Generating synthetic workloads to measure power utilization in a computer system

1. A method for generating a synthetic workload to test power utilization in a computer system, comprising:

monitoring power utilization of a reference computer system while the reference computer system executes a workload-of interest, wherein the monitoring process produces a power profile;

determining characteristics of the workload-of-interest from the power profile; and

using the determined characteristics to construct the synthetic workload, wherein, when the synthetic workload is executed on a target computer system, the synthetic workload has a power profile that closely matches the power profile of the reference computer system while the reference computer executes the workload-of-interest.

2. The method of claim 1 ,

wherein determining characteristics of the workload-of-interest involves performing a spectral decomposition on the power profile to produce a set of coefficients; and

wherein using the determined characteristics to construct the synthetic workload involves using the set of coefficients to construct the synthetic workload.

3. The method of claim 2 ,

wherein performing the spectral decomposition involves using the Fourier technique; and

wherein producing the set of coefficients involves producing coefficients for M largest Fourier harmonics produced by the Fourier technique.

4. The method of claim 3 , wherein using the set of coefficients to construct the synthetic workload involves constructing a workload comprised of M sinusoidal workload ads corresponding to the M largest Fourier harmonics.

5. The method of claim 1 , wherein monitoring the power utilization of the reference computer system involves:

periodically polling current sensors and associated voltage sensors within the reference computer system to generate dynamic traces of currents and associated voltages within the reference computer system; and

generating a dynamic trace of total power consumption of the reference computer system based on the dynamic traces of the currents and associated voltages.

6. The method of claim 1 , wherein monitoring the power utilization of the reference computer system involves using a power meter interposed between the reference computer system and a power source to monitor the power utilization.

7. The method of claim 1 , wherein monitoring the power utilization of the reference computer system involves:

monitoring instrumentation signals within the reference computer system, wherein the instrumentation signals do not include corresponding current and voltage signals that can be used to directly compute power consumption; and

estimating the power consumption for the computer system by inferring the power consumption from the instrumentation signals and from an inferential power model generated during a training phase.

8. The method of claim 1 , further comprising estimating the power utilization of a target computer system by:

executing the synthetic workload on the target computer system; and

monitoring the power utilization of the target computer system while the synthetic workload is executing.

9. The method of claim 1 ,

wherein monitoring the power utilization of the reference computer system additionally involves monitoring a relative mix between floating-point and integer operations as the reference computer system executes the workload-of-interest; and

wherein constructing the synthetic workload involves using the relative mix information to synthesize separate floating-point and integer components of the synthetic workload.

10. A computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for generating a synthetic workload to test power utilization in a computer system, the method comprising:

monitoring power utilization of a reference computer system while the reference computer system executes a workload-of interest, wherein the monitoring process produces a power profile;

determining characteristics of the workload-of-interest from the power profile; and

using the determined characteristics to construct the synthetic workload, wherein, when the synthetic workload is executed on a target computer system, the synthetic workload has a power profile that closely matches the power profile of the reference computer system while the reference computer system executes the workload-of-interest.

11. The computer-readable storage medium of claim 10 ,

wherein determining characteristics of the workload-of-interest involves performing a spectral decomposition on the power profile to produce a set of coefficients; and

wherein using the determined characteristics to construct the synthetic workload involves using the set of coefficients to construct the synthetic workload.

12. The computer-readable storage medium of claim 11 ,

wherein performing the spectral decomposition involves using the Fourier technique; and

wherein producing the set of coefficients involves producing coefficients for M largest Fourier harmonics produced by the Fourier technique.

13. The computer-readable storage medium of claim 12 , wherein using the set of coefficients to construct the synthetic workload involves constructing a workload comprised of M sinusoidal workloads corresponding to the M largest Fourier harmonics.

14. The computer-readable storage medium of claim 10 , wherein monitoring the power utilization of the reference computer system involves:

periodically polling current sensors and associated voltage sensors within the reference computer system to generate dynamic traces of currents and associated voltages within the reference computer system; and

generating a dynamic trace of total power consumption of the reference computer system based on the dynamic traces of the currents and associated voltages.

15. The computer-readable storage medium of claim 10 , wherein monitoring the power utilization of the reference computer system involves using a power meter interposed between the reference computer system and a power source to monitor the power utilization.

16. The computer-readable storage medium of claim 10 , wherein monitoring the power utilization of the reference computer system involves:

monitoring instrumentation signals within the reference computer system, wherein the instrumentation signals do not include corresponding current and voltage signals that can be used to directly compute power consumption; and

estimating the power consumption for the computer system by inferring the power consumption from the instrumentation signals and from an inferential power model generated during a training phase.

17. The computer-readable storage medium of claim 10 , further comprising estimating the power utilization of a target computer system by:

executing the synthetic workload on the target computer system; and

monitoring the power utilization of the target computer system while the synthetic workload is executing.

18. The computer-readable storage medium of claim 10 ,

wherein monitoring the power utilization of the reference computer system additionally involves monitoring a relative mix between floating-point and integer operations as the reference computer system executes the workload-of-interest; and

wherein constructing the synthetic workload involves using the relative mix information to synthesize separate floating-point and integer components of the synthetic workload.

19. An apparatus that generates a synthetic workload to test power utilization in a computer system, comprising:

a monitoring mechanism configured to monitor power utilization of a reference computer system while the reference computer system executes a workload-of interest, wherein the monitoring process produces a power profile;

an analysis mechanism configured to determine characteristics of the workload-of-interest from the power profile; and

a work-load-generation mechanism configured to use the determined characteristics to construct the synthetic workload, wherein, when the synthetic workload is executed on a target computer system, the synthetic workload has a power profile that closely matches the power profile of the reference computer system while the reference computer system executes the workload-of-interest.

20. The apparatus of claim 19 ,

wherein the analysis mechanism is configured to perform a spectral decomposition on the power profile to produce a set of coefficients; and

wherein the work-load-generation mechanism is configured to use the set of coefficients to construct the synthetic workload.

21. The apparatus of claim 19 ,

wherein the analysis mechanism is configured to use the Fourier technique to perform the spectral decomposition; and

wherein the analysis mechanism is configured to produce coefficients for M largest Fourier harmonics produced by the Fourier technique.

22. The apparatus of claim 21 , wherein the work-load-generation mechanism is configured to construct a workload comprised of M sinusoidal workloads corresponding to the M largest Fourier harmonics.