DataCenterGym: A Physics-Grounded Simulator for Multi-Objective Data Center Scheduling

This feed aggregates multiple 2026 arXiv papers across distributed systems, HPC, ML serving, and blockchain. Key items: DataCenterGym — a physics-grounded simulator and H-MPC scheduler for thermo- and power-aware geo-distributed datacenter scheduling; BlockRaFT — a RAFT-based intra-node distributed framework for scalable, crash-tolerant blockchain nodes with a concurrent Merkle-tree optimization; "Accuracy Is Speed" — a study of long-context LLM serving introducing Time-to-Correct-Answer (TTCA) and a routing design (LAAR) that trades accuracy for lower retry-driven latency; cuNNQS-SCI — full-G