Dynamic beats fixed: On phase-based algorithms for file migration

Abstract

We construct a deterministic 4-competitive algorithm for the online file migration problem, beating the currently best 20-year-old, 4.086-competitive MTLM algorithm by Bartal et al. (SODA 1997). Like MTLM, our algorithm also operates in phases, but it adapts their lengths dynamically depending on the geometry of requests seen so far. The improvement was obtained by carefully analyzing a linear model (factor-revealing LP) of a single phase of the algorithm. We also show that if an online algorithm operates in phases of fixed length and the adversary is able to modify the graph between phases, then the competitive ratio is at least 4.086.