Supercomputer Performance for Parallel Applications

如何在一个 将研究并行分时环境。 调度必须在给出并行应用程序之间进行平衡 访问他们需要的CPU，并保持交互性， 吞吐量 理解这一冲突需要 开发并行工作负载特性。 以来 工作负载的变化是非线性的， 这个动力系统问题将在克雷Y上实现， MP/864在SDSC。

The question of how processor scheduling should be done in a parallel timesharing environment will be investigated. Scheduling must be balanced between giving parallel applications access to CPUs they require, and maintaining interactivity and throughput. An understanding of this conflict requires the development of a parallel workload characterization. Since the workload changes nonlinearly, an iterative solution to resolve this dynamical system problem will be implemented on the Cray Y- MP/864 at SDSC.