Stochastic Weighted Caching with Per-Request Predictions

Abstract

Both the unweighted and weighted variants of caching are cornerstone problems in competitive analysis, with tight randomized competitive ratios of $Θ(\log k)$, where $k$ is the cache size. We study weighted caching with per-request predictions, where each page request is accompanied by a prediction of the next arrival time of the same page. For unweighted caching, such perfect predictions suffice to implement Belady's offline optimal rule. For weighted caching, however, previous work shows that a logarithmic lower bound persists even with perfect predictions on adversarial request sequences. Stochastic assumptions alone are also insufficient: the classical caching lower bounds already apply to uniform independent requests. In this work, we ask whether combining these two sources of additional information can overcome this barrier, and answer this question affirmatively. Specifically, we develop an efficient, deterministic cache eviction policy that achieves a competitive ratio of $15.75$ using perfect per-request predictions when page requests are drawn independently from an arbitrary, known static distribution in every step, and page costs are arbitrary. Our algorithm's performance degrades gracefully under erroneous predictions: we prove a bound of $O(\min\{1+E,\log k\})$ on the competitive ratio, where $E$ aggregates each page's average prediction error, and may be unknown to the algorithm.