Flex-MIG: Enabling Distributed Execution on MIG

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Myeongsu, Yeom, Ikjun, Kim, Younghoon
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917077847113728
author Kim, Myeongsu
Yeom, Ikjun
Kim, Younghoon
author_facet Kim, Myeongsu
Yeom, Ikjun
Kim, Younghoon
contents GPU clusters in multi-tenant settings often suffer from underutilization, making GPU-sharing technologies essential for efficient resource use. Among them, NVIDIA Multi-Instance GPU (MIG) has gained traction for providing hardware-level isolation that enables concurrent workloads without interference. However, MIG's hardware rigidity and the conventional one-to-one allocation model jointly lead to severe fragmentation and cluster-wide underutilization. We present Flex-MIG, a software-only framework that replaces one-to-one with a one-to-many allocation model and enables host-shared-memory collectives across MIG instances without hardware modification. Flex-MIG eliminates drain-required reconfiguration, reduces fragmentation, and improves makespan by up to 17% across diverse traces, showing that rethinking MIG's operational model as a software-coordinated layer substantially improves cluster efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2511_09143
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Flex-MIG: Enabling Distributed Execution on MIG
Kim, Myeongsu
Yeom, Ikjun
Kim, Younghoon
Distributed, Parallel, and Cluster Computing
D.4.7; C.1.4
GPU clusters in multi-tenant settings often suffer from underutilization, making GPU-sharing technologies essential for efficient resource use. Among them, NVIDIA Multi-Instance GPU (MIG) has gained traction for providing hardware-level isolation that enables concurrent workloads without interference. However, MIG's hardware rigidity and the conventional one-to-one allocation model jointly lead to severe fragmentation and cluster-wide underutilization. We present Flex-MIG, a software-only framework that replaces one-to-one with a one-to-many allocation model and enables host-shared-memory collectives across MIG instances without hardware modification. Flex-MIG eliminates drain-required reconfiguration, reduces fragmentation, and improves makespan by up to 17% across diverse traces, showing that rethinking MIG's operational model as a software-coordinated layer substantially improves cluster efficiency.
title Flex-MIG: Enabling Distributed Execution on MIG
topic Distributed, Parallel, and Cluster Computing
D.4.7; C.1.4
url https://arxiv.org/abs/2511.09143