BandPilot: Towards Performance- and Contention-Aware GPU Dispatching in AI Clusters

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Kunming, Liao, Hanlong, Xue, Junyu, Guo, Deke, Tang, Guoming
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915710452629504
author Zhang, Kunming
Liao, Hanlong
Xue, Junyu
Guo, Deke
Tang, Guoming
author_facet Zhang, Kunming
Liao, Hanlong
Xue, Junyu
Guo, Deke
Tang, Guoming
contents Modern multi-tenant AI clusters are increasingly communication-bound, driven by high-volume and multi-round GPU-to-GPU collective communication. Consequently, the GPU dispatcher's choice of a physical GPU subset for each tenant largely determines the job's effective collective bandwidth and thus its performance ceiling. Existing dispatchers predominantly rely on static, topology-aware heuristics that prioritize GPU resource compactness, assuming that minimizing physical distance maximizes communication bandwidth. However, we reveal that this assumption often fails due to complex system-level bottlenecks, such as non-linear NIC saturation and inter-node link heterogeneity.This paper presents BandPilot, a performance- and contention-aware GPU dispatching primitive that optimizes effective collective bandwidth for multi-tenant AI clusters. Specifically, BandPilot learns a data-efficient bandwidth model from sparse NCCL measurements via a hierarchical design. Guided by the model, a fast hybrid search combines an equilibrium-driven constructor with a pruned elimination search to navigate the combinatorial allocation space in real time. To account for multi-tenant interference, BandPilot virtually merges a candidate allocation with co-located cross-host jobs to conservatively estimate shared bottleneck capacity and predict contention-degraded bandwidth. Across a 32-GPU H100 cluster and heterogeneous simulations, BandPilot achieves 92-97% bandwidth efficiency relative to the best-found reference, improving average efficiency by 20-40% over topology-compactness heuristics.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15595
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BandPilot: Towards Performance- and Contention-Aware GPU Dispatching in AI Clusters
Zhang, Kunming
Liao, Hanlong
Xue, Junyu
Guo, Deke
Tang, Guoming
Distributed, Parallel, and Cluster Computing
Modern multi-tenant AI clusters are increasingly communication-bound, driven by high-volume and multi-round GPU-to-GPU collective communication. Consequently, the GPU dispatcher's choice of a physical GPU subset for each tenant largely determines the job's effective collective bandwidth and thus its performance ceiling. Existing dispatchers predominantly rely on static, topology-aware heuristics that prioritize GPU resource compactness, assuming that minimizing physical distance maximizes communication bandwidth. However, we reveal that this assumption often fails due to complex system-level bottlenecks, such as non-linear NIC saturation and inter-node link heterogeneity.This paper presents BandPilot, a performance- and contention-aware GPU dispatching primitive that optimizes effective collective bandwidth for multi-tenant AI clusters. Specifically, BandPilot learns a data-efficient bandwidth model from sparse NCCL measurements via a hierarchical design. Guided by the model, a fast hybrid search combines an equilibrium-driven constructor with a pruned elimination search to navigate the combinatorial allocation space in real time. To account for multi-tenant interference, BandPilot virtually merges a candidate allocation with co-located cross-host jobs to conservatively estimate shared bottleneck capacity and predict contention-degraded bandwidth. Across a 32-GPU H100 cluster and heterogeneous simulations, BandPilot achieves 92-97% bandwidth efficiency relative to the best-found reference, improving average efficiency by 20-40% over topology-compactness heuristics.
title BandPilot: Towards Performance- and Contention-Aware GPU Dispatching in AI Clusters
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2506.15595