Linux and High-Performance Computing
2026-03-25T08:52:24Z•8ac36e05bf0571a4fe935ebb417edde6b9e7993dd640e2a9441c848bc3f8b2f2
Arena training systemByzantine/rational consensusCFS bandwidth controllerCPU-induced bottlenecksGPU/CPU provisioningKV-cache reuseKubernetesLLM inferenceLinux historyMPIRAG servingSNAREcloud (AWS EC2)dynamic load balancingenergy-aware pruningfederated learninggradient codinghigh-performance computinginteractive HPCperformance engineeringprefetchingrank-aware schedulingrequests-only allocation
What happened
Collection of recent arXiv papers (HPC and ML systems) covering: historical analysis of Linux-based HPC (Beowulf vs Roadrunner); trends toward interactive/urgent HPC; rank-aware MPI resource scheduling on Kubernetes (mapping ranks to pods with proportional CPU requests; warns that Linux CFS hard limits cause massive slowdowns; recommends requests-only allocation; experiments on AWS c5.xlarge, k3s v1.35); PCR, a prefetch-enhanced KV-cache reuse system for low-latency RAG serving (prefix-tree cache, look-ahead LRU, layer-wise overlapping, queue-based SSD→DRAM prefetching; up to 2.47x TTFT speed‑
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_dc
- Record identifier
- 8ac36e05bf0571a4fe935ebb417edde6b9e7993dd640e2a9441c848bc3f8b2f2
- Enrichment time
- 2026-03-25T08:52:24Z
- 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.