Dynamic performance based cooling control for cluster processing devices

Patent metadata

Field	Value
Publication number	US-9335751-B1
Application number	US-201314012537-A
Country	US
Kind code	B1
Filing date	Aug 28, 2013
Priority date	Aug 28, 2013
Publication date	May 10, 2016
Grant date	May 10, 2016

Abstract

Methods, systems, and apparatus for cooling control in a datacenter. In one aspect, a method includes, for each processing device in a cluster of processing devices configured to perform a distributed task, wherein each processing device is thermally controlled by a cooling system that controls cooling to each processing device on an individual basis, determining whether the processing device is operating within a performance target for the cluster of processing devices; for each processing device determined to not be operating within the performance target for the cluster, generating a respective control signal to adjust the cooling delivery to the processing device to cause the performance of the processing device to be within the performance target for the cluster of processing devices; and for each processing device determined to be operating within the performance target for the cluster, maintaining the cooling delivery to the processing device.

First claim

What is claimed is:

1. A computer-implemented method, comprising:
for each processing device in a set of two or more processing devices configured to perform a distributed task, wherein each processing device is thermally controlled by a cooling system that controls cooling to each processing device on an individual basis, determining, by a cooling system control device that controls the cooling system, whether the processing device is operating within a performance target for the set of two or more processing devices, the determining comprising:
iteratively determining a load measure for the processing device, the load measure being indicative of a load on the processing device;
for each iteration for which the load measures meet a threshold load measure, determining an operating frequency measure for the processing device; and
determining whether the processing device is operating within a performance target for the set of two or more processing devices based one or more frequency measures of the processing device, wherein the performance target is a frequency range of a target operating frequency that is based on frequency measures obtained from the set of two or more processing devices operating at a common baseline cooling level;

for each processing device determined to not be operating within the performance target for the set of two or more processing devices, generating, by the cooling system control device, a respective control signal to adjust the cooling delivery to the processing device to cause the performance of the processing device to be within the performance target for the set of two or more processing devices; and
for each processing device determined to be operating within the performance target for the set of two or more processing devices, maintaining the cooling delivery.

2. The computer-implemented method of claim 1 , wherein:
determining whether the operating frequency measure of the processing device is within a frequency range of the target operating frequency comprises determining whether the operating frequency measure of the processing device is no less than a minimum percentage less than the target operating frequency; and
generating, by the cooling system control device, a respective control signal to adjust the cooling delivery to the processing device comprises generating a control signal to increase cooling delivery to the processing device.

3. The computer-implemented method of claim 1 , wherein:
determining whether the operating frequency measure of the processing device is within a frequency range of the target operating frequency comprises determining whether the operating frequency measure of the processing device is no greater than a maximum percentage greater than the target operating frequency; and
generating, by the cooling system control device, a respective control signal to adjust the cooling delivery to the processing device comprises generating a control signal to decrease cooling delivery to the processing device.

4. The computer-implemented method of claim 1 , wherein:
determining whether the operating frequency measure of the processing device is within a frequency range of the target operating frequency comprises determining whether the operating frequency measure of the processing device is greater than a minimum percentage less than the target operating frequency and less than a maximum percentage greater than the target operating frequency; and
generating, by the cooling system control device, a respective control signal to adjust the cooling delivery to the processing device comprises generating a first control signal to decrease cooling delivery to the processing device when the operating frequency is greater than a maximum percentage greater than the target operating frequency and generating a control signal to increase cooling delivery to the processing device when the operating frequency is less than a minimum percentage less than the target operating frequency.

5. The computer-implemented method of claim 1 , further comprising determining the target operating frequency from the operating frequencies of the processing devices of the set of two or more processing devices.

6. The computer-implemented method of claim 5 , wherein determining the target operating frequency from the operating frequencies of the processing devices of the set of two or more processing devices comprises determining a central tendency of the operating frequencies.

7. The computer-implemented method of claim 6 , wherein determining the target operating frequency from the operating frequencies of the processing devices of the set of two or more processing devices comprises determining each of the operating frequencies from frequency measures obtained from processing devices operating at a common baseline cooling level.

8. The computer-implemented method of claim 1 , wherein determining a load measure of a processing device comprises determining a power consumption level of the processing device, and wherein the threshold load measure is a threshold power consumption level.

9. A computer-implemented method, comprising:
for each processing device in a set of two or more processing devices configured to perform a distributed task, wherein each processing device is thermally controlled by a cooling system that controls cooling to disjoint sets processing devices on a disjoint set basis, determining, by a cooling system control device that controls the cooling system, whether the processing device is operating within a performance target for the set of two or more processing devices, the determining comprising:
iteratively determining a load measure for the processing device, the load measure being indicative of a load on the processing device;
for each iteration for which the load measures meet a threshold load measure, determining an operating frequency measure for the processing device; and
determining whether the processing device is operating within a performance target for the set of two or more processing devices based one or more frequency measures of the processing device, wherein the performance target is a frequency range of a target operating frequency that is based on frequency measures obtained from the set of two or more processing devices operating at a common baseline cooling level;

for each disjoint set of processing devices that includes a first processing device determined to not be operating within the performance target and operating below the performance target for the set of two or more processing devices, generating, by the cooling system control device, a respective control signal to increase the cooling delivery to the disjoint set of processing devices to cause the performance of the first processing device to increase and be within the performance target for the set of two or more processing devices; and
for each disjoint set of processing devices in which each processing device is determined to be operating within the performance target for the set of two or more processing devices, maintaining the cooling delivery.

10. The computer-implemented method of claim 9 , further comprising:
for each disjoint set of processing devices that includes a second processing device determined to not be operating within the performance target and above the performance target for the set of two or more processing devices set of two or more processing devices, maintaining the cooling delivery to the disjoint set of processing devices.

11. The computer-implemented method of claim 9 , further comprising:
for each disjoint set of processing devices that includes a second processing device determined to not be operating within the performance target

Assignees

• Google Inc

Inventors

• Farshchian Soheil
• Rice Jeremy

Classifications

• G06F1/206 Primary

  comprising thermal management · CPC title

• Y02B60/1275

  Cross-Sectional Technologies · mapped topic

• G06F11/3058

  Monitoring arrangements for monitoring environmental properties or parameters of the computing system or of the computing system component, e.g. monitoring of power, currents, temperature, humidity, position, vibrations (thermal management in cooling arrangements of a computing system G06F1/206) · CPC title

• G05B15/02 Primary

  electric · CPC title

• Y02D10/00

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