Computer system and data processing method for computer system
AssigneeHitachi, Ltd.
InventorsTakashi Yasui, Toshiyuki Ukai
Life of this patent
Estimating the term…
Abstract
A plurality of computers to execute jobs, a management computer to manage the execution of jobs and the disposition of data in the computers and a storage device storing data are interconnected via a network. The management program for the management computer divides the data into distributed data according to hint information and distributively disposes the distributed data and their replicas in memory storages allocated in memories of the computers. The computers execute the job using the distributed data allocated to their own memory. In the event of a fault in any of the computers, the management computer requests computers having the replicas of those distributed data disposed in the faulted computer to re-execute the job.
Claims10
A data processing method in a computer system having a plurality of first computers to distributively execute data processing, a second computer to manage the execution of the data processing by the plurality of the first computers, a storage device storing data used in the data processing, and a network interconnecting the plurality of the first computers, the second computer and the storage devices, the data processing method comprising the steps of:
- according to hint information given to the second computer and indicating a policy on the distributive disposition of the data in the plurality of the first computers, dividing the data held in the storage device into a plurality of pieces of distributed data;
- distributively disposing the distributed data and replicated data of each piece of the distributed data in memories of the plurality of the first computers;
- and
- performing data processing by the plurality of the first computers using the distributed data disposed in each of the first computers,
- wherein, in the event of a fault in any of the first computers, the data processing that was being executed in the faulty first computer at the time of the fault is re-executed by another first computer that holds the replicated data of the distributed data disposed in the faulty first computer,
- wherein the hint information includes information about whether the distributed data can be further distributively disposed, and
- wherein, when the hint information indicates that the distributed data can be further distributed, the distributively disposing step further includes dividing each of the pieces of the distributed data according to the hint information, creating the replicated data of the further divided distributed data, and disposing the pieces of the replicated data in first computers other than the first computer in which is disposed the original distributed data from which the replicated data has been created.
A computer system comprising:
- a storage device storing data;
- a plurality of first computers connected with the storage device and configured to manage data processing using the data;
- and
- a plurality of second computers connected with the first computers and configured to distributively execute the data processing,
- wherein the first computers each have a distributive disposition unit and an execution management unit,
- the distributive disposition unit configured to reference given hint information indicating a policy of distributive disposition of the data in a plurality of the second computers, divide the data held in the storage device into a plurality of pieces of distributed data and distributively dispose the pieces of the distributed data and each piece of replicated data of the distributed data into memories of the plurality of the second computers, and
- the execution management unit configured to request the second computers to execute the data processing and, in the event of a fault in any of the second computers, request another second computer to re-execute the data processing that was being executed in the faulted second computer at the time of fault,
- wherein the second computers each have a memory management unit and a data processing unit,
- the memory management unit being configured to hold the distributed data in a memory of its own computer in response to a request from the distributive disposition unit, and
- the data processing unit being configured to execute the data processing in response to a request from the execution management unit,
- wherein the hint information includes information indicating whether each piece of replicated data of the distributed data can be further distributively disposed.
Dependencies are inferred from the claim text.