Cedar is a heterogenous cluster, having 640 `skylake` nodes (192 GiB of memory, 24 cores/socket, 2 sockets/node), 752 `broadwell` nodes (minimum 128 GiB of memory, 16 cores/socket, 2 sockets/node), and 584 GPU devices.
- Group members can run their simulation on Cedar instead of own desktop, to have:
- simulations completed sooner thanks to parallelization;
- multiple simulations running at the same time;
- desktop free for other tasks.
Scheduler package for Cedar
- CedarScheduler is a python helper package for Cedar, developed in order to:
- simplify the procedure of submitting multiple similar jobs;
- speed up simulation.
The package, documentation and usage examples are located at arggit.
Installation and Usage
CedarScheduler package must be installed under your python environment on Cedar.
The steering file c_scheduler_config.yaml must be copied to home directory on Cedar.
Each job must be submitted in separate directory.
The examples of job setup are given in job_example folder:
A Web-Interface allowing to view Cedar statistics, submit new and monitor pending jobs is located at arggit.