This article explains how the three binding options -bsort, -bstart, and -bstop control where and in which order jobs are bound to CPUs in Gridware Cluster Scheduler. Together, they define the scheduler’s fill-up pattern — how hardware resources are used, balanced, or reused across sockets, cores, and NUMA nodes. Understanding these options is essential for…
Tag: NUMA
Understanding Binding in Gridware Cluster Scheduler
Modern compute nodes have grown increasingly complex — featuring heterogeneous cores, multi-level caches, and intricate NUMA topologies. In the previous post, Compute Nodes with Heterogeneous Topology in Gridware Cluster Scheduler, we looked at how these topologies are detected and represented in Gridware Cluster Scheduler. This post explores binding — how the scheduler decides where exactly…
Compute Nodes with Heterogeneous Topology in Gridware Cluster Scheduler
As CPUs evolve toward hybrid designs with mixed core types and increasingly complex memory hierarchies, HPC schedulers must also evolve.This post explains how Gridware Cluster Scheduler 9.1.0 meets that challenge—bringing detailed, topology-aware resource scheduling to modern heterogeneous compute nodes.