Architecture-aware transaction scheduling methodology in high-load blockchain networks

Abstract

The paper addresses the issue of Hyperledger Besu blockchain node performance degradation in high-load networks caused by the State Bloat phenomenon. An architecture-aware transaction scheduling methodology, RATEM, is proposed based on prefix-based request clustering to optimize Last Level Cache (LLC) utilization. The mechanisms of interaction between the proposed method, the JVM runtime environment, and hardware prefetching units are substantiated. Theoretical analysis indicates a potential 10–15% reduction in average RPC request processing latency.