arXiv:2606.18170 · v2
The Multipath Reliable Connection (MRC) Transport
Rip SohanEric SpadaEric DavisMark HandleyIdan BursteinTony HursonJithin JoseVivek KashyapRong PanSayantan SurSreevatsa AnantharamuAviv BarneaAdrian CaulfieldElazar CohenElliot EdmundsYamin FriedmanMahdieh GhaziMurali GarimellaTorsten HoeflerVipin JainAbdul KabbaniNoam KatzYanfang LeCharlie MbarikyGuglielmo MorandinMasoud MoshrefShane O'NeilMichael PapamichaelJonas PfefferleSiva Santosh PylaCostin RaiciuDavid RiddochKaren SchrammYuval ShpigelmanShahaf ShulerShy ShymanRaghava SivaramuAmin TootoonchianYang Wang
首次提交 2026-06-17 01:07
版本更新 2026-11-04 08:46
雷达观测 2026-11-05 11:11
摘要
MRC is an open, production-grade transport designed for large-scale AI/ML training over best-effort Ethernet. It extends RoCEv2 with explicit, composable primitives for per-packet multipath and sender-based congestion control, decouples packet delivery from semantic processing, adds multiple new capabilities for accelerated packet-loss recovery and adds resilience against port and path failures. This paper presents MRC and details its core capabilities and mechanisms.