Dynamic selection of hardware processors for stream processing

• US 9,569,221 B1
• Filed: 09/29/2014
• Issued: 02/14/2017
• Est. Priority Date: 09/29/2014
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Filippi, Enrica Maria, Chaudhri, Apoorv
• Primary Examiner(s): Ghaffari, Abu
• Application Number:
  US14/500,751
• Time in Patent Office: 869 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 1/329 by task scheduling
• CPC Class Codes: G06F 1/3293 by switching to a less power-consuming processor, e.g. sub-CPU
• 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: G06T 1/20 Processor architectures; Processor configuration, e.g. pipelining
• CPC Class Codes: G06T 15/005 General purpose rendering architectures
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: H04N 19/436 using parallelised computational arrangements
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A computing platform supports stream processing pipelines, each of which comprises a sequence of stream processing tools. Upon specification of a stream processing pipeline, multiple available hardware processors are evaluated to determine which of the processor is capable of executing each tool of the pipeline while satisfying specified performance goals. Among these processors, a hardware processor is selected for each pipeline tool that will minimize power consumption.

Claims

1. A computing platform comprising:

multiple hardware processors;

multiple image processing tools, the multiple image processing tools including a first image processing tool that is executable by an individual hardware processor of the multiple hardware processors;

an image processing pipeline configured to receive an image data stream and to process the image data stream with a sequence of the multiple image processing tools;

a resource manager configured to receive from an application (a) an instruction to execute the image processing pipeline and (b) a first performance goal that indicates an allowable latency of the first image processing tool;

the resource manager being further configured to perform acts based at least in part on receiving the instruction from the application, the acts comprising;

determining that a first hardware processor of the multiple hardware processors is configured to execute the first image processing tool with no more than the allowable latency;

determining that a second hardware processor of the multiple processors is configured to execute the first image processing tool with no more than the allowable latency;

determining that executing the first image processing tool on the first hardware processor uses less power than executing the first image processing tool on the second hardware processor; and

selecting the first hardware processor to execute the first image processing tool.

2. The computing platform of claim 1, wherein the selecting the first hardware processor is based at least on part on one or more of:

an available capacity of the first hardware processor;

a priority of the image processing pipeline relative to a second image processing pipeline;

or

whether the image processing pipeline executes continuously.

3. The computing platform of claim 1, the acts further comprising:

receiving a second performance goal for the image processing tool from the application;

determining that the first hardware processor is configured to satisfy the second performance goal when executing the first image processing tool;

determining that the second hardware processor is configured to satisfy the second performance goal when executing the first image processing tool;

wherein second performance goal indicates one or more of;

whether the image processing pipeline executes continuously;

or

a priority of the image processing pipeline relative to a second image processing pipeline.

4. A method, comprising:

receiving an instruction from an application to execute a processing pipeline, wherein the processing pipeline comprises multiple stream processing tools;

receiving an indication of a performance goal from the application, the performance goal indicating an allowable latency for a stream processing tool of the multiple stream processing tools;

determining that a first hardware processor is configured to execute the stream processing tool with no more than the allowable latency;

determining that a second hardware processor is configured to execute the stream processing tool with no more than the allowable latency;

determining that executing the stream processing tool on the first hardware processor uses less power than executing the stream processing tool on the second hardware processor; and

selecting the first hardware processor to execute the stream processing tool.

5. The method of claim 4, wherein selecting the first hardware processor comprises selecting the first hardware processor based at least on one or more of:

available capacity of the first hardware processor;

priority of the processing pipeline;

whether the processing pipeline executes continuously;

execution latency of the first hardware processor;

or

power usage of the first hardware processor.

6. The method of claim 4, wherein the performance goal further indicates one or more of:

whether the processing pipeline executes continuously;

or

priority of the processing pipeline relative to another processing pipeline.

7. The method of claim 4, further comprising compiling the stream processing tool to create an executable instruction sequence for execution by the first hardware processor, wherein compiling the stream processing tool is performed in response to selecting the first hardware processor.

8. The method of claim 4, further comprising receiving, from the application, one or more commands that specify the multiple stream processing tools of the processing pipeline.

9. The method of claim 4, wherein selecting the first hardware processor comprises selecting the first hardware processor based at least in part on projected performance of the first hardware processor when executing the stream processing tool in comparison to projected power usage of the first hardware processor when executing the stream processing tool.

10. The method of claim 4, further comprising determining a current processing load of the first hardware processor, wherein selecting the first hardware processor is based at least in part on the current processing load of the first hardware processor.

11. The method of claim 4, further comprising selecting a clock speed of the first hardware processor based at least on the performance goal.

12. The method of claim 4, wherein the first hardware processor comprises:

a general-purposes central-processing unit;

a processor core;

a graphics processing unit;

a digital signal processor;

a processor extension;

or

a vector processor.

13. A computing platform comprising:

a first hardware processor;

a second hardware processor;

a resource manager configured to perform acts comprising;

receiving an instruction from an application to execute a processing pipeline, wherein the processing pipeline comprises multiple stream processing tools;

receiving an indication of a performance goal from the application, the performance goal indicating an allowable latency for a stream processing tool of the multiple stream processing tools;

determining that the first hardware processor is configured to execute the stream processing tool with no more than the allowable latency;

determining that the second hardware processor is configured to execute the stream processing tool with no more than the allowable latency;

determining that executing the stream processing tool on the first hardware processor uses less power than executing the stream processing tool on the second hardware processor; and

selecting the first hardware processor to execute the stream processing tool,

determining that executing the stream processing tool on the first hardware processor uses less power than executing the stream processing tool on the second hardware processor; and

selecting the first hardware processor to execute the stream processing tool.

14. The computing platform of claim 13, wherein selecting the first hardware processor is based at least on part on one or more of:

available capacity of the first hardware processor;

priority of the processing pipeline;

whether the processing pipeline executes continuously;

execution latency of the first hardware processor;

power usage of the first hardware processor;

or

an allowable latency of the processing pipeline.

15. The computing platform of claim 13, wherein the performance goal further indicates one or more of:

whether the processing pipeline executes continuously;

or

priority of the processing pipeline relative to another processing pipeline.

16. The computing platform of claim 13, the acts further comprising compiling the stream processing tool to create an executable instruction sequence for execution by the first hardware processor, wherein compiling the stream processing tool is performed in response to selecting the first hardware processor.

17. The computing platform of claim 13, wherein selecting the first hardware processor is based at least in part on projected performance of the first hardware processor when executing the stream processing tool in comparison to projected power usage of the first hardware processor when executing the stream processing tool.

18. The computing platform of claim 13, the acts further comprising:

determining a current processing load of the first hardware processor; and

wherein selecting the first hardware processor is based at least in part on the current processing load of the first hardware processor.

19. The computing platform of claim 13, the acts further comprising selecting a clock speed of the first hardware processor based at least on the performance goal.

20. The computing platform of claim 13, wherein the first hardware processor comprises:

a general-purposes central-processing unit;

a processor core;

a graphics processing unit;

a digital signal processor;

a processor extension;

or

a vector processor.