No abstract

Using the MRC (Miss Rate Curve) to guide cache capacity allocation is a common method in the storage system. However, optimal resource allocation is an NP-complete problem due to the cache performance cliff. Existing studies ignore this phenomenon or they use partitioning technology to eliminate it without considering the performance potential behind the cliff. This paper delves into this potential and proposes a cliff-aware cache resource allocation algorithm based on the inherent relationship between the capacity and the hit rate. Experiments show that these requests where the latency is less than 130 µs is increased by 33.3%. The proposed method obtains a significant cost reduction in DRAM and improves the hitting ratio of the cache layer.