References & Citations
Computer Science > Distributed, Parallel, and Cluster Computing
Title: Efficient All-to-All Collective Communication Schedules for Direct-Connect Topologies
(Submitted on 24 Sep 2023 (v1), last revised 25 Apr 2024 (this version, v2))
Abstract: The all-to-all collective communications primitive is widely used in machine learning (ML) and high performance computing (HPC) workloads, and optimizing its performance is of interest to both ML and HPC communities. All-to-all is a particularly challenging workload that can severely strain the underlying interconnect bandwidth at scale. This paper takes a holistic approach to optimize the performance of all-to-all collective communications on supercomputer-scale direct-connect interconnects. We address several algorithmic and practical challenges in developing efficient and bandwidth-optimal all-to-all schedules for any topology and lowering the schedules to various runtimes and interconnect technologies. We also propose a novel topology that delivers near-optimal all-to-all performance.
Submission history
From: Liangyu Zhao [view email][v1] Sun, 24 Sep 2023 03:56:35 GMT (5158kb,D)
[v2] Thu, 25 Apr 2024 20:40:57 GMT (7353kb,D)
Link back to: arXiv, form interface, contact.