Supercomputer Workload Characterization

这项工作将根据主要作业类别的分类、作业组合参数变化的速率、作为作业大小和CPU限制的函数的队列组成以及资源需求和应用程序算法之间的相关性来提供工作量表征。目标是通过最优队列控制最小化挂钟扩展因子，通过在实际内存中保持适当的作业组合来最小化空闲时间，并在不惩罚所有进程(包括最大的进程)的情况下处理一般组合中需要大量资源的作业。

The work will provide a workload characterization based on categorization of major job classes, rates at which the job mix parameters change, composition of queues as a function of job size and CPU limits, and a correlation between resource demands and application algorithms. The goals are to minimize the wall clock expansion factor through optimal queue control, minimize idle time by keeping the appropriate job mix in real memory, and process large resource demanding jobs in the general mix, without penalizing all processes, including the largest ones.