Processor Management Techniques for Mesh-Connected Multiprocessors

This paper investigates various processor management techniques for improving the performance of mesh-connected multiprocessors. Three diierent techniques are analyzed. First, we use the smallest job rst (SJF) policy to improve the spatial parallelism in a mesh. Next, a policy called multitasking and multi-programming (M 2) is introduced. The M 2 scheme allows multiprogramming of jobs on various submeshes. Finally, a novel approach, called \limit allocation" is used for job allocation. With this policy, a job (sub-mesh) size is reduced if the job cannot be allocated. While all of the three approaches are viable alternatives , which in conjunction with any allocation algorithm can improve system performance beyond what is achievable with a complex allocation scheme and the usually assumed FCFS scheduling, the M 2 and limit allocation techniques are especially attractive for providing some additional features. The M 2 policy brings in the concept of time-sharing execution for better ef-ciency and the limit allocation shows how job size restriction can be beneecial for performance and fault-tolerance in a mesh topology. Moreover, the limit allocation scheme can outperform any other approach even using the simplest allocation policy.