside in NVMe.TM. disks 116.
[0039] When the host client 102 wants to write data to a particular
NVMe.TM. disk 116 in the storage appliance 114, CPU 106 of the host
client encapsulates the write command and sends the write command
to CPU 126 of the storage appliance 114 using the RDMA interface
110. Application 104 sends the data to DRAM 108 of the host client
so that the data is ready for transmission to the storage appliance
114 once the storage appliance 114 is ready to receive it.
[0040] Once the storage appliance 114 receives the write command
using the RDMA interface 130, the CPU 126 of the storage appliance
114 parses the encapsulated write command, de-capsulates it, and
extracts the data information. CPU 126 of the storage appliance 114
then initiates a transaction of the data using the RDMA interface
130 from DRAM 108 of the host client 102 to the non-volatile memory
308 of the storage appliance 114. Following the RDMA 130
transaction, the data will reside in the non-volatile memory 308 of
the storage appliance 114.
[0041] Once the data resides in the non-volatile memory 308 of the
storage appliance 114, non-volatile memory 308 sends CPU 126 of the
storage appliance 114 a status that indicates that the data has
been stored in non-volatile memory 308. The CPU 126 of the storage
appliance 114 then encapsulates the status and sends it to the CPU
106 of the host client 102 using the RDMA interface 130. The status
serves as an acknowledgement that the data has been stored in the
specified NVMe.TM. disk 116.
[0042] Since the storage appliance 114 has sent the acknowledgement
to the host client 102, in a background process, CPU 126 of the
storage appliance 114 initiates the transfer of the data from the
non-volatile memory 308 of the storage appliance 114 to the
NVMe.TM. disk 116 specified by the write command.
[0043] A process 200 of a write command using NVMe.TM. system 100
is illustrated in FIG. 2. The process 200 begins by sending an
encapsulated write command in step 402. For example, in NVMe.TM.
system 100, when the host client 102 wants to write data to a
particular NVMe.TM. disk 116 in the storage appliance 114, CPU 106
of the host client encapsulates the write command and sends the
write command to CPU 126 of the storage appliance 114 using the
RDMA interface 110. Application 104 sends the data to DRAM 108 of
the host client so that the data is ready for transmission to the
storage appliance 114 once the storage appliance 114 is ready to
receive it.
[0044] Process 200 continues by transferring data from the host
client 102 to the storage appliance 114 in step 404. For example,
in NVMe.TM. system 100, once the storage appliance 114 receives the
write command using the RDMA interface 130, the CPU 126 of the
storage appliance 114 parses the encapsulated write command,
de-capsulates it, and extracts the data information. CPU 126 of the
storage appliance then initiates a transaction of the data using
the RDMA interface 130 from DRAM 108 of the host client 102 to the
non-volatile memory 308 of the storage appliance 114. Following the
RDMA 130 transaction, the data will reside in the non-volatile
memory 308 of the storage appliance 114.
[0045] Process 200 continues by sending an acknowledgement to the
host client 102 from the CPU 126 of storage appliance 114 in step
406. For example, in NVMe.TM. system 100, once the data resides in
the non-volatile memory 308 of the storage appliance 114,
non-volatile memory 308 sends CPU 126 of the storage appliance 114
a status that indicates that the data has been stored in
non-volatile memory 308. The CPU 126 of the storage appliance 114
then encapsulates the status and sends it to the CPU 106 of the
host client 102 using the RDMA interface 130. The status serves as
an acknowledgement that the data has been stored in the specified
NVMe.TM. disk 116.
[0046] Process 200 finishes by transferring the data from
non-volatile memory 308 to one of the NVMe.TM. disks 116 in step
408. For example, in NVMe.TM. system 100, since the storage
appliance 114 has sent the acknowledgement to the host client 102,
in a background process, CPU 126 of the storage appliance 114
initiates the transfer of the data from the non-volatile memory 308
of the storage appliance 114 to the NVMe.TM. disk 116 specified by
the write command.
