A Comparative Study of Directory-Based Cache Coherence Protocols in Shared-Memory Multiprocessors

Abstract

The exponential growth of multiprocessor systems has made cache coherence a critical issue in shared-memory architectures. Directory-based protocols, in contrast to snooping-based approaches, offer scalable solutions for large-scale systems. This paper presents a comparative analysis of prominent directory-based cache coherence protocols including Full-Map, Limited Pointer, and Sparse Directory protocols. Each protocol is evaluated based on performance, memory overhead, latency, and scalability. With the growing demand for high-performance parallel computing, shared-memory multiprocessor systems have become increasingly prevalent. A key challenge in such systems is maintaining cache coherence when multiple processors simultaneously access shared data. This paper presents a comprehensive comparison of directory-based cache coherence protocols, focusing on their structure, performance, scalability, and communication overhead. We analyze three widely used protocols—Full-map, Limited Directory, and Sparse Directory—evaluating them based on memory overhead, latency, traffic reduction, and scalability. Simulation-based results demonstrate trade-offs in different coherence mechanisms, and highlight the protocols' suitability for different system configurations.