System and method for machine learning with NVMe-of ethernet SSD chassis with embedded GPU in SSD form factor

Patent metadata

Field	Value
Publication number	US-10417733-B2
Application number	US-201715672223-A
Country	US
Kind code	B2
Filing date	Aug 8, 2017
Priority date	May 24, 2017
Publication date	Sep 17, 2019
Grant date	Sep 17, 2019

Abstract

In one aspect of the present disclosure, a data storage and processing system is provided. The system includes a host server and a storage unit. The storage unit includes a drive comprising a memory and a drive processor, an external switch configured to couple the host server to the drive to send and receive data between the host server and the memory of the drive and a graphics processing unit. The drive processor is configured to send processing instructions and data from the drive memory to the graphics processing unit and the graphics processing unit is configured to process the data according to the processing instructions to generate result data.

First claim

What is claimed is:

1. A data storage and processing system comprising:
a host server; and
a storage unit, the storage unit comprising:
a drive comprising a drive memory and a drive processor;
an external switch configured to couple the host server to the drive to send and receive data between the host server and the drive memory of the drive; and
a graphics processing unit, wherein the drive processor is configured to send processing instructions and data from the drive memory to the graphics processing unit and the graphics processing unit is configured to process the data according to the processing instructions to generate result data,

wherein the drive comprises a table, and the drive processor is configured to:
identify a data chunk stored on the drive memory and a transform to be applied to the data chunk, wherein the table indicates that the graphics processing unit is capable of performing the transform; and
send the data chunk to the graphics processing unit.

2. The system of claim 1 , wherein the graphics processing unit comprises a U.2 connector, the graphics processing unit is connected to the drive through the U.2 connector, and the graphics processing unit receives the processing instructions and data through the U.2 connector.

3. The system of claim 1 , wherein the graphics processing unit is configured to send the result data to the drive processor and the drive processor is configured to store the result data in the drive memory.

4. The system of claim 1 , wherein the graphics processing unit is coupled to the external switch, and wherein the graphics processing unit is configured to send the result data to the host server utilizing the external switch.

5. The system of claim 4 , wherein the graphics processing unit is configured to send a status message to the drive processor upon completing the processing of the data according to the processing instructions.

6. The system of claim 1 , wherein the drive processor is configured to:
retrieve an address of the graphics processing unit from the table; and
send the data chunk to the address of the graphics processing unit.

7. The system of claim 1 , wherein the storage unit further comprises a baseboard management controller configured to determine capabilities of the graphics processing unit and to update the table based on the capabilities of the graphics processing unit.

8. The system of claim 7 , wherein:
the storage unit comprises a plurality of said drive, each drive having a separate table;
the storage unit comprises a plurality of said graphics processing unit; and
the baseboard management controller is configured to update each table of each drive of the plurality of drives.

9. The system of claim 7 , wherein the baseboard management controller and the graphics processing unit are configured to communicate using a Non Volatile Memory Express-Management Interface protocol, and wherein the baseboard management controller determines the capabilities of the graphics processing unit by utilizing a Non Volatile Memory Express-Management Interface Identify command.

10. The system of claim 7 , wherein the storage unit comprises a plurality of said graphics processing unit, and wherein the baseboard management controller is configured to determine a load for each graphics processing unit of the plurality of graphics processing units and update the table based on the load for each graphics processing unit of the plurality of graphics processing units.

11. The system of claim 1 , wherein the external switch is an Ethernet switch and the drives are Ethernet solid state drives.

12. The system of claim 1 , wherein the host server communicates with the storage unit using a remote direct access storage protocol.

13. The system of claim 1 , wherein the storage unit further comprises a chassis and a baseboard management controller, the graphics processing unit is a field replaceable unit, and the baseboard management controller is configured to detect a plug-in event corresponding to the graphics processing unit being inserted into the chassis.

14. A data storage and processing system comprising:
a host server; and
a storage unit, the storage unit comprising:
a drive comprising a drive memory and a drive processor;
an external switch configured to couple the host server to the drive to send and receive data between the host server and the drive memory of the drive; and
a graphics processing unit, wherein the drive processor is configured to send processing instructions and data from the drive memory to the graphics processing unit and the graphics processing unit is configured to process the data according to the processing instructions to generate result data,

wherein the host server is configured to send a trigger instruction to the drive processor, and the drive processor is configured to execute a data processing algorithm in response to the trigger instruction.

15. The data storage and processing system of claim 14 , wherein the data processing algorithm comprises steps which when executed cause the drive processor to send the data and the processing instructions to the graphics processing unit.

16. A method of operating a data storage and processing system comprising a drive and a graphics processing unit, the drive comprising a drive processor and a drive memory, the method comprising:
receiving, by the drive processor, data from a host server;
storing, by the drive processor, the data on the drive memory;
detecting, by the drive processor, a trigger;
executing, by the drive processor, an algorithm in response to the trigger, the algorithm identifying a data chunk of the data stored on the drive memory and a transform to apply to the data chunk;
searching, by the drive processor, a capability table for an address corresponding to a device for performing the transform; and
sending, by the drive processor, the data chunk and processing instructions for processing the data chunk to the address, the address being the address of the graphics processing unit.

17. The method of claim 16 , further comprising:
receiving, by the graphics processing unit, the data chunk and the processing instructions; and
processing, by the graphics processing unit, the data chunk according to the processing instructions to generate result data.

18. The method of claim 17 , further comprising:
sending, by the graphics processing unit, the result data to the drive processor; and
storing, by the drive processor, the result data on the drive memory.

19. The method of claim 16 , further comprising:
detecting, by a baseboard management controller, a capability of the graphics processing unit; and
updating, by the baseboard management controller, the capability table based on the capability of the graphics processing unit.

20. The method of claim 19 , wherein the data storage and processing system comprises a plurality of said drive, each drive of the plurality of drives comprises a capability table, and wherein the method comprises updating, by the baseboard management controller, each capability table of each drive of the plurality of drives based on the capability of the graphics processing unit.

21. A data storage and processing system comprising a drive and a graphics processing unit, the drive comprising a drive processor and a drive memory, the system comprising:
means for receiving data from a host server;
means for storing the data on the drive memory;
means for detecting a trigger;
means for executing an alg

Assignees

• Samsung Electronics Co Ltd

Inventors

• Kachare Ramdas P
• Olarig Sompong Paul
• Schwaderer David

Classifications

• G06F16/1847 Primary

  specifically adapted to static storage, e.g. adapted to flash memory or SSD · CPC title

• G06F13/1684 Primary

  using multiple buses · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F3/0688

  Non-volatile semiconductor memory arrays · CPC title

• G06T1/20 Primary

  Processor architectures; Processor configuration, e.g. pipelining · CPC title