An Efficient Hybrid Cache Replacement Policy for Cloud Block Storage

With the popularity of cloud services, Cloud Block Storage (CBS) systems have been widely deployed by cloud providers. Cloud cache plays a vital role in maintaining high and stable performance in CBS systems. Traditionally, most of the research work used heuristic cache algorithms. However, we observe that in a typical CBS system, these algorithms parameters cannot be adjusted flexibly, and the performance is poor when encountering complex cloud workloads. Recent advances in machine learning open up new and attractive approaches for solving classic problems in computing systems. However, the existing methods are either simple to implement, unable to reflect the differences and characteristics of existing heuristic algorithms, or too complex to make full use of some insight from classical heuristic algorithms. To solve these problems, we propose a lightweight learning-based hybrid cache replacement policy, called LHC which can intelligently switch different cache policies so as to effectively recognize different workloads or even different phases within the same workload by using Q-Learning in CBS systems. We have evaluated LHC by using I/O traces collected from Alibaba Cloud, one of the largest cloud providers in the world, and several open-source traces. Experimental results show that LHC not only improves hit rate by up to 81.4% with an average of 36.5%, but also reduces request latency and data traffic to the backend storage by up to 18.7% and 16.4% with an average of 12.6% and 11.2%, compared with the several cache methods.