ACCELERATION BENEFIT ESTIMATOR

• US 20140297405A1
• Filed: 03/26/2013
• Published: 10/02/2014
• Est. Priority Date: 03/26/2013
• Status: Active Grant

• Current Assignee: Empire Technology Development LLC (Allied Inventors Management, LLC)
• Original Assignee: Empire Technology Development LLC (Allied Inventors Management, LLC)
• Inventors: FINE, KEVIN
• Granted Patent: US 9,747,185 B2
• Time in Patent Office: Days
• US Class Current: 705/14.53
• CPC Class Codes: G06F 11/3003 Monitoring arrangements specially adapted to the computing system or computing system component being monitored
• CPC Class Codes: G06F 11/3062 where the monitored property is the power consumption power management in a computing system G06F1/3203
• CPC Class Codes: G06F 11/3409 for performance assessment
• CPC Class Codes: G06F 11/3419 by assessing time
• CPC Class Codes: G06F 11/3428 Benchmarking
• CPC Class Codes: G06F 2201/865 Monitoring of software
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Technologies related to acceleration benefit estimation are generally described. In some examples, data centers may identify applications that may benefit from Programmable Hardware Accelerators (PHAs), and test the identified applications by running accelerated versions thereof, i.e., versions that use one or more identified PHAs, and comparing performance of the accelerated versions to reference versions, i.e., versions that do not use the one or more identified PHAs. Data centers may report comparison results may be reported to data center customers to encourage customer adoption of PHAs.

Claims

1. A method for estimating programmable hardware acceleration benefit, the method comprising:

executing, by one or more computing devices in a data center, a reference version of an application to perform one or more tasks for a data center customer;

measuring one or more of an execution time, an execution cost, or an execution energy consumption for executing the reference version of the application;

identifying one or more programmable hardware accelerators for the application, wherein the identified programmable hardware accelerators are not being utilized during execution of the reference version of the application;

automatically executing, by the one or more computing devices, an accelerated version of the application to perform the one or more tasks performed by the reference version of the application, wherein the accelerated version of the application is accelerated by use of the one or more identified programmable hardware accelerators;

measuring one or more of an execution time, an execution cost, or an execution energy consumption for executing the accelerated version of the application;

calculating one or more of a difference in execution time, a difference in execution cost, or a difference in execution energy consumption between the reference version of the application and the accelerated version of the application; and

providing one or more of the difference in execution time, the difference in execution cost, or the difference in execution energy consumption to the data center customer.

2. The method of claim 1, further comprising determining whether programmable hardware resources in the data center are available for use by the data center customer.

3. The method of claim 1, wherein the identifying one or more programmable hardware accelerators for the application comprises analyzing code in the application for a segment that is at least partially replaceable by a programmable hardware accelerator.

4. The method of claim 3, further comprising retrieving a programmable hardware accelerator from a library of programmable hardware accelerators to at least partially replace the segment of the application.

5. The method of claim 3, further comprising generating a programmable hardware accelerator to at least partially replace the segment of the application.

6. The method of claim 3, wherein the analyzing code in the application comprises a flow extraction process.

7. The method of claim 3, wherein the analyzing code in the application comprises hash comparison process.

8. The method of claim 1, further comprising comparing outputs generated by the reference version of an application and the accelerated version of the application to determine whether the compared outputs are substantially identical.

9. The method of claim 1, wherein the automatically executing the accelerated version of the application comprises forking the reference version of the application to run accelerated segments of the application in parallel with un-accelerated segments executed by the reference version of the application.

10. The method of claim 1, wherein the automatically executing the accelerated version of the application is performed simultaneously with executing the reference version of the application.

11. The method of claim 1, wherein the measuring the reference version of the application, the identifying programmable hardware accelerators for the application, the measuring the accelerated version of the application, the calculating differences between the reference version of the application and the accelerated version of the application, and the providing the differences to the data center customer are performed automatically.

12. The method of claim 1, further comprising aggregating differences between the reference version of the application and the accelerated version of the application corresponding to multiple instances of executing the reference version of the application and the accelerated version of the application, and wherein the providing differences to the data center customer comprises providing a report summarizing differences over the multiple instances.

13. The method of claim 1, further comprising checking a device execution history of the one or more computing devices to verify use of the one or more identified programmable hardware accelerators during execution of the accelerated version of the application.

