Two shortcomings, low availability and low scalability, are very widespread in existing high-performance computing.rn We propose hardware model in Fig.1 and software model in Fig.2 to overcome the a bove-mentioned shortcomings. Fig.1 shows four items: (1) cluster manager, (2) network user terminals, (3) cluster servers and (4) SCSI share disk. Fig.2 shows various relationships among five modules: (1) cluster manager monitor, (2) cluster manager agent, (3) cluster manager client service, (4) cluster manager client, and (5) aware application interface.rn A job-center manager system has been built and successfully put into use. In its implementation, a very important feature is its migration function shown in Fig .3, which overcomes the two shortcomings as follows:rn (1) When failure occurs somewhere, it automatically shifts the jobs somewhere else until recovery is completed either automatically or manually. Thus availability is assured.rn (2) When performance suffers from overload, it shifts a suitable amount of load to other workstations or PCs. Thus scalability is assured.%可升级性与可用性是当前软件开发的一个目标，集群技术就是为了实现这一目标而提出的。 基 于集群技术，建立了新的作业管理系统的硬、软件运行环境。在集群服务器运行环境的基础 上，提出了作业迁移的模型以及集群节点的恢复算法，该系统已经在一个大型作业管理系统 中得到具体实现。
展开▼
