Saved in:
Bibliographic Details
Main Authors: Han, Shujie, Jiang, Feng, Lee, Patrick P. C., Zhang, Xiao, Huang, Zhijie, Zhao, Nannan, Zhao, Xiaonan, Pan, Lichen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.17821
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910229846818816
author Han, Shujie
Jiang, Feng
Lee, Patrick P. C.
Zhang, Xiao
Huang, Zhijie
Zhao, Nannan
Zhao, Xiaonan
Pan, Lichen
author_facet Han, Shujie
Jiang, Feng
Lee, Patrick P. C.
Zhang, Xiao
Huang, Zhijie
Zhao, Nannan
Zhao, Xiaonan
Pan, Lichen
contents Large Language Model (LLM) training is frequently interrupted by a heterogeneous spectrum of failures, from common GPU crashes to catastrophic cluster-wide outages. Existing checkpointing systems rely on monolithic, single-tier storage backend, forcing a trade-off between state-saving overhead and recovery speed. We propose TierCheck, a cluster-aware tiered checkpointing system that aligns storage placement with failure heterogeneity. TierCheck adopts a three-tier design that maintains lightweight differential checkpoints in local and peer memory for fast localized recovery, while asynchronously migrating heavyweight base checkpoints to remote persistent storage. It also ensures strict global consistency across tiers without stalling training, and achieves fast cluster-aware checkpoint restoration during recovery. Evaluations on models up to 40 billion parameters show that TierCheck achieves low training overhead, reduces end-to-end checkpointing time to under 10s, and supports high-frequency checkpointing, ultimately striking an optimal balance between low-overhead persistence and fast recovery.
format Preprint
id arxiv_https___arxiv_org_abs_2605_17821
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TierCheck: Tiered Checkpointing for Fault Tolerance in Large Language Model Training
Han, Shujie
Jiang, Feng
Lee, Patrick P. C.
Zhang, Xiao
Huang, Zhijie
Zhao, Nannan
Zhao, Xiaonan
Pan, Lichen
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Large Language Model (LLM) training is frequently interrupted by a heterogeneous spectrum of failures, from common GPU crashes to catastrophic cluster-wide outages. Existing checkpointing systems rely on monolithic, single-tier storage backend, forcing a trade-off between state-saving overhead and recovery speed. We propose TierCheck, a cluster-aware tiered checkpointing system that aligns storage placement with failure heterogeneity. TierCheck adopts a three-tier design that maintains lightweight differential checkpoints in local and peer memory for fast localized recovery, while asynchronously migrating heavyweight base checkpoints to remote persistent storage. It also ensures strict global consistency across tiers without stalling training, and achieves fast cluster-aware checkpoint restoration during recovery. Evaluations on models up to 40 billion parameters show that TierCheck achieves low training overhead, reduces end-to-end checkpointing time to under 10s, and supports high-frequency checkpointing, ultimately striking an optimal balance between low-overhead persistence and fast recovery.
title TierCheck: Tiered Checkpointing for Fault Tolerance in Large Language Model Training
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
url https://arxiv.org/abs/2605.17821