14. A non-transitory computer readable storage medium having computer executable instructions executable by a processor, the instructions that, when executed by the processor, implement a programmable hardware acceleration benefit estimator, which causes the processor to:

measure one or more of an execution time, an execution cost, or an execution energy consumption for executing a reference version of an application to perform one or more tasks for a data center customer;

identify one or more programmable hardware accelerators for the application, wherein the identified programmable hardware accelerators are not being utilized during execution of the reference version of the application;

automatically execute an accelerated version of the application to perform the one or more tasks performed by the reference version of the application, wherein the accelerated version of the application is accelerated by use of the one or more identified programmable hardware accelerators;

measure one or more of an execution time, an execution cost, or an execution energy consumption for executing the accelerated version of the application;

calculate one or more of a difference in execution time, a difference in execution cost, or a difference in execution energy consumption between the reference version of the application and the accelerated version of the application; and

provide one or more of the difference in execution time, the difference in execution cost, or the difference in execution energy consumption to the data center customer.

15. The non-transitory computer readable storage medium of claim 14, further comprising instructions that cause the processor to determine whether programmable hardware resources in the data center are available for use by the data center customer.

16. The non-transitory computer readable storage medium of claim 14, wherein the identifying one or more programmable hardware accelerators for the application comprises analyzing code in the application for a segment that is at least partially replaceable by a programmable hardware accelerator.

17. The non-transitory computer readable storage medium of claim 16, further comprising instructions that cause the processor to retrieve a programmable hardware accelerator from a library of programmable hardware accelerators to at least partially replace the segment of the application.

18. The non-transitory computer readable storage medium of claim 16, further comprising instructions that cause the processor to generate a programmable hardware accelerator to at least partially replace the segment of the application.

19. The non-transitory computer readable storage medium of claim 16, wherein the analyzing code in the application comprises a flow extraction process.

20. The non-transitory computer readable storage medium of claim 16, wherein the analyzing code in the application comprises hash comparison process.

21. The non-transitory computer readable storage medium of claim 14, further comprising instructions that cause the processor to compare outputs generated by the reference version of an application and the accelerated version of the application to determine whether the compared outputs are substantially identical.

22. The non-transitory computer readable storage medium of claim 14, wherein the automatically executing the accelerated version of the application comprises forking the reference version of the application to run accelerated segments of the application in parallel with un-accelerated segments executed by the reference version of the application.

23. The non-transitory computer readable storage medium of claim 14, wherein the automatically executing the accelerated version of the application is performed simultaneously with executing the reference version of the application.

24. The non-transitory computer readable storage medium of claim 14, wherein the measuring the reference version of the application, the identifying programmable hardware accelerators for the application, the measuring the accelerated version of the application, the calculating differences between the reference version of the application and the accelerated version of the application, and the providing the differences to the data center customer are performed automatically.

25. The non-transitory computer readable storage medium of claim 14, further comprising instructions that cause the processor to aggregate differences between the reference version of the application and the accelerated version of the application corresponding to multiple instances of executing the reference version of the application and the accelerated version of the application, and wherein the providing differences to the data center customer comprises providing a report summarizing differences over the multiple instances.

26. The non-transitory computer readable storage medium of claim 14, further comprising instructions that cause the processor to check a device execution history of the one or more computing devices to verify use of the one or more identified programmable hardware accelerators during execution of the accelerated version of the application.

27. A computing device configured to estimate programmable hardware acceleration benefit, comprising:

a processor;

a memory; and

a programmable hardware acceleration benefit estimator stored in the memory and executable by the processor, wherein the programmable hardware acceleration benefit estimator is configured to;

measure one or more of an execution time, an execution cost, or an execution energy consumption for executing a reference version of an application to perform one or more tasks for a data center customer;

identify one or more programmable hardware accelerators for the application, wherein the identified programmable hardware accelerators are not being utilized during execution of the reference version of the application;

automatically execute an accelerated version of the application to perform the one or more tasks performed by the reference version of the application, wherein the accelerated version of the application is accelerated by use of the one or more identified programmable hardware accelerators;

measure one or more of an execution time, an execution cost, or an execution energy consumption for executing the accelerated version of the application;

calculate one or more of a difference in execution time, a difference in execution cost, or a difference in execution energy consumption between the reference version of the application and the accelerated version of the application; and

provide one or more of the difference in execution time, the difference in execution cost, or the difference in execution energy consumption to the data center customer.

28. The computing device of claim 27, wherein the programmable hardware acceleration benefit estimator is configured to determine whether programmable hardware resources in the data center are available for use by the data center customer.