[0047] In prior art systems, the storage appliance 114 relies on
DRAM to store the data transferred from the host client 102 before
transferring the data to the NVMe.TM. disk 116. Since DRAM
maintains data only for as long as power is applied, prior art
systems must wait until the data has been transferred to the
NVMe.TM. disk 116 before sending the acknowledgement to the host
client 102 that the data has been stored. Since RDMA interfaces 110
and 130 have high bandwidth and low latency, the main time
consumption in prior art systems is in the disk access.
[0048] Process 200 allows for the latency to be just the transfer
latency between the host client 102 and the storage appliance 114.
The transfer of data from the non-volatile memory 308 to the
NVMe.TM. disk 116 does not affect the latency of process 200.
However, even in a storage appliance 114 implementation of
battery-backed DRAM, the non-volatile memory 308 is relatively
small compared to the NVMe.TM. disk 116. As such, it may be
required to use another storage appliance 114 to store the incoming
data while the storage appliance 114 that was intended to receive
the data transfers the data from the non-volatile memory 308 to the
NVMe.TM. disk 116.
[0049] FIG. 3 shows a schematic representation of an NVMe.TM.
system 300 comprising a host client 102, a target storage appliance
514, and a second storage appliances 520 communicating over Fabric
112. The host client 102 is a computing system that comprises an
application 104, a CPU 106, and DRAM 108. The target storage
appliance 514 and second storage appliance 520 are computing
systems that comprises a CPU 126, non-volatile memory 308, DRAM
518, and multiple NVMe.TM. disks 116. The host client 102, target
storage appliance 514, and second storage appliance 520 interface
using RDMA 110 and 130 which is a type of network interface
controller. DRAM 518 is similar to DRAM 108.
[0050] When the host client 102 wants to write data to a particular
NVMe.TM. disk 116 in the target storage appliance 514, CPU 106 of
the host client encapsulates the write command and sends the write
command to CPU 126 of the target storage appliance 514 using the
RDMA interface 110. Application 104 sends the data to DRAM 108 of
the host client so that the data is ready for transmission.
[0051] Once the target storage appliance 514 receives the write
command using the RDMA interface 130, the CPU 126 of the target
storage appliance 514 parses the encapsulated write command,
de-capsulates it, and extracts the data information. If the
non-volatile memory 308 of the target storage appliance 514 does
not have free space for the data that the host client 102 wants to
transfer, the CPU 126 of the target storage appliance re-directs
the write command to a CPU 126 of a second storage appliance 520
using the RDMA interface 110.
[0052] When the CPU 126 of the second storage appliance 520
receives the write command re-directed from the CPU 126 of the
target storage appliance 514, it determines if it has resources to
handle the command. If the second storage appliance 520 does not
have the resources to receive the data from the host client 102,
the second storage appliance 520 would inform CPU 126 of the target
storage appliance 514 so that CPU 126 of the target storage
appliance 514 can direct the command to another communicatively
coupled storage appliance (not shown) which can receive the
data.
[0053] If the second storage appliance 520 has the resources to
receive the data, CPU 126 of the second storage appliance 520 then
initiates a transaction of the data using the RDMA interface 130
from DRAM 108 of the host client 102 to the non-volatile memory 308
of the second storage appliance 520. Following the RDMA 130
transaction, the data will reside in the non-volatile memory 308 of
the second storage appliance 520.
[0054] Once the data resides in the non-volatile memory 308 of the
second storage appliance 520, non-volatile memory 308 sends CPU 126
of the target storage appliance 514 a status that indicates that
the data has been stored in non-volatile memory 308 of second
storage appliance 520. The CPU 126 of the target storage appliance
514 then encapsulates a status that indicates that the data has
been stored in non-volatile memory 308 of the second storage
appliance 520 and sends it to the CPU 106 of the host client 102
using the RDMA interface 130. The status serves as an
acknowledgement that the data has been stored in non-volatile
memory 308 of second storage appliance 520.
