Publication Date: 2019/10/22
Abstract: When all blocks are occupied inside a cache of a processor, the system uses a concept known as ‘Replacement policy’ to evict an existing block inside the cache memory to bring the required block from the main memory. There are many replacement policies applied to caches which itself have many levels. The policy ‘Application Behavior Aware Re-Reference Interval Prediction’ is an extension of existing policy of ‘Re-Reference Interval Prediction.’ The ABRIP policy has two levels of RRPV, implemented in two levels as we will see further. However, the ABRIP policy’s algorithm waits until all the cache blocks have max RRPV value hence ‘decaying’ the cache blocks. Proposed method intends to correct the decaying phenomenon by implementing Dynamic Non-decaying ABRIP or DND- ABRIP. Gem5 simulator was used in system emulation mode with SPEC2006 Benchmarks. We see a 0.1% improvement in IPC, 1.4% improvement in Read hits and 0.3% improvement in Write Hits for DND-ABRIP over ABRIP.
Keywords: Replacement Policy, Re-Reference Interval Prediction, Application Behavioral Re-Reference Interval Prediction, decaying, Dynamic Non-decaying ABRIP.
DOI: No DOI Available
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT19OCT1699.pdf
REFERENCES