When Scaling Fails: Network and Fabric Effects on Distributed GPU Training Performance

2026-03-06T07:23:53Zf091a4f6a685728855241da77b0dad1d6ede3c98daa82eb44fc16c878ee3085e
DRL auction manipulationEnvoy external processorLLM routingO-CUO-DUO-RANO-RUPII leakageRI S allocationadversarial MLautomotive/avionics safetycellular securitycongestion dynamicsdenial-of-servicedistributed GPU traininginference privacyjailbreak detectionlegacy 2G/3Gmessage groupingmulti-cloud routingnetwork fabricnetwork topologysynchronization amplificationtime-triggered schedulingvLLM Semantic Router

What happened

This collection of arXiv papers surveys systems research across distributed GPU training, cellular/O-RAN networks, connected-vehicle traffic prediction, RIS allocation with DRL, deterministic time-triggered scheduling, green mesh wireless architectures, and multi-model routing for LLMs (vLLM Semantic Router). Common themes with security relevance include expanded attack surfaces from disaggregation and multi-provider deployments (O-RAN, vLLM multi-cloud routing, Envoy external processors), legacy/under‑maintained infrastructure (2G/3G towers), and subtle network/fabric effects (topology, Conga

Why it matters

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

Evidence and limitations

Source ID
arxiv_cs_ni
Record identifier
f091a4f6a685728855241da77b0dad1d6ede3c98daa82eb44fc16c878ee3085e
Enrichment time
2026-03-06T07:23:53Z
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.