Saved in:
Bibliographic Details
Main Authors: Wu, Junchi, Wan, Xinfei, Li, Zhuoran, Jin, Yuyang, Sun, Guangyu, Liang, Yun, Zhou, Diyu, Zhuo, Youwei
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.24112
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915815104708608
author Wu, Junchi
Wan, Xinfei
Li, Zhuoran
Jin, Yuyang
Sun, Guangyu
Liang, Yun
Zhou, Diyu
Zhuo, Youwei
author_facet Wu, Junchi
Wan, Xinfei
Li, Zhuoran
Jin, Yuyang
Sun, Guangyu
Liang, Yun
Zhou, Diyu
Zhuo, Youwei
contents Many-core accelerators are essential for high-performance deep learning, but their performance is undermined by widespread fail-slow failures. Detecting such failures on-chip is challenging, as prior methods from distributed systems are unsuitable due to strict memory limits and their inability to track failures across the hardware topology. This paper introduces SLOTH, a lightweight, hardware-aware framework for practical on-chip fail-slow detection in many-core accelerators. SLOTH combines workload-aware instrumentation for operator-level monitoring with minimal overhead, on-the-fly trace compression to operate within kilobytes of memory, and a novel topology-aware ranking algorithm to pinpoint a failure's root cause. We evaluate SLOTH on a wide range of representative DNN workloads. The results demonstrate that SLOTH reduces the storage overhead by an average of 115.9$\times$, while achieving an average fail-slow detection accuracy of 86.77% and a false positive rate (FPR) of 12.11%. More importantly, SLOTH scales effectively across different many-core accelerator architectures, making it practical for large-scale deployments.
format Preprint
id arxiv_https___arxiv_org_abs_2510_24112
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards Efficient and Accurate Detection of On-Chip Fail-Slow Failures for Many-Core Accelerators
Wu, Junchi
Wan, Xinfei
Li, Zhuoran
Jin, Yuyang
Sun, Guangyu
Liang, Yun
Zhou, Diyu
Zhuo, Youwei
Hardware Architecture
Many-core accelerators are essential for high-performance deep learning, but their performance is undermined by widespread fail-slow failures. Detecting such failures on-chip is challenging, as prior methods from distributed systems are unsuitable due to strict memory limits and their inability to track failures across the hardware topology. This paper introduces SLOTH, a lightweight, hardware-aware framework for practical on-chip fail-slow detection in many-core accelerators. SLOTH combines workload-aware instrumentation for operator-level monitoring with minimal overhead, on-the-fly trace compression to operate within kilobytes of memory, and a novel topology-aware ranking algorithm to pinpoint a failure's root cause. We evaluate SLOTH on a wide range of representative DNN workloads. The results demonstrate that SLOTH reduces the storage overhead by an average of 115.9$\times$, while achieving an average fail-slow detection accuracy of 86.77% and a false positive rate (FPR) of 12.11%. More importantly, SLOTH scales effectively across different many-core accelerator architectures, making it practical for large-scale deployments.
title Towards Efficient and Accurate Detection of On-Chip Fail-Slow Failures for Many-Core Accelerators
topic Hardware Architecture
url https://arxiv.org/abs/2510.24112