AnchorTP: Resilient LLM Inference with State-Preserving Elastic Tensor Parallelism

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Wendong, Chen, Chujie, Xiao, He, Li, Kuan, Xiong, Jing, Zhang, Chen, Zhou, Wenyong, Tao, Chaofan, Bai, Yang, Yu, Bei, Wong, Ngai
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912710105497600
author Xu, Wendong
Chen, Chujie
Xiao, He
Li, Kuan
Xiong, Jing
Zhang, Chen
Zhou, Wenyong
Tao, Chaofan
Bai, Yang
Yu, Bei
Wong, Ngai
author_facet Xu, Wendong
Chen, Chujie
Xiao, He
Li, Kuan
Xiong, Jing
Zhang, Chen
Zhou, Wenyong
Tao, Chaofan
Bai, Yang
Yu, Bei
Wong, Ngai
contents Large Language Model (LLM) inference services demand exceptionally high availability and low latency, yet multi-GPU Tensor Parallelism (TP) makes them vulnerable to single-GPU failures. We present AnchorTP, a state-preserving elastic TP framework for fast recovery. It (i) enables Elastic Tensor Parallelism (ETP) with unequal-width partitioning over any number of GPUs and compatibility with Mixture-of-Experts (MoE), and (ii) preserves model parameters and KV caches in GPU memory via a daemon decoupled from the inference process. To minimize downtime, we propose a bandwidth-aware planner based on a Continuous Minimal Migration (CMM) algorithm that minimizes reload bytes under a byte-cost dominance assumption, and an execution scheduler that pipelines P2P transfers with reloads. These components jointly restore service quickly with minimal data movement and without changing service interfaces. In typical failure scenarios, AnchorTP reduces Time to First Success (TFS) by up to 11x and Time to Peak (TTP) by up to 59% versus restart-and-reload.
format Preprint
id arxiv_https___arxiv_org_abs_2511_11617
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AnchorTP: Resilient LLM Inference with State-Preserving Elastic Tensor Parallelism
Xu, Wendong
Chen, Chujie
Xiao, He
Li, Kuan
Xiong, Jing
Zhang, Chen
Zhou, Wenyong
Tao, Chaofan
Bai, Yang
Yu, Bei
Wong, Ngai
Distributed, Parallel, and Cluster Computing
Large Language Model (LLM) inference services demand exceptionally high availability and low latency, yet multi-GPU Tensor Parallelism (TP) makes them vulnerable to single-GPU failures. We present AnchorTP, a state-preserving elastic TP framework for fast recovery. It (i) enables Elastic Tensor Parallelism (ETP) with unequal-width partitioning over any number of GPUs and compatibility with Mixture-of-Experts (MoE), and (ii) preserves model parameters and KV caches in GPU memory via a daemon decoupled from the inference process. To minimize downtime, we propose a bandwidth-aware planner based on a Continuous Minimal Migration (CMM) algorithm that minimizes reload bytes under a byte-cost dominance assumption, and an execution scheduler that pipelines P2P transfers with reloads. These components jointly restore service quickly with minimal data movement and without changing service interfaces. In typical failure scenarios, AnchorTP reduces Time to First Success (TFS) by up to 11x and Time to Peak (TTP) by up to 59% versus restart-and-reload.
title AnchorTP: Resilient LLM Inference with State-Preserving Elastic Tensor Parallelism
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2511.11617