Parallel computing control apparatus and parallel computing system

Patent metadata

Field	Value
Publication number	US-2016299795-A1
Application number	US-201615073722-A
Country	US
Kind code	A1
Filing date	Mar 18, 2016
Priority date	Apr 9, 2015
Publication date	Oct 13, 2016
Grant date	—

Abstract

A parallel computing control apparatus determines, among a plurality of nodes, relay nodes located on a path from a first node allocated to a job to a second node. The parallel computing control apparatus obtains an index value indicating a use state of resources in the relay nodes. The parallel computing control apparatus calculates an estimated value of a transfer period taken to transfer data on the job from the first node to the second node, on the basis of the index value and the memory usage of the job.

First claim

What is claimed is:

1 . A parallel computing control apparatus comprising:
a communication interface which performs communication with a plurality of nodes, the plurality of nodes being connected over a network; and a processor which allocates one or more of the plurality of nodes to a job, wherein the processor
determines, among the plurality of nodes, a relay node located on a path from a first node to a second node, the first node being allocated to the job,
obtains an index value indicating a use state of resources in the relay node, and
calculates an estimated value of a transfer period taken to transfer data on the job from the first node to the second node, based on the index value and a memory usage of the job.

2 . The parallel computing control apparatus according to claim 1 , wherein:
the estimated value is calculated further using a coefficient that gives a weight to the index value; and when the data on the job is transferred from the first node to the second node after the estimated value is calculated, the processor updates the coefficient, based on a result of transferring the data.

3 . The parallel computing control apparatus according to claim 1 , wherein one or both of a first index value and a second index value is used as the index value, the first index value indicating an execution state of another job running on the relay node, the second index value indicating a use state of a communication link.

4 . The parallel computing control apparatus according to claim 1 , wherein the processor determines whether to transfer the data on the job to the second node, based on the estimated value and an idle period of the second node indicated in a schedule.

5 . A parallel computing system comprising:
a plurality of nodes connected over a network; and a control apparatus which allocates one or more of the plurality of nodes to a job, wherein the control apparatus
determines, among the plurality of nodes, a relay node located on a path from a first node to a second node, the first node being allocated to the job,
obtains an index value indicating a use state of resources in the relay node, and
calculates an estimated value of a transfer period taken to transfer data on the job from the first node to the second node, based on the index value and a memory usage of the job.

6 . A non-transitory computer-readable storage medium storing a computer program that causes a computer to execute a process comprising:
determining, among a plurality of nodes connected over a network, a relay node located on a path from a first node to a second node, the first node being allocated to a job; obtaining an index value indicating a use state of resources in the relay node; and calculating an estimated value of a transfer period taken to transfer data on the job from the first node to the second node, based on the index value and a memory usage of the job.

Assignees

• Fujitsu Ltd

Inventors

• Kagami Takahiro
• Hashimoto Tsuyoshi

Classifications

• G06F9/4856 Primary

  resumption being on a different machine, e.g. task migration, virtual machine migration (G06F9/5088 takes precedence) · CPC title

• G06F9/5027

  the resource being a machine, e.g. CPUs, Servers, Terminals · CPC title

• G06F9/5083 Primary

  Techniques for rebalancing the load in a distributed system · CPC title