Béluga
Availability : March, 2019 |
Login node : beluga.alliancecan.ca |
Globus Endpoint : computecanada#beluga-dtn |
Data Transfer Node (rsync, scp, sftp,...) : beluga.alliancecan.ca |
Portal : https://portail.beluga.calculquebec.ca/ |
Béluga is a general purpose cluster designed for a variety of workloads and situated at the École de technologie supérieure in Montreal. The cluster is named in honour of the St. Lawrence River's Beluga whale population.
Site-specific policies
By policy, Béluga's compute nodes cannot access the internet. If you need an exception to this rule, contact technical support explaining what you need and why.
Crontab is not offered on Béluga.
Each job on Béluga should have a duration of at least one hour (five minutes for test jobs) and a user cannot have more than 1000 jobs, running and queued, at any given moment. The maximum duration for a job on Béluga is 7 days (168 hours).
Storage
HOME Lustre filesystem, 105 TB of space |
|
SCRATCH Lustre filesystem, 2.6 PB of space |
|
PROJECT Lustre filesystem, 25 PB of space |
|
For transferring data via Globus, you should use the endpoint computecanada#beluga-dtn
, while for tools like rsync and scp you can use a login node.
High-performance interconnect
A Mellanox Infiniband EDR (100 Gb/s) network connects together all the nodes of the cluster. A central switch of 324 ports links the cluster's island topology with a maximum blocking factor of 5:1. The storage servers are networked with a non-blocking connection. The architecture permits multiple parallel jobs with up to 640 cores (or more) thanks to a non-blocking network. For jobs requiring greater parallelism, the blocking factor is 5:1 but even for jobs executed across several islands, the interconnection is high-performance.
Node characteristics
Turbo mode is activated on all compute nodes of Béluga.
nodes | cores | available memory | CPU | storage | GPU |
---|---|---|---|---|---|
160 | 40 | 92G or 95000M | 2 x Intel Gold 6148 Skylake @ 2.4 GHz | 1 x SSD 480G | - |
579 | 40 | 186G or 191000M | 2 x Intel Gold 6148 Skylake @ 2.4 GHz | 1 x SSD 480G | - |
10 | 6 x SSD 480G | ||||
51 | 40 | 752G or 771000M | 2 x Intel Gold 6148 Skylake @ 2.4 GHz | 1 x SSD 480G | - |
2 | 6 x SSD 480G | ||||
172 | 40 | 186G or 191000M | 2 x Intel Gold 6148 Skylake @ 2.4 GHz | 1 x NVMe SSD 1.6T | 4 x NVidia V100SXM2 (16G memory), connected via NVLink |
- To get a larger
$SLURM_TMPDIR
space, a job can be submitted with--tmp=xG
, wherex
is a value between 350 and 2490.
Monitoring jobs
To maximize the use of resources and reduce your waiting times in the queue, you can monitor your CPU and GPU past or current compute tasks in real time in the portal.
For each job you can monitor
- the use of compute cores,
- the use of memory,
- the use of GPUs.
When compute resources are little or not used at all, it is important to use the allocated resources and to adjust your requests. For example, if you request four cores (CPUs) but only use one, you must adjust your submission file accordingly.