Method and system for scheduling computer cluster operation

Abstract

The invention provides a method and system for scheduling computer cluster operation. The system comprises at least one computing cell block and a core switch, wherein each computing cell block is connected with the core switch for finishing operation and / or data transmission; the core switch is connected with the at least one computing cell block for collecting the working parameters of the at least one computing cell block and setting a weight for each working parameter; according to the working parameters and the corresponding weights, the core switch is used for compiling a load balance index corresponding to each computing cell block; and based on a preset strategy, the core switch is used for allocating tasks of the operation and / or data transmission to the at least one computer cell block in accordance with the load balance indexes. By using the technical scheme, the operation is allocated to the at least one computing cell block; the workload of an operating system administrator is reduced, with regard to a cluster system with a large number of nodes; the operating efficiency of an operation management system is improved; the blocking of cross-switch-port data interaction resulting from a too-busy or idle port can be avoided; and allocation strategies are more reasonable.

Description

technical field [0001] The invention relates to computer technology, in particular to a method and system for computer cluster job scheduling. Background technique [0002] With the increasing demand for computing, the scale of computer clusters used for computing is also increasing. The cluster based on the diskless working mode has become an inevitable trend in the development of high-performance clusters. In the diskless cluster, all computing nodes Operations such as communication, data access, and access are all performed through the switch. The data throughput of the switch will be very large, and the interaction between ports will be very frequent. If the cluster is relatively large, there will be hundreds or thousands of computing nodes. These nodes are connected through multiple edge switches and core switches, and the network topology formed is extremely complex, so data transmission between nodes across switches will be significantly delayed. [0003] In the prio...

AI Technical Summary

Problems solved by technology

[0005] The object of the present invention is to provide a computer cluster job scheduling method and system, which is used to solve the problem that in the prior art, the allocation strategy for assigning jobs to nodes can no longer meet the delay of data transmission or network problems encountered in the large-scale computing process. problems such as congestion, and the inability to maximize the use of available computing resources to maximize benefits

Images