Comparing Distributed Termination Detection Algorithms for Modern HPC Platforms

This paper revisits distributed termination detection algorithms in the context of High-Performance Computing (HPC) applications. We introduce an efficient variant Credit Distribution Algorithm (CDA) and compare it to original algorithm (HCDA) as well its two primary competitors: Four Counters (4C) Efficient Delay-Optimal Distributed (EDOD). analyze behavior each for some simplified task-based kernels show superiority CDA terms number control messages. then implementation these over a runtime system, PaRSEC advantages limitations approach real implementation.