Dynamic Hierarchical Birkhoff-von Neumann Decomposition for All-to-All GPU Communication
arXiv 2602.22756•1e9870ff6b0e7a09bacc1cdc1c92575e520062350def19503bd83df880e40d15
Birkhoff-von NeumannGPU-fabricsLLM trainingNICRDMAall-to-all communicationavailabilitycross-NIC failoverdistributed trainingdynamic frame sizingexactly-oncefault-tolerancekernel-bypassperformancerdma-corereceiver-opacityresilienceschedulingsystem-designzero-copy
Paper metadata
- arXiv ID
- 2602.22756
- Version
- Not specified by this published record
- Category
- Computer Science — Networking and Internet Architecture (cs.NI)
The PDF link points to arxiv.org. Baitaphish does not expose a private stored PDF.
Evidence and limitations
- Source ID
- arxiv_cs_ni
- Record identifier
- 1e9870ff6b0e7a09bacc1cdc1c92575e520062350def19503bd83df880e40d15
- Enrichment time
- 2026-03-04T19:51:06Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.