Replication in Graph Partitioning and Scheduling Problems

2026-05-04T08:52:21Zb427a7a2b01980de36d1ee67685b4307dde72857067e858fdec74504d65bc0a3
AIE-MLDAG-schedulingGEMMGPU-clustersIPUKV-cacheLLM-Emu','serving-emulation','profile-driven-sampling' , 'emulatLLM-servingMixture-of-ExpertsMoEPopdistPoplar-SDKSAGATempusVersal-AI-Edgeapproximation-complexitycomputational-fluid-dynamicsedge-inferenceexpert-placementgraph-partitioninginteger-linear-programmingreplicationsatellite-constellationspace-data-centersworkflow-scheduling

What happened

Collection of recent systems and algorithms papers (arXiv May 2026) focusing on distributed ML/AI execution, hardware acceleration, and resource-placement/scheduling tradeoffs. Key contributions include: theoretical and empirical analysis of replication for graph partitioning and DAG scheduling (showing large cost reductions but increased approximation hardness); adapting ML-accelerated CFD training to Graphcore IPUs (Poplar/Popdist optimizations); Space-XNet for two-level placement of MoE layers/experts across satellite constellations to cut inference latency; SAGA, a workflow-atomic GPU-club

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_dc
Record identifier
b427a7a2b01980de36d1ee67685b4307dde72857067e858fdec74504d65bc0a3
Enrichment time
2026-05-04T08:52:21Z
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.