Computational storage for distributed computing

• US 10,423,575 B2
• Filed: 03/02/2017
• Issued: 09/24/2019
• Est. Priority Date: 03/02/2017
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Hack, Michel H. T., Ren, Yufei, Tan, Wei, Wang, Yandong, Wu, Xingbo, Zhang, Li, Zhang, Wei
• Primary Examiner(s): Mackes, Kris E
• Application Number:
  US15/447,262
• Publication Number:
  US 20180253423A1
• Time in Patent Office: 936 Days
• CPC Class Codes: G06F 16/14 Details of searching files based on file metadata
• CPC Class Codes: G06F 16/1824 implemented using Network-attached Storage [NAS] architecture distributed or networked storage systems G06F3/067; protocols for distributed storage of data in a network H04L67/1097

Abstract

Computational storage techniques for distribute computing are disclosed. The computational storage server receives input from multiple clients, which is used by the server when executing one or more computation functions. The computational storage server can aggregate multiple client inputs before applying one or more computation functions. The computational storage server sets up: a first memory area for storing input received from multiple clients; a second memory area designated for storing the computation functions to be executed by the computational storage server using the input data received from the multiple clients; a client specific memory management area for storing metadata related to computations performed by the computational storage server for specific clients; and a persistent storage area for storing checkpoints associated with aggregating computations performed by the computation functions.

Claims

1. A computer-implemented method comprising:

setting up, by a computational storage server, a computational storage media in the computational storage server, wherein the computational storage media is set up to comprise a first memory area for storing the input data received from multiple clients, a second memory area designated for storing computation functions that are executed by the computational storage server using the input data received from the multiple clients, a client specific management metadata area in the computational storage media for storing metadata related to computations performed in the computational storage server for specific clients, and a persistent storage area designated for storing checkpoints for aggregating computations performed by the computation functions;

detecting, by the computational storage server, an indication from one or more clients from the multiple clients indicating a probe-ready status for the computational storage media;

in response to one or more clients from the multiple clients indicating the probe-ready status for the computational storage media, exposing, by the computational storage server, the computational storage media to the multiple clients as an available storage media;

receiving, by the computational storage server, input data from the multiple clients;

executing, within the computational storage server, one or more computation functions using the received input data as inputs;

aggregating, by the computational storage server, outputs of the one or more computation functions that used the received input data as inputs; and

returning, from the computational storage server to the multiple clients, aggregated outputs of the one or more computation functions that used the received input data as inputs.

2. The computer-implemented method of claim 1, further comprising:

setting up, by the computational storage server, the persistent storage area in the computational storage media that is designated for storing and returning the checkpoints used during said aggregating the outputs of the one or more computation functions.

3. The computer-implemented method of claim 1, wherein the computational storage server communicates using POSIX-compliant I/O interfaces.

4. The computer-implemented method of claim 1, wherein the computational storage server is a management computational storage server, wherein the one or more computation functions are a matrix of multiple different computation functions that execute using different computer languages, and wherein the computer-implemented method further comprises:

splitting, by the management computational storage server, the matrix into multiple sub-matrixes of computation functions, wherein each of the multiple sub-matrixes is written in a different computer language than other sub-matrixes from the multiple sub-matrixes;

storing, by the management computational storage server, each of the multiple sub-matrixes into a different computational storage server, wherein each different computational storage server is specific for a particular computer language;

receiving, by the management computational storage server, first data from a first client that is written in a first computer language;

receiving, by the management computational storage server, second data from a second client that is written in a second computer language;

transmitting, by the management computational storage server, the first data to a first computational storage server than contains a first sub-matrix computational function that runs under the first computer language;

transmitting, by the management computational storage server, the second data to a second computational storage server than contains a second sub-matrix computational function that runs under the second computer language;

receiving, by the management computational storage server, a first result from the first computational storage server and a second result from the second computational storage server, wherein the first result is an output of the first sub-matrix computational function using the first data, and wherein the second result is an output of the second sub-matrix computational function using the second data;

consolidating, by the management computational storage server, the first result and the second result into a consolidated result; and

sending, from the management computational storage server, the consolidated result to the first client and the second client.

5. The computer-implemented method of claim 4, further comprising:

receiving, by the first computational storage server, a set of global parameters from a global parameter server, wherein the set of global parameters comprises data that weights one or more operands used in the first sub-matrix computational function; and

executing, by multiple learner processors in the first computational storage server, the first sub-matrix computational function using the set of global parameters.

6. The computer-implemented method of claim 1, further comprising:

striping, by the computational storage server, the one or more computation functions across a RAID storage system.

7. A computer program product for aggregating computation operations and their outputs from a computational storage server, wherein the computational storage server is a management computational storage server, the computer program product comprising a computer readable storage device having program instructions embodied therewith, the program instructions readable and executable by a computer to cause:

receiving input data from multiple clients;

executing one or more computation functions using the received input data as inputs, wherein the one or more computation functions are a matrix of multiple different computation functions that execute using different computer languages;

aggregating, by the computational storage server, outputs of the one or more computation functions that used the received input data as inputs;

returning, to the multiple clients, aggregated outputs of the one or more computation functions that used the received input data as inputs;

splitting, by the management computational storage server, the matrix into multiple sub-matrixes of computation functions, wherein each of the multiple sub-matrixes is written in a different computer language than other sub-matrixes from the multiple sub-matrixes;

storing, by the management computational storage server, each of the multiple sub-matrixes into a different computational storage server, wherein each different computational storage server is specific for a particular computer language;

receiving, by the management computational storage server, first data from a first client that is written in a first computer language;

receiving, by the management computational storage server, second data from a second client that is written in a second computer language;

