CHIDDAM: a data mining based technique for cache hierarchy determination in commercial applications

In this paper, we present a cache hierarchy design using data mining (CHIDDAM) methodology to improve the memory efficiency for a given type of commercial application and its data set characteristics. Performance analysis, workload characterization, decision tree induction (a classification data mining method), and a greedy search algorithm are used in determining the optimal number of levels and sizes of the cache in the hierarchy. SimICS, a full system simulator, is used to simulate an ×86 machine loaded with Enterprise Linux operating system. Popular representative applications in data mining, such as, C4.5 and Apriori are used as benchmark applications for our simulation analysis. However, the proposed methodology is generic and can be applied to other applications for determine the design details of memory subsystems.