Prioritizing power delivery to processing units using historical workload information

Patent metadata

Field	Value
Publication number	US-10884482-B2
Application number	US-201816117125-A
Country	US
Kind code	B2
Filing date	Aug 30, 2018
Priority date	Aug 30, 2018
Publication date	Jan 5, 2021
Grant date	Jan 5, 2021

Abstract

A computer-implemented method (and associated computing system and computer program product) comprises distributing computing workloads among a plurality of processing units of a computing system. The method further comprises, responsive to detecting a reduced power supply capacity of the computing system: determining, based on historical workload information, a first workload of the computing workloads to prioritize the completion thereof, and prioritizing power delivery to a first processing unit of the plurality of processing units to which the first workload is distributed. The method further comprises powering down the first processing unit responsive to completion of the first workload.

First claim

What is claimed is:

1. A computer-implemented method comprising:
distributing computing workloads among a plurality of processing units of a computing system;
responsive to detecting a reduced power supply capacity of the computing system:
determining, based on historical workload information reflecting one or more previously completed workloads, a first workload of the computing workloads to prioritize the completion thereof; and
prioritizing power delivery to a first processing unit of the plurality of processing units to which the first workload is distributed; and

powering down the first processing unit responsive to completion of the first workload.

2. The method of claim 1 , further comprising:
responsive to detecting the reduced power supply capacity:
delivering a reduced amount of power to each of the plurality of processing units while determining the first workload,

wherein prioritizing power delivery to the first processing unit comprises increasing the power delivered from the reduced amount.

3. The method of claim 1 , wherein determining the first workload comprises:
determining that the first workload is expected to complete within a predetermined amount of time.

4. The method of claim 3 , wherein the first workload comprises a first function, the method further comprising:
determining the historical workload information, wherein determining the historical workload information comprises:
storing a start time and a stop time for one or more instances of previously completing the first function reflected in the one or more previously completed workloads; and
determining, based on the start time and the stop time, an average runtime of the first function,

wherein determining that the first workload is expected to complete within a predetermined amount of time is based on the average runtime.

5. The method of claim 1 , further comprising:
determining the historical workload information, wherein determining the historical workload information comprises:
storing distribution information indicating a distribution of the one or more previously completed workloads among the plurality of processing units,

wherein determining the first workload comprises:
identifying, based on the distribution information, which processing unit of the plurality of processing units is least likely to have a workload distributed to it after the reduced power supply capacity is detected, and
selecting the first workload from the identified processing unit.

6. The method of claim 1 , further comprising:
increasing the power delivered to at least a second processing unit of the plurality of processing units responsive to completion of the first workload.

7. The method of claim 1 , wherein the computing system comprises a plurality of power supplies, and
wherein detecting a reduced power supply capacity of the computing system comprises detecting a failure of at least one power supply of the plurality of power supplies.

8. A computing system comprising:
a plurality of power supplies; and
a plurality of processing units, wherein a first set comprising a first processing unit of the plurality of processing units is configured to:
distribute computing workloads among a second set of the plurality of processing units;
responsive to detecting a reduced power supply capacity of the plurality of power supplies:
determine, based on historical workload information reflecting one or more previously completed workloads, a first workload of the computing workloads to prioritize the completion thereof; and
prioritize power delivery to a selected processing unit of the second set to which the first workload is distributed; and

power down the selected processing unit responsive to completion of the first workload.

9. The computing system of claim 8 , wherein the first processing unit comprises a central processing unit, and
wherein the second set comprises a plurality of graphics processing units.

10. The computing system of claim 8 , wherein the first set is further configured to:
responsive to detecting the reduced power supply capacity:
deliver a reduced amount of power to each processing unit of the second set while determining the first workload,

wherein prioritizing power delivery to the selected processing unit of the second set comprises increasing the power delivered from the reduced amount.

11. The computing system of claim 8 , wherein determining the first workload comprises:
determining that the first workload is expected to complete within a predetermined amount of time.

12. The computing system of claim 11 , wherein the first workload comprises a first function, wherein the first set is further configured to:
determine the historical workload information, wherein determining the historical workload information comprises:
storing a start time and a stop time for one or more instances of previously completing the first function reflected in the one or more previously completed workloads; and
determining, based on the start time and the stop time, an average runtime of the first function,

wherein determining that the first workload is expected to complete within a predetermined amount of time is based on the average runtime.

13. The computing system of claim 8 , wherein the first set is further configured to:
determine the historical workload information, wherein determining the historical workload information comprises:
storing distribution information indicating a distribution of the one or more previously completed workloads among the second set,

wherein determining the first workload comprises:
identifying, based on the distribution information, which processing unit of the plurality of processing units is least likely to have a workload distributed to it after the reduced power supply capacity is detected, and
selecting the first workload from the identified processing unit.

14. The computing system of claim 8 , wherein detecting a reduced power supply capacity comprises detecting a failure of at least one power supply of the plurality of power supplies.

15. A computer program product for distributing power among a plurality of graphics processing units (GPUs) of a computing system, the computer program product comprising:
a computer-readable storage medium having computer-readable program code embodied therewith, the computer-readable program code executable by one or more computer processors to perform an operation comprising:
distributing computing workloads among the plurality of GPUs;
responsive to detecting a reduced power supply capacity of the computing system:
determining, based on historical workload information reflecting one or more previously completed workloads, a first workload of the computing workloads to prioritize the completion thereof; and
prioritizing power delivery to a first GPU of the plurality of GPUs to which the first workload is distributed; and

powering down the first GPU responsive to completion of the first workload.

16. The computer program product of claim 15 , the operation further comprising:
responsive to detecting the reduced power supply capacity:
delivering a reduced amount of power to each of the plurality of GPUs while determining the first workload,

wherein prioritizing power delivery to the first GPU comprises increasing the power delivered from the reduced amount.

17. The computer program product of claim 15 , wherein determining the first workload comprises:
determining that the first workload is expected to complete within a predetermined amount of tim

Assignees

• IBM

Inventors

• King Justin K
• Harrison Abby
• Olandt Jakob
• Duffy Brittany

Classifications

• G06F1/3228 Primary

  Monitoring task completion, e.g. by use of idle timers, stop commands or wait commands · CPC title

• G06F1/329

  by task scheduling · CPC title

• G06F1/3287 Primary

  by switching off individual functional units in the computer system · CPC title

• G06F9/5094

  where the allocation takes into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title

• Y02D10/00

  Energy efficient computing, e.g. low power processors, power management or thermal management · CPC title