11-13, 16:00–16:30 (Asia/Bangkok), Stage 3
In this session, we introduce how we optimized zkEVM provers in production to significantly reduce prover costs, a major expense in running zkEVM. Topics include diagnosing zkEVM bottlenecks using CPU and memory profiling, leveraging DAGs for parallelization, and efficient memory management with a memory pool, fine-tuned garbage collection, and in-memory swapping for gigantic memory usage. These optimizations reduced zkEVM prover runtime by 75%, representing a substantial performance gain.
Leo Jeong, Ph.D. in CSE, is a Senior Research Engineer at Linea, Consensys. With expertise in zkEVM performance optimization and distributed systems, he has spent the past year diagnosing performance bottlenecks and optimizing zkEVM provers in production. Previously, at Samsung Research, he optimized the performance of their data center and big data platform. Leo also served on the technical program committee of the IEEE International Conference on Blockchain and Cryptocurrency.