Temporal Locality in Today's Content Caching: Why it Matters & How to Model It

摘要

If timing, as we are often reminded, is everything, it's also a critical part of today's caching system. The main reason is that the popularity of an item - like a video on YouTube, as considered by the authors of this paper - evolves quickly and it can easily fool a simplistic cache management strategy. In particular a heuristic model called independent reference is often used in evaluating caching networks even today, and this precisely states that popularity is fixed with time and that a new request is independent of the past. In reality requests typically occur in bursts, a phenomenon sometimes called temporal locality. How much of this locality matters in today's caching? Which methods can be used in practice to extract and recognize it when it occurs? Are there models that retain some of the simplicity of the independent reference model while accounting for temporal locality?