We've just released Open Cluster Scheduler 9.1.0 and Gridware Cluster Scheduler 9.1.0. This is a big one.
The headline feature is a completely redesigned topology-aware binding framework. Binding is now scheduler-driven -- cores, sockets, dies, L2/L3 caches, and NUMA domains are treated as consumable resources and assigned during scheduling, not after the fact by execution daemons. The new -b... option family gives you precise control over placement, including support for hybrid CPUs with performance and efficiency cores. If your jobs are sensitive to cache locality or memory bandwidth, this will make a real difference.
On the security side, TLS encryption is now available for all internal component communication with automatic certificate management. Munge authentication has been added for containerized and user-namespace environments. And a new DoS protection mechanism rate-limits GDI requests to keep qmaster stable under pressure.
qtelemetry provides native Prometheus and Grafana metrics export -- host metrics, job metrics, qmaster internals -- with a pre-configured Grafana dashboard ready to go.
Full systemd integration lands in this release: daemons run as native services, jobs can execute under systemd control with cgroup-based accounting and device isolation, and usage collection is configurable between systemd, the built-in collector, or both.
Other highlights include qalter -when now to decrease resources of running jobs (free a license mid-job), Department View for multi-tenancy, a faulty job loadsensor for automatic archival of failed job data, and extensible JSON accounting with custom metrics support.
Gridware Cluster Scheduler adds enterprise-only features on top: Qontrol (a modern web UI for cluster configuration), qgpu enhancements for NVIDIA GPU management via DCGM, FlexNet license-manager integration, qtelemetry, TLS encryption, and DoS protection. Platform support spans x86-64, ARM64, ppc64le, s390x, and RISC-V64 across a wide range of distributions, with SLES 15 SP7 certification.
Open Cluster Scheduler source code: github.com/hpc-gridware/clusterscheduler
Open Cluster Scheduler pre-compiled packages: hpc-gridware.com/download-main
To trial Gridware Cluster Scheduler, contact HPC-Gridware directly through their website or by email.
Full release blog post: hpc-gridware.com/hpc-gridware-releases-gridware-cluster-scheduler-9-1-0-and-open-cluster-scheduler-9-1-0