Nonclairvoyant scheduling

• Published in: Theoretical computer science Vol. 130; no. 1; pp. 17 - 47
• Main Authors: Motwani, Rajeev, Phillips, Steven, Torng, Eric
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.08.1994
• ISSN: 0304-3975; 1879-2294
• DOI: 10.1016/0304-3975(94)90151-1

Virtually all research in scheduling theory has been concerned with clairvoyant scheduling where it is assumed that the characteristics of a job (in particular, its execution time, release time and dependence on other jobs) are known a priori. This assumption is invalid for scheduling problems that arise in time-sharing operating systems where the scheduler must provide fast turnaround for processes being generated by the users without any knowledge of the future behavior of these processes. We study preemptive, nonclairvoyant scheduling schemes where the scheduler has no knowledge of the jobs' characteristics. We develop a model for evaluating scheduling strategies for single and multiprocessor systems. This model compares the nonclairvoyant scheduler against the optimal clairvoyant scheduler, and it takes into account various issues such as release time, execution time, preemption cost, and the inter-dependence between jobs. Within this model we study some standard scheduling algorithms described in the systems literature, and we provide some theoretical justification for their effectiveness in practice by presenting some randomized and deterministic upper and lower bounds.