GD-GhOST: A Goal-Oriented Self-Tuning Caching Algorithm

A popular solution to internet performance problems is the widespread caching of data. Many caching algorithms have been proposed in the literature, most attempting to optimize for one criteria or another, and recent efforts have explored the automation and self-tuning of caching algorithms in response to observed workloads. We extend these efforts to consider the goal of optimizing for selectable performance criteria. With our proposed algorithm, we have shown performance matching and exceeding the best performance of the known greedy dual-size algorithms for either object or byte hit ratios across different web workloads. GD-GhOST consistently outperforms the other algorithms tested, at its worst observed performance GD-GhOST exhibited equivalent miss rates to those of the best applicable Greedy-Dual variant, while achieving miss rates that were 25% lower than the worst performing variant. For byte miss rates, GD-GhOST consistently demonstrated rates lower than the best applicable Greedy-Dual variant. At its best, GD-GhOST offered byte miss rates 10% lower than the best variant