Learning In Chaos: Efficient Autoscaling and Self-Healing for Multi-Party Distributed Training

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Feng, Wenjiao, Xiao, Rongxing, Li, Zonghang, Yu, Hongfang, Sun, Gang, Luo, Long, Guizani, Mohsen, Ho, Qirong, Liu, Steve
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918140270608384
author Feng, Wenjiao
Xiao, Rongxing
Li, Zonghang
Yu, Hongfang
Sun, Gang
Luo, Long
Guizani, Mohsen
Ho, Qirong
Liu, Steve
author_facet Feng, Wenjiao
Xiao, Rongxing
Li, Zonghang
Yu, Hongfang
Sun, Gang
Luo, Long
Guizani, Mohsen
Ho, Qirong
Liu, Steve
contents Node and link churn in multi-party, cross-region clusters over wide-area networks (WANs) often disrupts distributed training. However, checkpoint-based recovery and cloud-centric autoscaling react slowly and assume centralized control, which is misaligned with the self-governed setup where institutions can freely join and leave. This paper proposes Chaos, a multi-party distributed training system with self-healing and autoscaling, enabling robust and elastic training under churn. It speeds up autoscaling via multi-neighbor state replication and model sharding. We formalize the sharding and assignment as a MINLP that captures WAN heterogeneity, and reduce it to a tractable MILP by analyzing its monotonicity on a divisibility chain. By establishing an equivalence, we derive a greedy algorithm that follows optimality rules and yields the optimal solution in polynomial time. Chaos uses a cluster monitor to track resource and topology changes, and handles scaling events through peer negotiation protocols, enabling fully self-governed autoscaling among institutions. Experiments show that Chaos has substantially lower scale-out delay than Pollux, Elan, and Autoscaling, and handles scale-in, connect-link, and disconnect-link events within 20ms. It also delivers the lowest idle time, showing superior resource use and scalability as the cluster grows.
format Preprint
id arxiv_https___arxiv_org_abs_2505_12815
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning In Chaos: Efficient Autoscaling and Self-Healing for Multi-Party Distributed Training
Feng, Wenjiao
Xiao, Rongxing
Li, Zonghang
Yu, Hongfang
Sun, Gang
Luo, Long
Guizani, Mohsen
Ho, Qirong
Liu, Steve
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
68T99
I.2.11
Node and link churn in multi-party, cross-region clusters over wide-area networks (WANs) often disrupts distributed training. However, checkpoint-based recovery and cloud-centric autoscaling react slowly and assume centralized control, which is misaligned with the self-governed setup where institutions can freely join and leave. This paper proposes Chaos, a multi-party distributed training system with self-healing and autoscaling, enabling robust and elastic training under churn. It speeds up autoscaling via multi-neighbor state replication and model sharding. We formalize the sharding and assignment as a MINLP that captures WAN heterogeneity, and reduce it to a tractable MILP by analyzing its monotonicity on a divisibility chain. By establishing an equivalence, we derive a greedy algorithm that follows optimality rules and yields the optimal solution in polynomial time. Chaos uses a cluster monitor to track resource and topology changes, and handles scaling events through peer negotiation protocols, enabling fully self-governed autoscaling among institutions. Experiments show that Chaos has substantially lower scale-out delay than Pollux, Elan, and Autoscaling, and handles scale-in, connect-link, and disconnect-link events within 20ms. It also delivers the lowest idle time, showing superior resource use and scalability as the cluster grows.
title Learning In Chaos: Efficient Autoscaling and Self-Healing for Multi-Party Distributed Training
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
68T99
I.2.11
url https://arxiv.org/abs/2505.12815