arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.28781cs.DCcs.LG
收藏

当GPU沉默失败时:超越数值遥测的可观测性感知早期预警

When GPUs Fail Quietly: Observability-Aware Early Warning Beyond Numeric Telemetry

Michael Bidollahkhani, Freja Nordsiek, Julian M. Kunkel

AI总结:

本文提出一种结合GPU遥测热漂移特征和监控流水线退化指标的早期预警框架,通过结构化遥测崩溃检测提升故障预警时间,数据集已公开。

摘要:

GPU节点是现代HPC和AI工作负载的核心,但许多故障不会立即表现为硬故障。一些不稳定性逐渐显现为弱热或效率漂移,而另一类故障则突然发生,几乎没有数值前兆。在脱离类故障中,GPU在驱动或互连层面不可用,主导的可观察信号是结构性的,包括设备指标消失和监控负载完整性退化。本文提出一种可观测性感知的早期预警框架,联合建模(i)GPU遥测中的利用感知热漂移特征,以及(ii)监控流水线退化指标,如刮取延迟增加、样本丢失、时间序列间隙和设备指标消失。该框架在GWDG生产遥测数据上进行评估,其中GPU、节点、监控和调度器信号可相互关联。结果表明,脱离故障表现出极小的数值前兆,主要通过结构化遥测崩溃观测,而联合建模相比仅GPU检测提升了早期预警时间。本文所用数据集已公开于https://doi.org/10.5281/zenodo.19052367。

英文摘要:

GPU nodes are central to modern HPC and AI workloads, yet many failures do not manifest as immediate hard faults. While some instabilities emerge gradually as weak thermal or efficiency drift, a significant class occurs abruptly with little or no numeric precursor. In these detachment-class failures, GPUs become unavailable at the driver or interconnect level and the dominant observable signal is structural, including disappearance of device metrics and degradation of monitoring payload integrity. This paper proposes an observability-aware early-warning framework that jointly models (i) utilization-aware thermal drift signatures in GPU telemetry and (ii) monitoring-pipeline degradation indicators such as scrape latency increase, sample loss, time-series gaps, and device-metric disappearance. The framework is evaluated on production telemetry from GPU nodes at GWDG, where GPU, node, monitoring, and scheduler signals can be correlated. Results show that detachment failures exhibit minimal numeric precursor and are primarily observable through structural telemetry collapse, while joint modeling increases early-warning lead time compared to GPU-only detection. The dataset used in this study is publicly available at https://doi.org/10.5281/zenodo.19052367.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