Sharif Digital Repository

From the programmer’s point of view, main memory allocation to a program is contigious, however this is non-contigious and the program is scattered in physical memory here and there. Assuming main memory partitioned into regions, each program accesses in its life time different regions which are not necessariliy close and occupy different percentages of the cache capacity. Considering what the replacement policy is chosen, the cache would be partitioned differently among regions, e.g., the commonly used LRU policy partitions the cache among regions on a demand basis, giving the more cache resource to regions whose miss ratio is higher, which is not necessarily optimal. In this thesis, we...