English

Fault-tolerant Reduce and Allreduce operations based on correction

Distributed, Parallel, and Cluster Computing 2026-02-27 v1

Abstract

Implementations of Broadcast based on some information dissemination algorithm -- e.g., gossip or tree-based communication -- followed by a correction algorithm has been proposed previously. This work describes an approach to apply a similar idea to Reduce. In it, a correction-like communication phase precedes a tree-based phase. This provides a Reduce algorithm which is tolerant to a number of failed processes. Semantics of the resulting algorithm are provided and proven. Based on these results, Broadcast and Reduce are combined to provide Allreduce.

Keywords

Cite

@article{arxiv.2602.22445,
  title  = {Fault-tolerant Reduce and Allreduce operations based on correction},
  author = {Martin Kuettler and Hermann Haertig},
  journal= {arXiv preprint arXiv:2602.22445},
  year   = {2026}
}
R2 v1 2026-07-01T10:53:02.475Z