A Fast-and-Effective Early-Stage Multi-level Cache Optimization Method Based on Reuse-Distance Analysis

Abstract: In this paper, we propose a practical and effective approach allowing designers to optimize multi-level cache size at the early system design phase. Our key contribution is to generalize the reuse distance analysis method and develop an effective and practical cache design optimization approach. We adopt a simple scanning search method to locate optimal cache solutions in terms of cache size, power consumption, or average data access delay. The proposed approach is particularly useful for early-phase system designers and is verified to be 150 to 250 times faster than the traditional simulation-based approach. In addition, we also introduce a simplified analytical model and provide designers insights about how cache design parameters may affect the expected results. As a result, designers can make an adequate decision in the early system design phase.