Dynamic allocation of FPGA resources

Patent metadata

Field	Value
Publication number	US-11042413-B1
Application number	US-201916508100-A
Country	US
Kind code	B1
Filing date	Jul 10, 2019
Priority date	Jul 10, 2019
Publication date	Jun 22, 2021
Grant date	Jun 22, 2021

Abstract

Utilization metrics associated with a plurality of eligible compute tasks executable on a plurality of field-programmable gate arrays are determined. The utilization metrics are evaluated to dynamically identify a reallocation of an identified field-programmable gate array resource of the field-programmable gate arrays to handle a selected eligible compute task of the eligible compute tasks. The identified field-programmable gate array resource is automatically reprogrammed to handle the selected eligible compute task.

First claim

What is claimed is:

1. A system, comprising:
field-programmable gate arrays; and
a processor configured to:
determine utilization metrics associated with a plurality of eligible compute tasks executable on the field-programmable gate arrays;
evaluate the utilization metrics to dynamically identify a reallocation of an identified field-programmable gate array resource of the field-programmable gate arrays to handle a selected eligible compute task of the eligible compute tasks; and
automatically reprogram the identified field-programmable gate array resource to handle the selected eligible compute task;
wherein the processor is configured to evaluate the utilization metrics to dynamically identify the reallocation of the identified field-programmable gate array resource including by being configured to at least in part determine whether a first eligible compute task has been performed more frequently than a second eligible compute task.

2. The system of claim 1 , wherein the utilization metrics are associated with one or more of the following: queue depths of the field-programmable gate arrays or execution times to perform the eligible compute tasks.

3. The system of claim 1 , wherein the processor is further configured to evaluate one or more of the following: time of day, calendar date, or time zone to dynamically identify the reallocation of the identified field-programmable gate array resource.

4. The system of claim 1 , wherein the processor is configured to automatically reprogram the identified field-programmable gate array resource including by being configured to transfer a bitstream to the identified field-programmable gate array resource from a persistent storage device.

5. The system of claim 4 , wherein the bitstream is a part of a plurality of bitstreams included in a function repository.

6. The system of claim 1 , wherein the processor is further configured to invoke the identified field-programmable gate array resource to handle the selected eligible compute task.

7. The system of claim 1 , wherein the processor is further configured to report a result associated with the selected eligible compute task.

8. The system of claim 7 , wherein the result is formatted as an application programming interface response object.

9. The system of claim 1 , wherein the field-programmable gate arrays are located in a single server cluster.

10. The system of claim 1 , wherein the field-programmable gate arrays are located across multiple server clusters.

11. The system of claim 1 , wherein the processor is further configured to receive a request to perform the selected eligible compute task on the reprogrammed field-programmable gate array resource.

12. The system of claim 11 , wherein the request is received via an application programming interface.

13. The system of claim 11 , wherein the request originates from a software application.

14. A system, comprising:
field-programmable gate arrays; and
a processor configured to:
determine utilization metrics associated with a plurality of eligible compute tasks executable on the field-programmable gate arrays, wherein the utilization metrics are formulated as cost functions that include, as components, queue depths of the field-programmable gate arrays and execution times to perform the eligible compute tasks;
evaluate the utilization metrics to dynamically identify a reallocation of an identified field-programmable gate array resource of the field-programmable gate arrays to handle a selected eligible compute task of the eligible compute tasks; and
automatically reprogram the identified field-programmable gate array resource to handle the selected eligible compute task.

15. The system of claim 14 , wherein the cost functions include, as a component, reprogramming times of the field-programmable gate arrays.

16. The system of claim 14 , wherein the processor is configured to automatically reprogram the identified field-programmable gate array resource including by being configured to transfer a bitstream to the identified field-programmable gate array resource from a persistent storage device.

17. The system of claim 14 , wherein the processor is further configured to report a result associated with the selected eligible compute task.

18. The system of claim 14 , wherein the field-programmable gate arrays are located across multiple server clusters.

19. The system of claim 14 , wherein the processor is further configured to receive a request to perform the selected eligible compute task on the reprogrammed field-programmable gate array resource.

20. A system, comprising:
field-programmable gate arrays; and
a processor configured to:
determine utilization metrics associated with a plurality of eligible compute tasks executable on the field-programmable gate arrays;
evaluate the utilization metrics to dynamically identify a reallocation of an identified field-programmable gate array resource of the field-programmable gate arrays to handle a selected eligible compute task of the eligible compute tasks;
evaluate current events that generate social media activity to dynamically identify the reallocation of the identified field-programmable gate array resource; and
automatically reprogram the identified field-programmable gate array resource to handle the selected eligible compute task.

Assignees

• Facebook Inc

Inventors

• Byagowi Ahmad
• Lambeta Michael Maroye
• Mroz Martin

Classifications

• G06F2209/509

  Offload · CPC title

• G06F2209/501

  Performance criteria · CPC title

• G06F9/5027 Primary

  the resource being a machine, e.g. CPUs, Servers, Terminals · CPC title

• G06F9/505

  considering the load · CPC title

• G06F30/34

  for reconfigurable circuits, e.g. field programmable gate arrays [FPGA] or programmable logic devices [PLD] · CPC title