hare.epcc.ed.ac.uk (Edinburgh)
hare is a SUSe 9.x Linux HPC cluster consisting of 68 AMD Opteron processing cores.
System Organisation
- 12× 4-processor compute nodes with 8 GB Memory and 160 GB disk space
- 2× 8-processor compute nodes with 16 GB memory and 400 GB disk space
- 2× 2-processor interactive nodes with 4 GB memory and 800 GB disk space
- 1× 4 TB RAID disk array.
Compilers
The Portland Group compilers are available on the system.
Libraries
The MPI, and LAPACK libraries are also installed.
Queues
The queuing system is set up to try and acheive a balance between parallel and serial jobs on the cluster. There are basically four types of queue on the system, these are: input testing queues, serial queues, parallel queues and fat node queues.
Input Testing Queue
This queue is used for testing the input for a job before submitting to one of the longer queues. Jobs on this queue are virtually guaranteed to run instantaneously. Please use this queue as there is nothing more disappointing than waiting hours/days for a slot and then your job failing due to a typo!
- test-run.q
- Serial queue with a time limit of 3 hours.
Serial Queues
As the name suggests, these queues run serial jobs on the cluster. They can also be used for array jobs. Array job users should note that there is a maximum of 24 processors available through the serial queues.
The different queues have different priorities so please use the one that corresponds most closely to the time estimated for your job as this will help your job run more quickly.
If you wish to use your own scripts to submit jobs to these queues you should note that there are resources that must be requested to have access to the queues. See the Job Submission topic
- serial-medium.q
- Serial queue with a time limit of 1 week.
- serial-long.q
- Serial queue with a time limit of 4 weeks.
Parallel Queues
These queues run parallel (either SMP or distributed memory) jobs on the cluster. Note that there are a maximum of 4 processors in SMP mode or 48 processors in distributed memory mode.
The different queues have different priorities so please use the one that corresponds most closely to the time estimated for your job as this will help your job run more quickly.
If you wish to use your own scripts to submit jobs to these queues you should note that there are resources that must be requested to have access to the queues. In addition, resource reservation should be switched on when submitting jobs to these queues (and the specific software job submission scripts do this). See the Job Submission topic
- parallel-short.q
- Parallel queue with a time limit of 3 hours.
- parallel-medium.q
- Parallel queue with a time limit of 1 week.
- parallel-long.q
- Parallel queue with a time limit of 4 weeks.
Fat Node Queues
These queues run parallel SMP jobs on the fat (8 processor) nodes on the cluster. They have been particularly designed with large Gaussian 03 jobs in mind. There are a maximum of 8 processors available for each job.
The two different queues have different priorities so please use the one that corresponds most closely to the time estimated for your job as this will help your job run more quickly.
If you wish to use your own scripts to submit jobs to these queues you should note that there are resources that must be requested to have access to the queues. You should also bear in mind that each job has exclusive access to a fat node and should therefore use all 8 processors if possible. See the Job Submission topic
- fat-medium.q
- Fat node queue with a time limit of 2 days.
- fat-long.q
- Fat node queue with a time limit of 2 weeks.
Please see the Support Pages for more information on submitting jobs to the queue system.
The cluster uses the Sun Grid Engine for controlling and submitting jobs.
