• Summary
• Related Documents
  • Active Bibliography
  • Co-citation
• Version History

Abstract:

A multiprocessor system is unlikely to have access to information about the execution characteristics of the jobs it is to schedule. In this work, we are interested in scheduling algorithms for batch jobs that require no such knowledge (such algorithms are called nonclairvoyant) Preemptive scheduling (i.e., redistribution of processors) is important to reduce mean response time in multiprocessor systems, especially in the widely available network of workstations. Preemption is a method to adapt to the uncertain and changing nature of jobs and workloads. Unfortunately, preemption may incur large overheads if it is applied frequently. To account for the cost preemptions, we consider a number of simple scheduling algorithms classified by the number of preemptions they are allowed, ranging from none to an infinite number. The Equi-partition algorithm [18], which partitions the processors evenly between the uncompleted jobs, is an example of a simple scheduler that is non-clairvoyant and preempts only when jobs complete. Motwani et al. [15] show that the mean response time of jobs is within two of optimal for fully parallelizable jobs. Since parallel programs can have a wide variety of execution characteristics in practice, we consider a number of classifications of jobs according to how well they are able to utilize processors. Moreover a job may have both sequential and parallel phases in its computation. Hence, we allow jobs to have multiple phases, each of which may have different execution characteristics.

Citations