It is well known that on-line preemptive scheduling algorithms can achieve efficient performance. A classic example is the Shortest Remaining Processing Time (SRPT) algorithm which is optimal for flow time scheduling, assuming preemption is costless. In real systems, however, preemption has significant overhead. In this paper we suggest a new model where preemption is costly. This introduces new considerations for preemptive scheduling algorithms and inherently calls for new scheduling strategies. We present a simple on-line algorithm and present lower bounds for on-line as well as efficient off-line algorithms which show that our algorithm performs close to optimal.
展开▼
