Current policies of the computing cluster
SSH
The use of ssh to the computing hosts isn’t allowed;
QUEUES
there are three level of queues
- allgroups queue: allocated on every host of the cluster and allowed for every user. All the user that submit jobs on this queue are suggested of user a checkpoint software.
- reserved queues: allocated on a subset of hosts (not overlapping with other reserved queues). The jobs submitted on these queues have priority higher than the jobs on the allgroup queue and will be started at the first available slot but without enforce the stop of any jobs
- owner queues: allocated on a subset of hosts (not overlapping with other owner queues). The jobs submitted on these queues have priority higher than any jobs in other queues and will be started alt the first available slot even enforcing the stop of other job(s) of other queues.
- interactive queue: you can ask for an interactive session on one host of one hour (or less) of time. This can be used to debug a program or for check a running job
- debug queue: Queue reserved for debugging purpose. Maximum run time: 1 day (24 hours).
- testing queue: sometimes you will find a ‘testing’ queue. In this queue are grouped machines/hosts under testing. Feel free of submit your jobs but consider that the queue and jobs can disappear without warning. A report of your tests is welcome.
GPU
The GPU can be used with two separate setup showed in the Examples page:
- shared mode: in this case you can select a GPU on a particular host and request for a certaint percentage of GPU Video RAM (30%, 50%, etc.), other users can use the remaining video RAM with other jobs. Using this setup all the GPU will be used at the best performances for all the users;
- reserved mode: discouraged, use only if sure. In this case the full GPU is reserved for a single job