Guard: Scalable Straggler Detection and Node Health Management for Large-Scale Training

2026-05-18

AI Summary

This work addresses the challenge of detecting “slow faults”—subtle node degradations that silently impair performance but evade conventional health checks in large-scale model training. The authors propose a novel health management system that synergistically combines lightweight online performance monitoring with offline, systematic node scanning to jointly identify both abrupt failures and long-term slow faults. By leveraging runtime performance tracing, FLOPs utilization analysis, and exhaustive node-level stress testing, the approach substantially enhances system observability and scalability. Experimental results demonstrate that deployment of this method increases peak FLOPs utilization by up to 1.7×, reduces training step duration variance from 20% to 1%, and effectively extends mean time between failures while lowering operational overhead.

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 1.7x, 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.

Problem

• straggler detection
• node health management
• fail-slow
• large-scale training
• performance degradation

Innovation

• straggler detection
• node health management
• fail-slow
• large-scale training
• performance monitoring