We now compare performance of SARC to two state-of-the-art LRU variants, namely, LRU Top and LRU Bottom (see Section III-A), using the SPC-1 Like benchmark on Shark hardware that is running our experimental software implementations.
SARC minimizes the number of misses. The effect of such minimization can be studied from two perspectives, namely, that of the client and that of the storage controller. To a client, at a fixed load, the most important metric is the average response time. To study the performance seen by a client over a large spectrum of real-life operating scenarios, for the SPC-1 Like benchmark, we compare throughput versus average response time curves for all the three algorithms. To a storage controller, a crucial metric is the internal load on the RAID arrays. Within Shark, we measure the rate of tracks being staged to the cache due to read requests (both random and sequential). We will demonstrate how SARC convincingly outperforms both the LRU variants from the perspective of the client as well as the storage controller.