基于边际增益的二级缓存动态分配策略

Second-level Cache Dynamic Allocation Strategy Based on Marginal Gain

现有的ULC机制可有效减少多级缓存的数据冗余,并解决存储服务器端缓存访问的局部性较弱问题,但在存储服务器连接多个应用服务器的情况下,现有ULC在分配缓存容量时不能使存储服务器端缓存资源的边际收益最大化。为此,提出一种多应用共享缓存的二级缓存动态分配策略MG-ULC。该策略以ULC机制为基础,给出以边际增益为考虑因素的缓存分配的理论依据,并根据各应用的访问模式在二级缓存的边际增益动态分配缓存容量。实验结果表明,随着各应用服务器访问模式的变化,MG-ULC能比ULC更合理地分配二级缓存,从而达到更高的缓存利用率。

The existing Unified Level-aware Caching（ULC） protocol can effectively solve the problems of redundantly cache blocks in multilevel hierarchy and weaken locality at storage server cache. However, when there are multiple application servers sharing one storage server, the ULC method during allocating cache capacity cannot gain the maximal marginal profits of the storage server cache. So a second-level cache dynamic allocation strategy called MG-ULC is proposed, and it is designed for storage servers in which multiple applications share the same cache resources. Based on the ULC protocol, the theoretical foundation of cache allocation is given for marginal gain factor, and the MG-ULC dynamically allocates cache capacity in accordance with the second-level cache marginal gain of each application＇s access pattern. Experimental results show that, as each application＇s access pattern changes, the MG-ULC can allocate second-level cache more rationally than the ULC, thereby realizing a higher cache utilization rate.