Exascale PMI on a heterogeneous sub-exascale Slurm cluster
summary
Describes deploying PMIx alongside Slurm on a VUB tier-2 HPC cluster that mixes TCP and InfiniBand, explaining how PMIx speeds up launching and controlling distributed tasks compared with traditional MPI bootstrapping. Covers practical lessons and deployment tips for using open-source PMIx with Slurm to achieve low launch latency, small data footprint, and tighter resource-manager integration on a heterogeneous sub-exascale system.
AI-written summary · about
related talks
41mFrom Hardware to Advantage: Hybrid Infrastructure as Enabler for Industry Adoption
49mMicroservices on InfiniBand: 800 Gbps in Distributed Training
38mMPI Meets Machine Learning: Unlocking PyTorch distributed for scaling AI workloads
16mTowards Optimal Rack-scale μs-level CPU Scheduling through In-Network Workload...
25mScalable IT Workload Automation
20mAdvancing HPC The Hybrid Scheduler Approach at the D.E.Shaw Group
44mBoosting your applications with distributed caches/datagrids
10mMPI Meets Python: How to Use MPI in the Python Universe