Presentation

DescriptionMaintaining high and stable performance within the storage system is challenging. The optimization of cache capacity overload is a concise solution to the performance enhancement of the storage system. Some rule-based methods are proposed to solve this problem but still face challenges because the real-world workloads might be time-varying and the storage architecture is so complex and coupled that engineers are unable to find the bottleneck and tune it. We propose a light-weight learning-based control framework for cache capacity overload problem, called L-QoCo. Extensive experiments show that L-QoCo posses strong adaptability and significantly improves the performance of the storage systems.