Ethereum execution clients implement identical Ethereum Virtual Machine (EVM) semantics, yet can exhibit substantially different performance in practice. We study this divergence through a three-step systems analysis of two representative clients, Geth and Reth. First, across historical synchronization, live synchronization, and transaction replay, we show that Reth is markedly faster whenever repeated local execution dominates, reducing historical local processing by 3.69× and replay time by 6.87×, while Geth is 2.70× faster end-to-end in live synchronization because verification and canonical commitment lie on the critical path. Second, by replaying 100,000 mainnet blocks and attributing runtime to individual opcodes, we show that this performance gap is overwhelmingly concentrated in read-side state access: SLOAD alone explains 95.8% of the measured inter-client difference, with further asymmetry concentrated in account- and code-lookup opcodes. Third, in a public-view builder case study, we use this diagnosis to reduce how much candidate evaluation reaches the expensive exact-simulation stage: on a frozen test split of 886 blocks, an XGBoost prefilter at top-0.1% ratio to top-2.0% ratio reduces checked candidates by 254,708 to 604,776 relative to an effective-priority-fee baseline at matched Flashbots-hit levels, while speeding the ranking-and-selection stage by about 1.75×. These results show that execution-client divergence is best understood as a critical-path problem: the same semantic interface can expose very different bottlenecks depending on how state access, verification, and workload structure interact.
Chon Kit Lao, Nora Sinong Lu, Jingyi Ning et al.· International Conference on...· 0 citations
Training GNNs on large-scale graphs imposes significant memory constraints for storing substantial amounts of graph structures and node features. This often necessitates the use of memory extensions such as SSDs, leading to a memory hierarchy with disparities in capacity and access speed. Existing approaches focus on mitigating the read amplification of SSDs used as memory extensions to enhance overall performance. However, these methods fail to achieve optimal performance on heterogeneous memory architectures such as DRAM–NVM systems and overlook the efficient utilization of fast memory. In this paper, we propose Malope, an efficient memory-aware and locality-preserved GNN training framework designed for heterogeneous memory systems. First, Malope introduces a memory-aware graph partitioning strategy that preserves multi-hop connectivity and maximizes fast memory utilization. Second, Malope presents a novel locality-preserved GNN training mechanism that reorganizes mini-batches to enhance data locality, thereby improving fast memory hit rates and minimizing partition switching overhead. Additionally, Malope integrates pipelined GNN training and partition switching to minimize data transfer overhead under low bandwidth conditions. Lastly, Malope enables fine-grained model persistence, built on reorganized mini-batch training, for rapid failure recovery. Experimental results on large real-world datasets show that Malope significantly outperforms state-of-the-art GNN training frameworks, achieving an impressive average speedup of <inline-formula><tex-math notation="LaTeX">$1.51\times$</tex-math><alternatives><mml:math><mml:mrow><mml:mn>1</mml:mn><mml:mo>.</mml:mo><mml:mn>51</mml:mn><mml:mo>×</mml:mo></mml:mrow></mml:math><inline-graphic xlink:href="cheng-ieq1-3705364.gif"/></alternatives></inline-formula>.
Junkun Shen, Yuezhi Che, Haoran Zhou et al.· IEEE Transactions on Paralle...· 0 citations