[0055] In a background process, CPU 126 of the second storage
appliance 520 initiates the transfer of the data from the
non-volatile memory 308 of the second storage appliance 520 to the
DRAM 518 of the target storage appliance 514. Once the data is
stored in the DRAM 518 of the target storage appliance 514, CPU 126
of the target storage appliance 514 initiates the transfer of the
data from the DRAM 518 of the target storage appliance 514 to the
NVMe.TM. disk 116 of the target storage appliance specified by the
write command.
[0056] A process 400 of a write command using NVMe.TM. system 300
is illustrated in FIG. 4. The process 400 begins by sending an
encapsulated write command from the host client 102 to the target
storage appliance 514 in step 602. For example, in NVMe.TM. system
300, when the host client 102 wants to write data to a particular
NVMe.TM. disk 116 in the target storage appliance 514, CPU 106 of
the host client encapsulates the write command and sends the write
command to CPU 126 of the target storage appliance 514 using the
RDMA interface 110. Application 104 sends the data to DRAM 108 of
the host client so that the data is ready for transmission.
[0057] Process 400 continues by broadcasting a query with the
encapsulated write command from the target storage appliance 514 to
the second storage appliance 520 in step 604. For example, in
NVMe.TM. system 300, once the target storage appliance 514 receives
the write command using the RDMA interface 130, the CPU 126 of the
target storage appliance 514 parses the encapsulated write command,
de-capsulates it, and extracts the data information. If the
non-volatile memory 308 of the target storage appliance 514 does
not have free space for the data that the host client 102 wants to
transfer, the CPU 126 of the target storage appliance re-directs
the write command to a CPU 126 of a second storage appliance 520
using the RDMA interface 130.
[0058] Process 400 continues by determining the resource
availability of the second storage appliance 520 in step 606. For
example, in NVMe.TM. system 300, when the CPU 126 of the second
storage appliance 520 receives the write command re-directed from
the CPU 126 of the target storage appliance 520, it determines if
it has resources to handle the command. If the second storage
appliance 520 does not have the resources to receive the data from
the host client 102, the second storage appliance 520 would inform
CPU 126 of the target storage appliance 514 so that CPU 126 of the
target storage appliance 514 can direct the command to another
communicatively coupled storage appliance (not shown) which can
receive the data.
[0059] Process 400 continues by transferring data from the DRAM 108
of the host client 102 to the non-volatile memory 308 of the second
storage appliance 520 in step 608. For example, in NVMe.TM. system
300, if the second storage appliance 520 has the resources to
receive the data, CPU 126 of the second storage appliance 520 then
initiates a transaction of the data using the RDMA interface 130
from DRAM 108 of the host client 102 to the non-volatile memory 308
of the second storage appliance 520. Following the RDMA 130
transaction, the data will reside in the non-volatile memory 308 of
the second storage appliance 520.
[0060] Process 400 continues by sending an acknowledgement to the
host client 102 from the CPU 126 of target storage appliance 514 in
step 610. For example, in NVMe.TM. system 300, once the data
resides in the non-volatile memory 308 of the second storage
appliance 520, non-volatile memory 308 sends CPU 126 of the target
storage appliance 114 a status that indicates that the data has
been stored in non-volatile memory 308 of second storage appliance
520. The CPU 126 of the target storage appliance 514 then
encapsulates a status that indicates that the data has been stored
in non-volatile memory 308 of the second storage appliance 520 and
sends it to the CPU 106 of the host client 102 using the RDMA
interface 130. The status serves as an acknowledgement that the
data has been stored in non-volatile memory 308 of second storage
appliance 520.
[0061] Process 400 continues by transferring the data from the
non-volatile memory 308 of the second storage appliance to the DRAM
518 of the target storage appliance 514 in step 612. For example,
in NVMe.TM. system 300, in a background process, CPU 126 of the
second storage appliance 520 initiates the transfer of the data
from the non-volatile memory 308 of the second storage appliance
520 to the DRAM 518 of the target storage appliance 514.
