An Integrated Processor Management Scheme for Parallel Computers

大规模并行多计算机的性能在很大程度上取决于其处理器管理方案。不幸的是，当前一代商用多计算机所采用的方案不能有效地利用系统资源。处理器管理涉及处理器分配和作业调度。处理器分配也影响处理器间通信延迟。研究人员已经各自独立地解决了这些问题。然而，评估这些问题对系统性能的综合影响是必要的。因此，本计画聚焦于处理器分配、作业调度和处理器间通讯的综合研究。研究目标将分三个阶段完成。第一阶段包括设计和开发有效的处理器分配和作业调度方案，同时为处理器间通信采用合适的消息路由算法。在第二阶段，研究分配、调度和进程间通信对系统性能的综合影响。提出的研究的第三阶段涉及这些方案的实验验证。＊＊＊

The performance of massively parallel multicomputers depends largely on their processor management scheme. Unfortunately, the schemes employed in the current generation of commercial multicomputers do not efficiently utilize the system resources. Processor management is concerned with processor allocation and job scheduling. Processor allocation also affects the interprocessor communication latency. Researchers have addressed these issues independently of each other. However, it is essential to evaluate the combined effect of these issues on system performance. Thus, this project focuses on an integrated study of processor allocation, job scheduling, and interprocessor communication. The research objectives will be accomplished in three phases. The first phase encompasses the design and development of efficient processor allocation and job scheduling schemes while employing a suitable message routing algorithm for interprocessor communication. In the second phase, the integrated effect of allocation, scheduling, and interprocess communication on system performance is to be studied. The third phase of the proposed research is concerned with the experimental validation of these schemes. ***