The cluster is capable to run all workload types (single core, SMP, MPI, HTC) and evolves in a continuous upgrade model.
Hydra is a cluster being continuously upgraded. Traditional clusters are purchased in one go and remain fixed for their lifetime (generally 4-5 years) until replaced by a new one. However, research is changing and evolving all the time. We have and will therefore invest regularly in Hydra to extend or adapt the cluster on a yearly basis to better meet users' needs.
The cluster is composed of about 160 compute nodes, covering several CPU generations. Part of the cluster has Infiniband network for MPI jobs. A GPFS storage of ~800 TB offers fast and large capacities for IO demanding jobs. There are nodes with 16, 20 or 24 cores and fitted with memory capacity from 64 GB to 1.5 TB of RAM. Several master nodes run the core services such as Moab, Torque or the LDAP servers.
Hydra is freely accessible to all VUB and ULB users. You need to activate your NetID for Hydra before being able to access it. See our documentation for more details. We impose limitations on the number and type of jobs that can be executed since this is a shared environment and therefore everyone need a chance to get jobs started.
|Number of running jobs||Maximum 2500 jobs|
|Number of concurrent usable cores||From 200 to 1000|
|Number of cores per MPI job||One MPI job can use up to 96 cores|
|Number of nodes per MPI job||One MPI job can use up to 12 nodes|
|Number of jobs in the queue per user||2500 for single core & SMP jobs and 40 for MPI jobs|
|Job array size||2500 jobs per array|
|Memory per job||Default is 1 GB and maximum is 1500 GB.|
|Home space||Quota (soft and hard) set to 100 GB|
|Work space||Quota set to 500 GB soft and 2 TB hard with a grace period of 15 days.|
|Walltime for single core and SMP jobs||Maximum allowed is 336 hours|
|Walltime for MPI jobs||Maximum allowed is 168 hours|