A hardware-efficient dual-source data replication and local broadcast mechanism in distributed shared caches

Massive data access and edge computing have overwhelmed traditional edge computing devices. Researchers begin to study the deployment of many-core architectures with distributed shared caches in the chip of edge computing devices. The performance of traditional coherence protocols used in the distributed shared cache has become a potential bottleneck of system performance. To meet the requirements of low hardware overhead and high performance, we proposed a hardware-efficient dual-source data replication and local broadcast mechanism(DRLB). DRLB alleviates the inherent limitations of previous data replication mechanisms and reduces the overhead of coherence protocols. We verified the correctness and feasibility of DRLB by hardware implementation and proposed a hardware optimization of the non-blocking data replication mechanism. We evaluated DRLB and found that it can reduce the execution time by 16%, but only causes 0.82% of extra meta storage overhead, which also outperforms previous state-of-the-art data replication mechanism. What is more, the optimized DRLB can reduce the execution time by an average of 23%.