Software Cache Eviction Policy Based on Stochastic Approach
This paper develops an innovative approach for guiding high-level software caches' eviction policy. The decision on which data to keep in the cache is made according to a stochastic analysis over the application data access behaviour. This approach shows it is possible to achieve high cache hit ratios with a reduced cache size. The effectiveness of the policy is tested and validated through the execution of two distinct benchmarks - the TPC-W and the oo7 benchmarks. The newly developed approach is flexible enough to be applied to any high-level software cache in an object-oriented system.