Design and implementation of SSD aware Heterogeneous cache Algorithm: A Two-level caching algorithm for RAID storage systems

Existing conventional RAID storage systems are unable to provide expected storage performance under the explosive growth in data volumes. One of the solution is introducing SSDs (Solid State Drives), a promising storage medium which provides high performance and power efficiency, as a cache to loosen the performance bottleneck of RAID storage systems. However, using DRAM as a universe buffer area roughly, regardless of different characteristics between SSDs and the RAID, is inefficient. This paper proposes SHA (SSD aware Heterogeneous cache Algorithm), which divides the DRAM area into two regions: the SSD region and the RAID region, and adaptively adjusts these two regions according to the workload. The SSD region uses a victim-cost based replacement algorithm, and cached blocks are evaluated by their access patterns and the price of evicting them. The RAID region uses a LRU algorithm at the stripe granularity. This algorithm explores the parallelism of RAIDs and aggregates writes to reduce write penalties. Besides, a hot data filter is used to detect the temporal locality of evicted blocks from the RAID region, and blocks with temporal locality are sent to SSDs. This paper introduces the design of the SHA in detail, and performs intensively performance evaluations on the SHA.