Paging on a RAM with Limited Resources

Abstract

The paging problem is that of deciding which pages to keep in a cache of k pages to minimize the number of page faults for a two-level memory system. We consider efficient paging algorithms that require limited space and time on a unit-cost RAM with word size w. We present an implementation of the randomized marking algorithm of Fiat et al. that uses only k + o(k) bits of space while supporting page requests in O(1) worst-case time. In addition, we present a novel k-competitive deterministic algorithm called Surely Not Recently Used (SNRU) that approximates the Least Recently Used (LRU) algorithm. For any constant 1/k ≤ ε < 1/2, SNRU(∈) ensures that the ⌈(1/2 - ε)k⌉ most recently requested pages are in the cache, while using only O(1/ε) worst-case time per request and 2k + o(k) bits of space.

This research was supported in part by the IBM Toronto Centre for Advanced Studies.