Cluster computational capacity level switching based on demand prediction and stability constraint and power consumption management

• US 8,788,855 B2
• Filed: 09/23/2011
• Issued: 07/22/2014
• Est. Priority Date: 09/23/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Cong, Lanlan, Zhu, Renqi, Zhang, Yang
• Primary Examiner(s): KIM, KENNETH S
• Application Number:
  US13/241,614
• Publication Number:
  US 20130080814A1
• Time in Patent Office: 1,033 Days
• Field of Search: None
• US Class Current: 713/320
• CPC Class Codes: G06F 1/3206 Monitoring of events, devices or parameters that trigger a change in power modality
• CPC Class Codes: G06F 1/3287 by switching off individual functional units in the computer system
• CPC Class Codes: G06F 1/329 by task scheduling
• CPC Class Codes: G06F 9/5094 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 Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Power consumption and computational availability of a cluster are automatically managed by monitoring consumption, predicting demand based on a job queue and/or historic data, and checking certain conditions and constraints. Threshold conditions for switching the cluster'"'"'s capacity level are based on the current level and the predicted demand. Switching levels is avoided when a stability constraint would be violated. Cluster capacity levels can be uniformly or nonuniformly spaced, and may have decreasing gaps between levels as capacity decreases. Capacity level definitions and level switching conditions and constraints may be defaults and/or be provided by an administrator.

Claims

1. A computer-readable storage medium configured with data and with instructions that when executed by at least one processor causes the processor(s) to perform a process for managing computational availability of a computational cluster which is operating at a current cluster computational capacity level, the process comprising the steps of:

predicting demand for computational availability of the computational cluster based at least on a job queue length, the job queue length being the number of jobs in a job queue, each job comprising code which is subject to scheduling;

determining that a threshold condition for switching the computational cluster to a different cluster computational capacity level is satisfied, the threshold condition based at least on the current cluster computational capacity level and the predicted demand;

ascertaining whether switching the computational cluster to the different cluster computational capacity level will violate a stability constraint of the computational cluster, the stability constraint based at least on a history of cluster computational capacity level switches within a period of time which is not earlier than a specified history cutoff; and

performing at least one of the following;

(i) switching the computational cluster to the different cluster computational capacity level after determining that the threshold condition for switching is satisfied and ascertaining that switching the computational cluster to the different cluster computational capacity level will not violate the stability constraint, (ii) recommending to a user that the computational cluster be switched to the different cluster computational capacity level, (iii) recommending to a user that the computational cluster not be switched to the different cluster computational capacity level.

2. The configured medium of claim 1, wherein the process further comprises monitoring electrical power consumption of the computational cluster at sample points.

3. The configured medium of claim 1, wherein the process further comprises monitoring electrical power consumption of the computational cluster for a time period of at least ten minutes.

4. The configured medium of claim 1, wherein the predicting step predicts demand for computational availability of the computational cluster based on job queue length and on at least one of the following:

job priority, job wait time.

5. The configured medium of claim 1, wherein the predicting step predicts demand for computational availability of the computational cluster based on job queue length and on historic power consumption data.

6. The configured medium of claim 1, wherein the process further includes switching the computational cluster to one of at least two other cluster computational capacity levels (at least three levels being present), and the cluster computational capacity levels are uniformly spaced.

7. The configured medium of claim 1, wherein the process further includes switching the computational cluster to one of at least two other cluster computational capacity levels (at least three levels being present), and the cluster computational capacity levels are nonuniformly spaced, with decreasing gaps between levels as capacity decreases.

