Abstract

We deal with a natural generalization of one of the seminal problems in the study of on-line computation, list management [ST85a, IRSW91, Irani91]. We consider a doubly linked list with a pointer pointing to the last item accessed. Our goal, as with list management, is to minimize the total number of pointers traversed over a sequence of searches. This problem models issues of dynamic adaptive disk optimization. In this work we initiate the study of this problem in the competitive setting. Our main result is showing non-trivial upper bounds on the competitive ratio. For restricted access patterns we give algorithms that have much better competitive ratios. To complement those bounds we show that there is no constant-competitive algorithm for the problem with arbitrary access patterns, even against an adversary that does not rearrange its items. Department of Computer Science, Tel Aviv University, Tel Aviv. Research supported in part by a grant from the Israel Academy of Sciences. E-M...

Citations