🤖 AI Summary
Blockchain’s serial execution severely limits throughput, while existing parallel execution engines suffer from resource contention among transactions—particularly under hot-contract workloads—yielding only marginal improvements (e.g., +60%). This paper proposes a resource-access-pattern-aware dynamic block construction algorithm. It combines static analysis with lightweight runtime resource fingerprinting to construct a modular scheduling graph, enabling intelligent transaction partitioning and reordering. The approach requires no client-side hints or optimistic conflict detection and integrates transparently into mainstream parallel execution engines. Experimental evaluation across diverse representative workloads demonstrates over 2× throughput improvement versus baseline serial execution and significantly outperforms state-of-the-art parallel schemes. Crucially, it effectively alleviates the execution bottleneck induced by single hot contracts, marking a substantial advance in scalable blockchain execution.
📝 Abstract
Many blockchains such as Ethereum execute all incoming transactions sequentially significantly limiting the potential throughput. A common approach to scale execution is parallel execution engines that fully utilize modern multi-core architectures. Parallel execution is then either done optimistically, by executing transactions in parallel and detecting conflicts on the fly, or guided, by requiring exhaustive client transaction hints and scheduling transactions accordingly. However, recent studies have shown that the performance of parallel execution engines depends on the nature of the underlying workload. In fact, in some cases, only a 60% speed-up compared to sequential execution could be obtained. This is the case, as transactions that access the same resources must be executed sequentially. For example, if 10% of the transactions in a block access the same resource, the execution cannot meaningfully scale beyond 10 cores. Therefore, a single popular application can bottleneck the execution and limit the potential throughput. In this paper, we introduce Anthemius, a block construction algorithm that optimizes parallel transaction execution throughput. We evaluate Anthemius exhaustively under a range of workloads, and show that Anthemius enables the underlying parallel execution engine to process over twice as many transactions.