8. A computer-readable storage medium configured with data and with instructions that when executed by at least one processor causes the processor(s) to perform a process for managing computational availability, the process comprising the steps of:

obtaining at least three cluster computational capacity level definitions, the defined cluster computational capacity levels including a nonzero minimum computational capacity level;

predicting demand for computational availability of the computational cluster based on at least one of the following;

(i) information from a job queue containing jobs which comprise code that is subject to scheduling, (ii) historic power consumption data;

determining that a threshold condition for switching the computational cluster from a current cluster computational capacity level to a different cluster computational capacity level is satisfied, the threshold condition based at least on the predicted demand and on an indication of current cluster power consumption;

ascertaining that switching the computational cluster to the different cluster computational capacity level will not violate a user-configurable stability constraint of the computational cluster, the user-configurable stability constraint based at least on a history of cluster computational capacity level switches within a user-specified period of time; and

recommending to a user whether or not the computational cluster should be switched to the different cluster computational capacity level.

9. The configured medium of claim 8, wherein the process further comprises at least one of the following:

monitoring electrical power consumption of the computational cluster for at least four sample points, and identifying the current cluster computational capacity level based on results of the monitoring;

monitoring electrical power consumption of the computational cluster for a time period of at least thirty minutes, and identifying the current cluster computational capacity level based on results of the monitoring.

10. The configured medium of claim 8, wherein the predicting step predicts demand for computational availability of the computational cluster based at least on job priority information from a job queue.

11. The configured medium of claim 8, wherein the obtaining step comprises at least one of the following listed steps:

receiving a cluster computational capacity level definition from a user;

using a previously specified default cluster computational capacity level definition;

getting a cluster computational capacity level definition which specifies a percentage of total cluster capacity;

getting a cluster computational capacity level definition which specifies capacity in absolute terms that include a number of storage bytes;

getting a cluster computational capacity level definition which specifies capacity in absolute terms that include a number of processors.

12. The configured medium of claim 11, wherein the process comprises at least two of the listed steps.

13. The configured medium of claim 8, wherein the obtaining level obtains definitions of at least three cluster computational capacity levels which are nonuniformly spaced.

14. A computer system comprising:

logical processor(s);

a memory in operable communication with the logical processor(s);

at least three cluster computational capacity level definitions residing in the memory;

a demand predictor residing in the memory and including demand prediction code which when executed by the processor(s) will calculate a predicted demand for computational availability of the computational cluster, based on information from a job queue containing jobs which comprise code that is subject to scheduling;

a threshold condition for switching the computational cluster from a current one of the defined cluster computational capacity levels to a different one of the defined cluster computational capacity levels, the threshold condition residing in the memory and based on at least the predicted demand, and wherein the threshold condition comprises a stability constraint that is based on at least one of the following;

a count of switches between cluster computational capacity levels, a frequency of switches between cluster computational capacity levels; and

code which upon execution by a processor performs at least one of the following;

(i) switches the computational cluster to the different cluster computational capacity level after determining that the threshold condition for switching is satisfied and ascertaining that switching the computational cluster to the different cluster computational capacity level will not violate the stability constraint, (ii) recommends to a user that the computational cluster be switched to the different cluster computational capacity level, (iii) recommends to a user that the computational cluster not be switched to the different cluster computational capacity level.

15. The system of claim 14, comprising definitions of at least four consecutive cluster computational capacity levels which conform with at least one of the following:

a uniform spacing between levels, a geometric sequence spacing between levels, a logarithmic spacing between levels.

16. The system of claim 14, wherein the threshold condition is a condition for switching to a lower cluster computational capacity level, and the threshold condition comprises a condition that is met when a sampled power consumption level does not exceed a specified threshold for at least a specified number of consecutive samples and the number of jobs in a job queue does not exceed a specified number.

17. The system of claim 14, wherein the threshold condition is a condition for switching to a higher cluster computational capacity level, and the threshold condition comprises a condition that is met when the number of jobs in a job queue is at least a specified number.

18. The system of claim 14, wherein the threshold condition is a condition for switching to a lower cluster computational capacity level, and the threshold condition comprises a condition that is met when power consumption over a measured time period does not exceed a specified threshold and the number of jobs in a job queue does not exceed a specified number.

19. The system of claim 14, wherein the threshold condition is based at least in part on the number of jobs in at least one job queue.

20. The system of claim 14, wherein the demand predictor calculation is based on job queue length, job priority, and job wait time.