Random Adaptive Cache Placement Policy

TL;DR

Traditional cache replacement policies struggle with complex access patterns, leading to underutilized cache sets. RAC introduces a hybrid approach that blends random eviction with a V-Way-inspired architecture featuring decoupled tag and data storage in a $16$-way set-associative cache, organized with a $2048$-entry tag table ($32$ ways) and a $2048$-entry data table ($16$ ways). It uses a shared pool and forward/reverse pointers to dynamically adapt to workload spikes, achieving up to $80.82\%$ hit rate on challenging traces, albeit with only modest IPC gains. These results suggest RAC’s potential to improve cache utilization and memory access latency, while highlighting opportunities for further IPC optimization and real-time, possibly ML-assisted, replacement decisions.

Abstract

This paper presents a new hybrid cache replacement algorithm that combines random allocation with a modified V-Way cache implementation. Our RAC adapts to complex cache access patterns and optimizes cache usage by improving the utilization of cache sets, unlike traditional cache policies. The algorithm utilizes a 16-way set-associative cache with 2048 sets, incorporating dynamic allocation and flexible tag management. RAC extends the V-Way cache design and its variants by optimizing tag and data storage for enhanced efficiency. We evaluated the algorithm using the ChampSim simulator with four diverse benchmark traces and observed significant improvements in cache hit rates up to 80.82% hit rate. Although the improvements in the instructions per cycle (IPC) were moderate, our findings emphasize the algorithm's potential to enhance cache utilization and reduce memory access times.

Figures (1)

• Figure 1: Visualization of Random Adaptive Cache Implementation