Cluster Of Application-oriented Resource Management System Design And Realization

In recent years, people have become increasingly demanding computer processing of high performance, single server has been increasingly unable to meet the needs of users, and super-computers are very expensive, by contrast, cluster system has high performance-price ratio, high availability, and high expansibility characteristics, thus gradually become the mainstream of high performance computing. Cluster system is a group of computers which are interconnected through high-speed network, and they provide application services to user as a whole. Oil exploration is an industry needs mass data processing; it has the characteristics of large amount of calculations and complicated algorithm, the demand for high performance computing is increasing.The cluster resource management system's main task is to provide unified management of cluster resources, and achieve higher performance through the improvement of system resource utilization rate. The cluster resource management system in this paper is a key support system for Southwest Petroleum Geophysical Corporation in Seismic data processing and interpretation system, and it's a dedicated system for seismic data processing features. According to user needs, this system provides unified management of cluster resources, through the rational allocation of users to submit jobs, increase the resource utilization rate, thereby improving the efficiency of the system. The system mainly implements the following functions:(1) This system realizes the management of resources'objects in server and nodes, all computing resources in cluster are managed by server which contains nodes and jobs'management, computing resources are managed by compute nodes which contain node's resources and job information.(2) According to the characteristics of seismic data processing and the software and hardware requirements in the project, based on the general scheduling algorithm we design a scheduling strategy for the system. In order to facilitate the replacement of scheduling algorithm and reduce server response time of various requests, the server object management and scheduling process are designed separately, realized several job scheduling algorithms and considered the system load balance and resource reserve strategy.(3) In order to guarantee the system to provide continued and reliable services, reduce the loss caused by the cluster server or compute node failure, we design and implement dual-machine backup function and node's failure processing function, which implemented critical data backup and recovery in server and transfer of unfinished job in failure node.This system uses a client/server design model, clients only communicate with server, and all resource objects are unified managed by server. The system designed with event-driven mode, the operation of each node in the cluster depends on received message, and considered to reduce coupling between modules in order to facilitate system upgrades.