Scheduling a metacomputer by an implicit voting system

We address the problem of scheduling hardware requests to a metacomputer environment. The environment may consist of a heterogeneous set of computing resources. A high level abstraction of the metacomputer is assumed to take care of the heterogeneity. The scheduling problem is considered in two parts. Some requests are extracted from the waiting room at first and scheduled afterwards by using bin packing to allocate the hardware resources. The bin packing algorithm is dynamically turned by the current mixture of the requests to tradeoff utilization versus response time. The extraction algorithm ensures prioritizing without starving lower priority jobs. The Implicit Voting System is compared to a well known bin packing algorithm by simulation. IVS improves the system utilization by 30% and simultaneously the average waiting time by 9%. Its behavior is also validated by integration into the Computing Center Software, CCS. Currently, this software package serves a pool of different transputer based MPP-systems to a large user community.<
>