Cache Prediction for XIV

Enterprise block storage systems have been experiencing a revolution with the introduction of solid-state-disks, and the demise of high RPM disks. The IBM XIV storage system in particular has embraced this concept with the introduction of large capacity SSD based caches. Customers can purchase the additional cache, for a price. The natural customer question is therefore, “will SSD caching improve the performance of my workload?”. To try to answer this question, and others like it, XIV has incorporated block-tracing into its platform, with complementary analysis and prediction tools. We have constructed an efficient cache prediction tool, that works for a modern controller two level cache: RAM over SSD. It scales to terabytes of cache, and massive multi day traces. Given a customer workload, it can quickly figure out if the addition of SSD cache is likely to improve cache hit rates.