System and method to enable processor management policy in a multi-processor environment

• US 20060036878A1
• Filed: 08/11/2004
• Published: 02/16/2006
• Est. Priority Date: 08/11/2004
• Status: Active Grant

• Current Assignee: Tahoe Research Limited (f/k/a Learndale Limited) (Vector Capital Corporation)
• Original Assignee: Intel Corporation
• Inventors: Zimmer, Vincent J., Rothman, Michael A.
• Granted Patent: US 7,739,527 B2
• Time in Patent Office: Days
• US Class Current: 713/300
• CPC Class Codes: G06F 1/3203 Power management, i.e. event-based initiation of a power-saving mode
• 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: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

In some embodiments, the invention involves off-loading processor workloads to reduce power requirements of a multi-processor system. In one embodiment, a multi-processor system redistributes computing among the multiple processors and changes the state of one or more processors to sleep mode. In another embodiment, a multi-processor system throttles the processor speed of under-utilized processors to reduce power consumption. Other embodiments are described and claimed.

Claims

1. A system for reducing power consumption in a multi-processor environment, comprising:

a plurality of processors being interchangeable with respect to execution of enterprise tasks, wherein at least one processor is capable of sleep mode, and wherein a processor in sleep mode uses less power than a processor not in sleep mode;

a workload profiler to monitor workloads on the respective plurality of processors, and to determine whether each workload exceeds a predetermined threshold, the workloads comprising a plurality of enterprise tasks, and the workload profiler to redistribute enterprise tasks among the plurality of processors by initiating a power event.

2. The system as recited in claim 1, wherein the workload profiler is configured to initiate sleep mode in one or more processors in conjunction with the redistribution of enterprise tasks among the plurality of processors.

3. The system as recited in claim 1, wherein the workload profiler is configured to initiate wake mode in one or more sleeping processors in conjunction with the redistribution of enterprise tasks among the plurality of processors.

4. The system as recited in claim 1, wherein the workload profiler is configured to initiate a throttling of processor speed in one or more processors in conjunction with the redistribution of enterprise tasks among the plurality of processors.

5. The system as recited in claim 1, wherein the workload profiler is configured to select at least one of initiating sleep mode and initiating a throttling of processor speed in one or more processors, the selection being in conjunction with the redistribution of enterprise tasks among the plurality of processors.

6. The system as recited in claim 1, wherein the workload profiler is bundled with a virtual machine manager (VMM) in a virtualization environment.

7. The system as recited in claim 6, wherein each processor operates in a separate virtual machine (VM).

8. The system as recited in claim 1, wherein the workload profiler is an out of band process.

9. The system as recited in claim 1, wherein the workload profiler comprises an operating system (OS) agent.

10. The system as recited in claim 1, wherein the predetermined threshold is processor-specific.

11. A method for reducing power consumption in a multi-processor environment, comprising:

periodically probing a plurality of processors, by a workload profiler, to determine a workload for each processor, the periodicity of the probing being user selectable;

determining whether a user selectable first threshold exceeds one or more workloads;

if the user selectable first threshold exceeds one or more workloads, then redistributing workloads among the plurality of processors, and initiating a sleep state in one or more idle processors; and

if the one or more workloads exceeds the user selectable first threshold, then waking one or more processors from a sleep state, and redistributing workloads among the plurality of processors.

12. The method as recited in claim 11, further comprising:

determining whether a user selectable second threshold has been exceeded and throttling processor speed for one or more processors based on the user selectable first and second threshold and determined workloads.

13. The method as recited in claim 12, wherein the first threshold is different from the second threshold.

14. The method as recited in claim 12, wherein the first and second thresholds are processor-specific.

15. The method as recited in claim 11, wherein the workload profiler is bundled with a virtual machine manager (VMM), the VMM controlling the plurality of processors in a virtualization environment.

16. The method as recited in claim 15, wherein each processor operates in a separate virtual machine (VM).

17. The method as recited in claim 11, wherein the workload profiler is an out of band process.

18. The method as recited in claim 11, wherein the workload profiler comprises an operating system (OS) agent.

19. The method as recited in claim 11, wherein redistributing workloads further comprises:

initiating an interprocessor interrupt;

transferring workload from a first processor to a second processor;

notifying the plurality of processors that the first processor has gone off-line.

20. A machine accessible medium containing code for managing power consumption in a multiprocessor platform, that when executed cause a machine to:

periodically probe a plurality of processors, by a workload profiler, to determine a workload for each processor, the periodicity of the probing being user selectable;

determine whether a user selectable first threshold exceeds one or more workloads;

if the user selectable first threshold exceeds one or more workloads, then redistribute workloads among the plurality of processors, and initiate a sleep state in one or more idle processors; and

if the one or more workloads exceeds the user selectable first threshold, then wake one or more processors from a sleep state, and redistribute workloads among the plurality of processors.

21. The machine accessible medium as recited in claim 20, further comprising code that when executed cause a machine to:

determine whether a user selectable second threshold has been exceeded and throttling processor speed for one or more processors based on the user selectable first and second threshold and determined workloads.

22. The machine accessible medium as recited in claim 21, wherein the first threshold is different from the second threshold.

23. The machine accessible medium as recited in claim 21, wherein the first and second thresholds are processor-specific.

24. The machine accessible medium as recited in claim 20, wherein the workload profiler is bundled with a virtual machine manager (VMM), the VMM controlling the plurality of processors in a virtualization environment.

25. The machine accessible medium as recited in claim 24, wherein each processor operates in a separate virtual machine (VM).

26. The machine accessible medium as recited in claim 20, wherein the workload profiler is an out of band process.

27. The machine accessible medium as recited in claim 20, wherein the workload profiler is an operating system (OS) agent.

28. The machine accessible as recited in claim 20, wherein redistributing workloads further code that causes the machine to:

initiate a interprocessor interrupt;

transfer workload from a first processor to a second processor;

notify the plurality of processors that the first processor has gone off-line.

29. A method for reducing power consumption in a multi-processor environment, comprising:

periodically probing a plurality of processors, by a workload profiler, to determine a workload for each processor, the periodicity of the probing being user selectable;

determining whether a power event should be performed based on the determined workloads and a set of user selectable profiler thresholds;

if a power event is determined to be performed, then redistributing workloads among the plurality of processors and performing the power event, wherein the power event is at least one of the events selected from the group of initiating sleep mode in a processor, waking a processor, and throttling a processor.

30. The method as recited in claim 29, wherein the workload profiler is bundled with a virtual machine manager (VMM), the VMM controlling the plurality of processors in a virtualization environment.