MABF: H/W based page locality detection techniques with minimal performance degradation

Abstract

Techinically, role of main memory is to decrease gab between speed of processor and one of disk. Even though size of memory gets larger, data processing at server computing environment for big data, which is one of the hottest keywords in the world nowdays, or memory size at mobile computing environment still make it important to manage memory resource effectively. One of ways for it is to use Page Locality Detection. MABF(Memory Accessed Based Filtering), which we propose, support several key features, which are to detect page locality per thread, not process and to give TLB information, which is to see if each page number is in TLB, as well as the number of each page access. MABF also achieves at least about 90% accuracy with minimal performance degradation.