[0062] Process 400 finishes by transferring the data from the DRAM
518 to one of the target storage appliance 514 to one of the
NVMe.TM. disks 116 of the target storage appliance 514 in step 614.
For example, in NVMe.TM. system 300, Once the data is stored in the
DRAM 518 of the target storage appliance 514, CPU 126 of the target
storage appliance 514 initiates the transfer of the data from the
DRAM 518 of the target storage appliance 514 to the NVMe.TM. disk
116 of the target storage appliance specified by the write
command.
[0063] FIG. 5 shows a schematic representation of an NVMe.TM.
system 500 comprising two host clients 102, a target storage
appliance 514, and a second storage appliances 520 communicating
over Fabric 112. The host clients 102 are computing systems that
comprise an application 104, a CPU 106, and DRAM 108. The target
storage appliance 514 and second storage appliances 520 are
computing systems that comprises a CPU 126, DRAM 518, and multiple
NVMe.TM. disks. The host clients 102, target storage appliance 514,
and second storage appliance 520 interface using RDMA 110 and 130
which is a type of network interface controller. NVMe.TM. system
500 comprises a non-volatile memory tier 708 and NVMe.TM. tier 716.
Non-volatile memory tier 708 is a collection of multiple
non-volatile memory devices 308. NVMe.TM. tier 716 is a collection
of multiple NVMe.TM. disks 116. NVMe.TM. system 500 may comprise
more than two host clients 102 and more than one second storage
appliance 520.
[0064] NVMe.TM. system 500 allows for the allocation of NVRAM in a
large scale storage deployment, enabling the non-volatile memory
tier 708. When a host client 102 wants to write data to a
particular NVMe.TM. disk 116 in a target storage appliance 514, CPU
106 of the host client encapsulates the write command and sends the
write command to CPU 126 of the target storage appliance 514 using
the RDMA interface 110. The target storage appliance 514 will
utilize the non-volatile memory tier 708 to transfer the data from
the DRAM 108 of the host client 102 and, further, transfer the data
from the non-volatile memory tier 708 to the particular NVMe.TM.
disk of the NVMe.TM. tier 716.
[0065] For example, if the target storage appliance 514 has enough
local resources, the target storage appliance 514 will store the
data in the local portion of the non-volatile memory tier 708. The
target storage appliance 514 will then send an acknowledgment
status to the host client 102. If the target storage appliance 514
does not have enough local resources, the target storage appliance
514 will broadcast a query to all of the second storage appliances
520 in the network, asking if there is a second storage appliance
520 that can receive the data. One of the second storage appliances
520 with available resources sends a signal to the target storage
appliance 514 indicating that it can receive the data. The process
continues as described by process 400.
[0066] In order to allocate the valuable and limited resources of
the non-volatile memory tier 708, fair and user defined mechanisms
should be defined. FIG. 6 shows a schematic representation of an
NVRAM device 802. NVRAM 802 is an example of a user defined
mechanism for allocating the valuable and limited resources of the
non-volatile memory tier 708. NVRAM 802 is a non-volatile
semiconductor memory device comprising storage space divided
a-priori by an administrator into reserved area 804, local area
812, and remote area 814. NVRAM 802 is an example of a non-volatile
memory device 308.
[0067] Reserved area 804 defines the guaranteed storage area
available for each host client 102. The reserved area 804 is
further divided into multiple areas, each area dedicated to a host
client 102. Host 1 806 is storage area dedicated to the first host
client 102. Host 2 808 is storage area dedicated to the second host
client 102. Host N 810 is storage area dedicated to the N.sup.th
host client 102.
[0068] Local area 812 is storage area allocated for data sent to
NVRAM 802 for storage in a local NVMe.TM. disk 116. Local area 812
functions as extra space when the reserved area 804 that
corresponds to a particular host client 102 is full. For example,
if host 1 storage area 806 corresponding to host client 102 is full
and NVRAM 802 receives data from host client 102, the new data may
be stored in the local area 812.
