Phase Partitioning Methods for I/O Cache Optimization

System designers face large challenges in the storage hierarchy when putting new optimizations into practice. We consider a recent storage cache optimization, designed to reduce cache requirements and boost I/O performance, which adapts to individual program behavior. This scheme leverages a quantum-based design, and we observe that the choice of time quanta has a significant effect on performance and overheads. Accordingly, intelligent designs must judiciously select re-optimization points. We therefore develop new phase detection schemes that identify valued re-optimization points through interaction models between applications and this caching technique. We evaluate online and offline variants in the context of enterprise I/O workloads and observe hit-rate gains over 20% for a range of cache sizes.