Scholarlink

<p>The rapid growth of AI models has imposed increasingly stringent performance demands on data centers. Modern data centers adopt Remote Direct Memory Access (RDMA) to reduce CPU overhead and network latency. RDMA operates over a lossless network, and RDMA congestion control (CC) protocols are key enablers for achieving low-latency and high-throughput data delivery. Through in-depth experiment analysis, we reveal that existing RDMA CC protocols still suffer from sluggish congestion response and convergence speed, limiting their ability to support latency-sensitive AI applications. In this paper, we propose a switch-driven CC algorithm named FACC. FACC enables switches to precisely identify flows that actually cause congestion and promptly notifies the congestion information to senders. At the sender, FACC leverages the intrinsic packet conservation property of a lossless network to assess the extent of network congestion. Then, FACC employs a PI controller to adaptively adjust the sending rate, thereby achieving rapid congestion elimination and improving both the transmission rate and convergence speed. We conduct extensive experiments to evaluate the performance of FACC. The results show that FACC improves convergence speed while achieving at most 86.6% lower flow completion time (FCT) compared with state-of-the-art approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Rate-adaptive RDMA congestion control for AI clusters

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月29日

查看全文
  • Xin He,
  • Zihao Zhang,
  • Junchang Wang,
  • Zheng Wu,
  • Weibei Fan,
  • Yiping Zuo

摘要

The rapid growth of AI models has imposed increasingly stringent performance demands on data centers. Modern data centers adopt Remote Direct Memory Access (RDMA) to reduce CPU overhead and network latency. RDMA operates over a lossless network, and RDMA congestion control (CC) protocols are key enablers for achieving low-latency and high-throughput data delivery. Through in-depth experiment analysis, we reveal that existing RDMA CC protocols still suffer from sluggish congestion response and convergence speed, limiting their ability to support latency-sensitive AI applications. In this paper, we propose a switch-driven CC algorithm named FACC. FACC enables switches to precisely identify flows that actually cause congestion and promptly notifies the congestion information to senders. At the sender, FACC leverages the intrinsic packet conservation property of a lossless network to assess the extent of network congestion. Then, FACC employs a PI controller to adaptively adjust the sending rate, thereby achieving rapid congestion elimination and improving both the transmission rate and convergence speed. We conduct extensive experiments to evaluate the performance of FACC. The results show that FACC improves convergence speed while achieving at most 86.6% lower flow completion time (FCT) compared with state-of-the-art approaches.