29. The computing device of claim 27, wherein the identifying one or more programmable hardware accelerators for the application comprises analyzing code in the application for a segment that is at least partially replaceable by a programmable hardware accelerator.

30. The computing device of claim 28, wherein the programmable hardware acceleration benefit estimator is configured to retrieve a programmable hardware accelerator from a library of programmable hardware accelerators to at least partially replace the segment of the application.

31. The computing device of claim 28, wherein the programmable hardware acceleration benefit estimator is configured to generate a programmable hardware accelerator to at least partially replace the segment of the application.

32. The computing device of claim 28, wherein the analyzing code in the application comprises a flow extraction process.

33. The computing device of claim 28, wherein the analyzing code in the application comprises hash comparison process.

34. The computing device of claim 27, wherein the programmable hardware acceleration benefit estimator is configured to compare outputs generated by the reference version of an application and the accelerated version of the application to determine whether the compared outputs are substantially identical.

35. The computing device of claim 27, wherein the automatically executing the accelerated version of the application comprises forking the reference version of the application to run accelerated segments of the application in parallel with un-accelerated segments executed by the reference version of the application.

36. The computing device of claim 27, wherein the automatically executing the accelerated version of the application is performed simultaneously with executing the reference version of the application.

37. The computing device of claim 27, wherein the measuring the reference version of the application, the identifying programmable hardware accelerators for the application, the measuring the accelerated version of the application, the calculating differences between the reference version of the application and the accelerated version of the application, and the providing the differences to the data center customer are performed automatically.

38. The computing device of claim 27, wherein the programmable hardware acceleration benefit estimator is configured to aggregate differences between the reference version of the application and the accelerated version of the application corresponding to multiple instances of executing the reference version of the application and the accelerated version of the application, and wherein the providing differences to the data center customer comprises providing a report summarizing differences over the multiple instances.

39. The computing device of claim 27, wherein the programmable hardware acceleration benefit estimator is configured to check a device execution history of the one or more computing devices to verify use of the one or more identified programmable hardware accelerators during execution of the accelerated version of the application.

40. A method for estimating programmable hardware acceleration benefit in a data center, the method comprising:

providing customer-activated acceleration of one or more applications executed by one or more computing devices in the data center, wherein the customer-activated acceleration employs programmable hardware accelerators to perform one or more application tasks;

selecting an application for which the customer-activated acceleration is not activated by a data center customer;

measuring one or more of an execution time, an execution cost, or an execution energy consumption for executing a reference version of the application without the customer-activated acceleration;

automatically executing, by the one or more computing devices, an accelerated version of the selected application;

measuring one or more of an execution time, an execution cost, or an execution energy consumption for executing the accelerated version of the application; and

providing, to the data center customer, a difference in execution time, a difference in execution cost, or a difference in execution energy consumption between the reference version of the application and the accelerated version of the application.

41. The method of claim 40, further comprising identifying one or more programmable hardware accelerators for the application, wherein the programmable hardware accelerators are not used by the reference version of the application.

42. The method of claim 41, wherein the identifying one or more programmable hardware accelerators for the application comprises analyzing code in the application for a segment that is at least partially replaceable by a programmable hardware accelerator.

43. The method of claim 40, further comprising comparing outputs generated by the reference version of an application and the accelerated version of the application to determine whether the compared outputs are substantially identical.

44. The method of claim 40, wherein the automatically executing the accelerated version of the application comprises forking the reference version of the application to execute accelerated segments of the application in parallel with un-accelerated segments executed by the reference version of the application.

45. The method of claim 40, wherein the automatically executing the accelerated version of the application is performed simultaneously with executing the reference version of the application.

46. The method of claim 40, further comprising aggregating differences between the reference version of the application and the accelerated version of the application corresponding to multiple instances of executing the reference version of the application and the accelerated version of the application, and wherein the providing differences to the data center customer comprises providing a report summarizing differences over the multiple instances.

47. The method of claim 40, further comprising checking a device execution history of the one or more computing devices to verify use of the one or more identified programmable hardware accelerators during execution of the accelerated version of the application.

48. The method of claim 40, further comprising determining whether programmable hardware resources in the data center are available for use by the data center customer.

49. The method of claim 40, wherein the selecting an application for which the customer-activated acceleration is not activated by a data center customer comprises selecting an application with an identified application type or an application comprising an identified application segment type, wherein the identified application type or the identified application segment type is established, prior to selecting the application, as operable with at least one identified programmable hardware accelerator.