Data transformation caching in an artificial intelligence infrastructure

• US 10,671,435 B1
• Filed: 07/20/2018
• Issued: 06/02/2020
• Est. Priority Date: 10/19/2017
• Status: Active Grant

• Current Assignee: Pure Storage, Inc.
• Original Assignee: Pure Storage, Inc.
• Inventors: Gold, Brian, Watkins, Emily, Jibaja, Ivan, Ostrovsky, Igor, Kim, Roy
• Primary Examiner(s): Shin, Kyung H
• Application Number:
  US16/040,996
• Time in Patent Office: 683 Days
• CPC Class Codes: G06F 16/1794 Details of file format conversion
• CPC Class Codes: G06F 16/245 Query processing
• CPC Class Codes: G06F 16/248 Presentation of query results
• CPC Class Codes: G06F 16/972 Access to data in other repository systems, e.g. legacy data or dynamic Web page generation
• CPC Class Codes: G06F 3/0604 Improving or facilitating administration, e.g. storage management
• CPC Class Codes: G06F 3/0608 Saving storage space on storage systems
• CPC Class Codes: G06F 3/0646 Horizontal data movement in storage systems, i.e. moving data in between storage devices or systems
• CPC Class Codes: G06F 3/0649 Lifecycle management
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
• CPC Class Codes: G06F 3/0679 Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP]
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06F 9/5027 the resource being a machine, e.g. CPUs, Servers, Terminals
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06Q 30/0243 Comparative campaigns
• CPC Class Codes: G06T 1/20 Processor architectures; Processor configuration, e.g. pipelining
• CPC Class Codes: G06T 1/60 Memory management
• CPC Class Codes: G06T 2200/28 involving image processing hardware

Abstract

Data transformation caching in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (‘GPU’) servers, including: identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to a dataset; generating, in dependence upon the one or more transformations, a transformed dataset; storing, within one or more of the storage systems, the transformed dataset; receiving a plurality of requests to transmit the transformed dataset to one or more of the GPU servers; and responsive to each request, transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset.

Claims

1. A method of data transformation caching in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (‘

GPU’

) servers, the method comprising;

identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to a dataset;

generating, in dependence upon the one or more transformations, a transformed dataset;

storing, within one or more of the storage systems, the transformed dataset;

receiving a plurality of requests to transmit the transformed dataset to one or more of the GPU servers; and

responsive to each request, transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset.

2. The method of claim 1 wherein generating, in dependence upon the one or more transformations, a transformed dataset further comprises generating, by the storage system in dependence upon the one or more transformations, transformed dataset.

3. The method of claim 1 wherein transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset further comprises transmitting the transformed dataset from the one or more storage systems directly to application memory on the GPU servers.

4. The method of claim 3 wherein transmitting the transformed dataset from the one or more storage systems directly to application memory on the GPU servers further comprises transmitting the transformed data dataset from the one or more storage systems to the GPU servers via remote direct memory access (‘

RDMA’

).

5. The method of claim 1 further comprising executing, by one or more of the GPU servers, one or more machine learning algorithms associated with the machine learning model using the transformed dataset as input.

6. The method of claim 1 further comprising:

scheduling, by a unified management plane, one or more transformations for one or more of the storage systems to apply to the dataset; and

scheduling, by the unified management plane, execution of one or more machine learning algorithms associated with the machine learning model by the one or more GPU servers.

7. The method of claim 1 further comprising providing, by a unified management plane to the one or more GPU servers, information describing the dataset, the one or more transformations applied to the dataset, and the transformed dataset.

8. An artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (‘

GPU’

) servers, the artificial intelligence infrastructure configured to carry out the steps of;

identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to a dataset;

generating, in dependence upon the one or more transformations, a transformed dataset;

storing, within one or more of the storage systems, the transformed dataset;

receiving a plurality of requests to transmit the transformed dataset to one or more of the GPU servers; and

responsive to each request, transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset.

9. The artificial intelligence infrastructure of claim 8 wherein generating, in dependence upon the one or more transformations, a transformed dataset further comprises generating, by the storage system in dependence upon the one or more transformations, transformed dataset.

10. The artificial intelligence infrastructure of claim 8 wherein transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset further comprises transmitting the transformed dataset from the one or more storage systems directly to application memory on the GPU servers.

11. The artificial intelligence infrastructure of claim 10 wherein transmitting the transformed dataset from the one or more storage systems directly to application memory on the GPU servers further comprises transmitting the transformed data dataset from the one or more storage systems to the GPU servers via remote direct memory access (‘

RDMA’

).

12. The artificial intelligence infrastructure of claim 8 wherein the artificial intelligence infrastructure is further configured to carry out the step of executing, by one or more of the GPU servers, one or more machine learning algorithms associated with the machine learning model using the transformed dataset as input.

13. The artificial intelligence infrastructure of claim 8 wherein the artificial intelligence infrastructure is further configured to carry out the steps of:

scheduling, by a unified management plane, one or more transformations for one or more of the storage systems to apply to the dataset; and

scheduling, by the unified management plane, execution of one or more machine learning algorithms associated with the machine learning model by the one or more GPU servers.

14. The artificial intelligence infrastructure of claim 8 wherein the artificial intelligence infrastructure is further configured to carry out the step of providing, by a unified management plane to the one or more GPU servers, information describing the dataset, the one or more transformations applied to the dataset, and the transformed dataset.

15. An apparatus for data transformation offloading in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (‘

GPU’

) servers, the apparatus comprising a computer processor, a computer memory operatively coupled to the computer processor, the computer memory having disposed within it computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the steps of;

identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to a dataset;

generating, in dependence upon the one or more transformations, a transformed dataset;

storing, within one or more of the storage systems, the transformed dataset;

receiving a plurality of requests to transmit the transformed dataset to one or more of the GPU servers; and

responsive to each request, transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset.

16. The apparatus of claim 15 wherein generating, in dependence upon the one or more transformations, a transformed dataset further comprises generating, by the storage system in dependence upon the one or more transformations, transformed dataset.

17. The apparatus of claim 15 wherein transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset further comprises transmitting the transformed dataset from the one or more storage systems directly to application memory on the GPU servers.

18. The apparatus of claim 15 further comprising computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the steps of:

scheduling, by a unified management plane, one or more transformations for one or more of the storage systems to apply to the dataset; and

scheduling, by the unified management plane, execution of one or more machine learning algorithms associated with the machine learning model by the one or more GPU server.

19. The apparatus of claim 15 further comprising computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the step of providing, by a unified management plane to the one or more GPU servers, information describing the dataset, the one or more transformations applied to the dataset, and the transformed dataset.

20. The apparatus of claim 15 further comprising computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the step of executing, by one or more of the GPU servers, one or more machine learning algorithms associated with the machine learning model using the transformed dataset as input.