Information processing device, job scheduling method, and job scheduling program

Patent metadata

Field	Value
Publication number	US-9244740-B2
Application number	US-201414196406-A
Country	US
Kind code	B2
Filing date	Mar 4, 2014
Priority date	Mar 4, 2013
Publication date	Jan 26, 2016
Grant date	Jan 26, 2016

Abstract

An information processing device includes: a measurement unit 10 for measuring respective use rates of a plurality of coprocessors each for executing a job, respective use rates of a plurality of interface cards each for passing data input or output by each of the plurality of coprocessors, and respective latencies and respective throughputs in communication between the plurality of coprocessors and the plurality of interface cards; and a determination unit 20 for determining a coprocessor that is to execute the job from among the plurality of coprocessors, based on a result of the measurement by the measurement unit 10.

First claim

What is claimed is:

1. An information processing device comprising:
a processor; and
a non-transitory computer-readable medium storing computer-executable code that the processor executes to realize a plurality of units comprising:
a measurement unit for measuring respective use rates of a plurality of coprocessors each for executing a job, respective use rates of a plurality of interface cards where data input or output by each of the plurality of coprocessors pass, and respective latencies and respective throughputs in communication between the plurality of coprocessors and the plurality of interface cards; and
a determination unit for determining a coprocessor that is to execute the job from among the plurality of coprocessors, based on a result of the measurement by the measurement unit, and determining an interface card where the data input or output by the coprocessor pass from among the plurality of interface cards;
wherein the measurement unit includes a job profile creation unit for measuring an execution time, a data transfer amount, and a number of input or output request issuances of the job executed by a coprocessor, and creating a job profile including a result of the measurement, and
wherein, when determining the coprocessor that is to execute the job and the interface card where the data input or output by the coprocessor pass, the determination unit calculates, for each of all combinations of the plurality of coprocessors and the plurality of interface cards, an estimated execution time of the job to be executed, and selects a combination of a coprocessor and an interface card with a shortest estimated execution time, using the job profile.

2. The information processing device according to claim 1 , wherein the measurement unit includes: a communication affinity measurement unit for measuring a latency and a throughput in communication between each of the plurality of coprocessors and each of the plurality of interface cards, and creating communication affinity information including a result of the measurement; a transfer amount monitor unit for measuring a data transfer amount of each of the plurality of interface cards, and creating transfer amount information including a result of the measurement; and a coprocessor use rate monitor unit for acquiring a core use rate from each of the plurality of coprocessors, and registering the core use rate in coprocessor use rate information, and
wherein the determination unit acquires the communication affinity information, the transfer amount information, and the coprocessor use rate information from the measurement unit as the result of the measurement.

3. A job scheduling method comprising:
measuring respective use rates of a plurality of coprocessors each for executing a job, respective use rates of a plurality of interface cards where data input or output by each of the plurality of coprocessors pass, and respective latencies and respective throughputs in communication between the plurality of coprocessors and the plurality of interface cards;
determining a coprocessor that is to execute the job from among the plurality of coprocessors, based on a result of the measurement, and determining an interface card where the data input or output by the coprocessor pass from among the plurality of interface cards;
measuring an execution time, a data transfer amount, and a number of input or output request issuances of the job executed by a coprocessor, and creating a job profile including a result of the measurement; and
when determining the coprocessor that is to execute the job and the interface card where the data input or output by the coprocessor pass, calculating, for each of all combinations of the plurality of coprocessors and the plurality of interface cards, an estimated execution time of the job to be executed, and selecting a combination of a coprocessor and an interface card with a shortest estimated execution time, using the job profile.

4. A non-transitory computer-readable recording medium having recorded thereon a job scheduling program for causing a computer to execute:
a process of measuring respective use rates of a plurality of coprocessors each for executing a job, respective use rates of a plurality of interface cards where data input or output by each of the plurality of coprocessors pass, and respective latencies and respective throughputs in communication between the plurality of coprocessors and the plurality of interface cards;
a process of determining a coprocessor that is to execute the job from among the plurality of coprocessors, based on a result of the measurement, and determining an interface card where the data input or output by the coprocessor pass from among the plurality of interface cards;
a process of measuring an execution time, a data transfer amount, and a number of input or output request issuances of the job executed by a coprocessor, and creating a job profile including a result of the measurement; and
a process of, when determining the coprocessor that is to execute the job and the interface card where the data input or output by the coprocessor pass, calculating, for each of all combinations of the plurality of coprocessors and the plurality of interface cards, an estimated execution time of the job to be executed, and selecting a combination of a coprocessor and an interface card with a shortest estimated execution time, using the job profile.

Assignees

• Nec Corp

Inventors

• Akimoto Shunsuke

Classifications

• G06F9/505 Primary

  considering the load · CPC title

• G06F2209/501

  Performance criteria · CPC title