MCCL: Measurement-based Collective Communication Library
Proceedings of the ACM on Networking · 19.11.2026 · 4 (CoNEXT4) · 1-25
DOI: 10.1145/3856969 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the ACM on Networking, online erschienen am 19.11.2026. Verfasst von Michele Ferrero und 5 weiteren Autor:innen. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Collective communication is a critical performance bottleneck in large-scale distributed training. Existing libraries and synthesis frameworks compute open-loop schedules once, assuming a static topology and peak link capacities. In production, however, persistent gray failures can silently degrade link throughput and latency for hours, causing pre-computed schedules to lose much of their efficiency during execution. In this paper, we present Measurement-based Collective Communication Library (MCCL), a collective communication library that closes the loop between runtime network measurement and dynamic schedule synthesis. MCCL passively measures GPU-to-GPU throughput, detects persistent degradation, re-synthesizes the affected communication-intensive schedules with warm-started synthesizers, and hot-swaps them without interrupting training. We validate the complete loop on a multi-node hardware prototype and use simulations to evaluate cluster-scale scenarios. In our experiments, MCCL incurs negligible overhead (
Zitieren
APA
Ferrero, M., Castellano, G., Medagliani, P., Martin, S., Morabito, R., & Gallo, M. (2026). MCCL: Measurement-based Collective Communication Library. Proceedings of the ACM on Networking, 4(CoNEXT4), 1-25. https://doi.org/10.1145/3856969
Vancouver
Ferrero M, Castellano G, Medagliani P, Martin S, Morabito R, Gallo M. MCCL: Measurement-based Collective Communication Library. Proceedings of the ACM on Networking. 2026;4(CoNEXT4):1-25. doi:10.1145/3856969
BibTeX
@article{Ferrero2026,
author = {Ferrero, Michele and Castellano, Gabriele and Medagliani, Paolo and Martin, Sebastien and Morabito, Roberto and Gallo, Massimo},
title = {MCCL: Measurement-based Collective Communication Library},
journal = {Proceedings of the ACM on Networking},
year = {2026},
volume = {4},
number = {CoNEXT4},
pages = {1--25},
doi = {10.1145/3856969}
}
EUVIMED