[0069] Remote area 814 is storage area allocated for data sent to
NVRAM 802 for storage in a remote NVMe.TM. disk 116. Remote area
814 functions as extra space when the reserved area 804 of a target
storage appliance 514 is full and a second storage appliance 520 is
needed to receive the data. For example, if the NVRAM 802 of a
target storage appliance 514 is full, the NVRAM 802 of a second
storage appliance 520 can receive the data and store it in the
remote storage area 814.
[0070] A process 700 of allocating space for a write command in an
NVRAM device 802 is illustrated in FIG. 7. The process 700 begins
by sending an encapsulated write command from the host client 102
to the target storage appliance 514 in step 902. For example, CPU
106 of the host client encapsulates the write command and sends the
write command to CPU 106 of the target storage appliance 514 using
the RDMA interface 110.
[0071] The process 700 continues by determining the command
preference of the encapsulated write command in step 904. The
encapsulated write command from the host client 102 may have a
command preference that specifies if the write command is intended
for storage at the target storage appliance 514 or a second storage
appliance 520. If the command preference is to send the data to a
second storage appliance 520, the process 700 continues to step
906. If the command preference is to send the data to the target
storage appliance 514, the process 700 continues to step 908.
[0072] At step 906, the target storage appliance 514 broadcasts a
query with the write command to a second storage appliance 520. For
example, the CPU 106 of the target storage appliance 514 sends the
write command to a CPU 106 of a second storage appliance 520 using
the RDMA interface 110. The CPU 106 of the target storage appliance
514 may initiate the RDMA interface 110 to broadcast a query to the
second storage appliance 520 using the RDMA interface 110. After
step 906, process 700 continues as described by process 400,
described above.
[0073] At step 908, process 700 continues by determining if there
is available reserved area 804 dedicated to the host client 102. If
there is available reserved area 804 dedicated to the host client
102, process 700 continues to step 910. If there is no available
reserved area 804 dedicated to the host client 102, process 700
continues to step 912.
[0074] At step 910, process 700 continues by allocating space from
the reserved area 804 for data storage. For example, if host 1
storage area 806 corresponding to host client 102 is available,
host 1 storage area 806 may be used to store the incoming data from
host client 102. After step 910, process 700 continues as described
by process 200, described above.
[0075] At step 912, process 700 continues by determining if there
is available local area 812. If there is no available local area
812, process 700 continues to step 906. If there is available local
area 812, process 700 continues to step 914.
[0076] At step 914, process 700 continues by determining if the
host client 102 has reached a threshold limit of allocation in the
local area 812. If the host client 102 has reached a threshold
limit of allocation in the local area 812, process 700 continues to
step 906. If the host client 102 has not reached a threshold limit
of allocation in the local area 812, process 700 continues to step
916.
[0077] At step 916, process 700 determines if the priority of the
write command is higher than a threshold priority. If the priority
of the write command is not higher than the threshold priority,
process 700 continues to step 906. If the priority of the write
command is higher than the threshold priority, process 700
continues to step 918.
[0078] Process 700 finishes by allocating space from the local area
812 for data storage in step 918. For example, if host 1 storage
area 806 corresponding to host client 102 is full and NVRAM 802
receives data from host client 102, the new data may be stored in
the local area 812.
[0079] A similar process occurs when a command is received from a
second host client 102. In this situation, the target storage
appliance 514 checks remote area 814 for available space. If there
is space available in remote area 814, the target storage appliance
514 allocates space from the remote area 814 for the incoming
command.
[0080] Other objects, advantages and embodiments of the various
aspects of the present invention will be apparent to those who are
skilled in the field of the invention and are within the scope of
the description and the accompanying Figures. For example, but
without limitation, structural or functional elements might be
rearranged consistent with the present invention. Similarly,
principles according to the present invention could be applied to
other examples, which, even if not specifically described here in
detail, would nevertheless be within the scope of the present
invention.
* * * * *