Efficient Data Placement and Replacement Algorithms for Multiple-Level Memory Hierachy

Improvement in memory speeds has not kept pace with increasing processor speed. Consequently, the gap between processor and memory speed is growing. In order to meet the computational needs of the next decade, the design projects of PetaFlop computer introduce many levels of memory hierarchy with huge amount of speed differences between levels. As one of the design point for PetaFlop computers, this paper proposes algorithms for static data replacement on deep memory hierarchy. Two memory models are studied. The first model consists of a uni-processor with one physical k-level memory hierarchy. The second model contains a multiprocessor with multi-level memory hierarchy, where some levels may be physically shared by multiple processors. We design data replacement algorithms under both platforms. Our experimental results show the effectiveness of our approaches by comparing total memory accessing time obtained by the schemes such as OPT, LRU and the proposed algorithms.