transmitting, by the management computational storage server, the first data to a first computational storage server than contains a first sub-matrix computational function that runs under the first computer language;

transmitting, by the management computational storage server, the second data to a second computational storage server than contains a second sub-matrix computational function that runs under the second computer language;

receiving, by the management computational storage server, a first result from the first computational storage server and a second result from the second computational storage server, wherein the first result is an output of the first sub-matrix computational function using the first data, and wherein the second result is an output of the second sub-matrix computational function using the second data;

consolidating, by the management computational storage server, the first result and the second result into a consolidated result;

sending, from the management computational storage server, the consolidated result to the first client and the second client;

receiving, by the first computational storage server, a set of global parameters from a global parameter server, wherein the set of global parameters comprises data that weights one or more operands used in the first sub-matrix computational function; and

executing, by multiple learner processors in the first computational storage server, the first sub-matrix computational function using the set of global parameters.

8. The computer program product of claim 7, wherein the program instructions are further readable and executable by the computer to cause:

setting up, by the computational storage server, a computational storage media in the computational storage server, wherein the computational storage media is set up to comprise a first memory area for storing the input data received from the multiple clients, a second memory area designated for storing computation functions that are executed by the computational storage server using the input data received from the multiple clients, a client specific management metadata area in the computational storage media for storing metadata related to computations performed in the computational storage server for specific clients, and a persistent storage area designated for storing checkpoints for aggregating computations performed by the computation functions;

detecting, by the computational storage server, an indication from one or more clients from the multiple clients indicating a probe-ready status for the computational storage media;

in response to one or more clients from the multiple clients indicating the probe-ready status for the computational storage media, exposing, by the computational storage server, the computational storage media to the multiple clients as an available storage media; and

aggregating, by the computational storage server, outputs of one or more aggregated computation functions at the checkpoints based on the metadata.

9. The computer program product of claim 7, wherein the multiple clients transmit the updated data to the computational storage server using POSIX-compliant I/O interfaces.

10. The computer program product of claim 7, wherein the method further comprises:

striping, by the computational storage server, the one or more computation functions across a RAID storage system.

11. The computer program product of claim 7, wherein the program instructions are provided as a service in a cloud environment.

12. A computer system comprising:

one or more processors;

one or more computer readable memories, operably coupled to the one or more processors, wherein the one or more computer readable memories store program instructions for execution by at least one of the one or more processors, the stored program instructions comprising;

program instructions to set up a computational storage media in a computational storage server, wherein the computational storage media is set up to comprise a first memory area for storing input data received from multiple clients, a second memory area designated for storing computation functions executed by the computational storage server using the input data received from the multiple clients, a client specific management metadata area in the computational storage media for storing metadata related to computations performed in the computational storage server for specific clients, and a persistent storage area designated for storing checkpoints for aggregating computations performed by the computation functions;

program instructions to detect an indication from one or more clients from the multiple clients indicating a probe-ready status for the computational storage media;

program instructions to, in response to one or more clients from the multiple clients indicating the probe-ready status for the computational storage media, expose the computational storage media to the multiple clients as an available storage media;

program instructions to receive input data from the multiple clients;

program instructions to execute one or more of the computation functions using the received input data as inputs;

program instructions to aggregate outputs of the one or more computation functions that used the received input data as inputs; and

program instructions to return, to the multiple clients, aggregated outputs of the one or more computation functions that used the received input data as inputs.

13. The computer system of claim 12, further comprising:

program instructions to return, to the computational storage server that is aggregating the outputs of the one or more computation functions, checkpoints used during said aggregating the outputs of the one or more computation functions.

14. The computer system of claim 12, wherein the multiple clients transmit the updated data to the computational storage server using POSIX-compliant I/O interfaces.

15. The computer system of claim 12, wherein the computer system is a management computational storage server, wherein the one or more computation functions are a matrix of multiple different computation functions that execute using different computer languages, and wherein the computer system further comprises:

program instructions to split, by the management computational storage server, the matrix into multiple sub-matrixes of computation functions, wherein each of the multiple sub-matrixes is written in a different computer language than other sub-matrixes from the multiple sub-matrixes;

program instructions to store, by the management computational storage server, each of the multiple sub-matrixes into a different computational storage server, wherein each different computational storage server is specific for a particular computer language;

program instructions to receive, by the management computational storage server, first data from a first client that is written in a first computer language;

program instructions to receive, by the management computational storage server, second data from a second client that is written in a second computer language;

program instructions to transmit, by the management computational storage server, the first data to a first computational storage server than contains a first sub-matrix computational function that runs under the first computer language;

program instructions to transmit, by the management computational storage server, the second data to a second computational storage server than contains a second sub-matrix computational function that runs under the second computer language;

program instructions to receive, by the management computational storage server, a first result from the first computational storage server and a second result from the second computational storage, wherein the first result is an output of the first sub-matrix computational function using the first data, and wherein the second result is an output of the second sub-matrix computational function using the second data;

program instructions to consolidate, by the management computational storage server, the first result and the second result into a consolidated result; and

program instructions to send, from the management computational storage server, the consolidated result to the first client and the second client.

16. The computer system of claim 15, further comprising:

program instructions to receive, by the first computational storage server, a set of global parameters from a global parameter server, wherein the set of global parameters comprises data that weights one or more operands used in the first sub-matrix computational function; and

program instructions to execute, by multiple learner processors in the first computational storage server, the first sub-matrix computational function using the set of global parameters.

17. The computer system of claim 12, further comprising:

program instructions to stripe the one or more computation functions across a RAID storage system.

18. The computer system of claim 12, wherein the stored program instructions are provided as a service in a cloud environment.