He will also demonstrate Bright's unique data aware scheduling capability for the Cloud, eliminating the need to manually manage data movement. The combined Bright Cluster Manager and open source SLURM workload manager deliver a seamless, intuitive solution for provisioning, scheduling, monitoring and managing the extended cluster and data within one intuitive GUI and cluster management shell, on-premise or in the Cloud.
"Bright Cluster Manager is delivered with SLURM as a pre-configured, sys admin-selectable option", stated Martijn de Vries. "Bright's tight integration of SLURM, and other workload managers, makes it easy to get the most productivity from clusters, whether on premise or in the Cloud."
When the system administrator selects SLURM as the workload manager of choice, Bright automatically installs and configures the software. Bright continually updates configurations throughout the life of the cluster. The system administrator can then take full advantage of the workload manager's capabilities using the Bright GUI or cluster management shell, without the need to learn additional commands or procedures. Further, SLURM can be managed via Bright's SOAP API. Both Bright Cluster Manager and SLURM can be extended into the Cloud to provide additional capacity, with just a few mouse clicks within the Bright GUI.
The tight integration of Bright's health checking capabilities with SLURM also provides protection against the "Black Hole Node Syndrome", when normally undetected node issues cause job crashes, sometimes extending to full job queue flushes. Working closely with the workload manager, Bright's pre-job health checks detect system problems before nodes actually fail, and then sidelines these nodes before the job is started.
As another means of maintaining high throughput, Bright Cluster Managers automatic failover capability also manages the seamless failover of the workload manager, preventing head node crashes from interrupting productivity.
Martijn de Vries' presentation is at 11:00 AM on Tuesday, October 9th. Also presenting at the conference are several HPC experts from large national laboratories, research universities, and the leading SLURM developers.
The meeting is a two-day event for SLURM users and contributors and other HPC experts, to be held on October 9th and 10th at the Barcelona Supercomputing Center. More information is available at www.bsc.es/marenostrum-support-services/hpc-events-trainings/others/slurm-user-group-meeting-2012