GUARD: SCALABLE STRAGGLER DETECTION AND NODE HEALTH MANAGEMENT FOR LARGE-SCALE TRAINING

Guanliang Liu, Abhinandan Patni, congzhu lin, Zoe zeng, Jack Wittmayer, yinghong liu, josh wu, Anthony Ko, Alexander Zhipa, Ashvin Nihalani, Binxuan Huang, Cong cheng, Mi Sun, Vijay Rajakumar, Rejith George Joseph, Parthasarathy Govindarajen

Keywords: Large-scale, distributed training

Abstract:

Training frontier-scale foundation models involves coordinating tens of thousands of GPUs over multi-month
runs, where even minor performance degradations can accumulate into substantial efficiency losses. Existing
health-check mechanisms, such as NCCL tests or GPU burn-in, primarily focus on functional correctness and
often fail to detect fail-slow behaviors that silently degrade system performance. In this paper, we present
Guard, a scalable system for detecting stragglers and ensuring node health in large-scale training clusters. Guard
combines lightweight online performance monitoring during training with an offline node-sweep mechanism that
systematically evaluates and qualifies nodes before they participate in production workloads. This design enables
Guard to detect both acute failures and long-running fail-slow behaviors that traditional diagnostics cannot capture.
Deployed on large-scale foundation model pretraining workloads, Guard improves mean FLOPs utilization by
up to 1.7×, reduces run-to-run training step variance from 20% to 1%, increases mean time to failure (MTTF),
and significantly reduces operational and debugging overhead. These results demonstrate that proactive straggler
detection and systematic node qualification are critical for maintaining stable and efficient large-scale training.

Topics: Model Training: Large-scale, distributed ML and RL training

Submission Number: 98