OptCCL: Scalable Synthesis of Optimal Collective Communication Algorithms
Richard Shapley, Rachit Agarwal, David B. Shmoys
2026年份
摘要
We present OptCCL, a technique to synthesize collective communication algorithms that are optimal for a given host and network hardware and topology. OptCCL is general: it enables synthesizing optimal algorithms for all existing collectives, for all existing hardware, and even for multiple concurrent collectives sharing host and network resources. And yet, OptCCL is scalable: it synthesizes optimal algorithms for hundreds of GPUs within tens of minutes.
问问这篇 Paper
问问你的智能体。
Lune 读过与它相关的顶会 Paper,每个回答都会注明依据哪几篇。
相关 Paper
- Synthesizing optimal collective algorithmsZixian Cai, Zhengyang Liu, Saeed Maleki, Madanlal Musuvathi 等PPoPP 2021 · 被引用 64 次
- TACCL: Guiding Collective Algorithm Synthesis using Communication SketchesAashaka Shah, Vijay Chidambaram, Meghan Cowan, Saeed Maleki 等NSDI 2023
- SyCCL: Exploiting Symmetry for Efficient Collective Communication SchedulingJiamin Cao, Shangfeng Shi, Jiaqi Gao, Weisen Liu 等SIGCOMM 2025 · 被引用 15 次
- MSCCLang: Microsoft Collective Communication LanguageMeghan Cowan, Saeed Maleki, Madanlal Musuvathi, Olli Saarikivi 等ASPLOS 2023 · 被引用 40 次
- TCCL: Discovering Better Communication Paths for PCIe GPU ClustersHeehoon Kim, Junyeol Ryu, Jaejin LeeASPLOS 2024 · 被引用 26 次
