This is a quick start guide to our clusters for users already familiar with operating in an HPC environment.
Accessing A Cluster§
Before accessing the Legion cluster, it is necessary to apply for an account. Once you have received notification that your account has been created, you may log in via SSH to:
- Legion: legion.rc.ucl.ac.uk
- Myriad: myriad.rc.ucl.ac.uk
- Grace: grace.rc.ucl.ac.uk
Your username and password are the same as those for your central UCL user id. Legion, Myriad, and Grace are only accessible from within UCL’s network. If you need to access them from outside, you need to log in via the Gateway, a departmental machine, or install the IS VPN service.
More details on connecting to these services are provided on the Accessing RC Systems page.
Managing data on the our clusters§
Users on our clusters have access to three pools of storage. They have a home
directory which is mounted read only on the compute
nodes and therefore cannot be written to by running jobs. They have a
"scratch" area which is writable from jobs and intended for live data and job output.
There is a link to this area called "Scratch" within the user’s home
directory. Finally, within a job users have access to temporary local storage on the
nodes (environmental variable
$TMPDIR) which is cleared at the end of
Legion has slow external network connections, so there is a dedicated transfer node with 10 gigabit network links to and from Legion available at:
Transfers to Grace and Myriad may use any of the normal login nodes.
For more details on the fairly complicated data management structures within Legion, see the Managing Data on RC Systems page.
Our clusters run an operating system based on Red Hat Enterprise Linux 7 with the Son of Grid Engine batch scheduler. UCL-supported and provided packages are made available to users through the use of the modules system.
||lists available modules|
||loads a module|
||removes a module|
The module system handles dependency and conflict information.
You can find out more about the modules system on Legion on the RC Systems user environment page.
Compiling your code§
We provide Intel and GNU compilers, and OpenMPI and Intel MPI through the modules system, with the usual wrappers. For a full list of the development tools available see here or in the development tools/compilers sections of the modules system.
You can find out more about compiling code on a cluster on the Compiling page.
Job scheduling policy and projects§
A fair-share resource allocation model has been implemented on all our clusters. See Resource Allocation for more information and context.
Jobs submitted to the scheduler (with "qsub") are shell scripts with
directives preceded by
#!/bin/bash -l # Request ten minutes of wallclock time (format hours:minutes:seconds). #$ -l h_rt=0:10:0 # Request 1 gigabyte of RAM per process. #$ -l mem=1G # Set the name of the job. #$ -N SomeScience_1_16 # Select the MPI parallel environment and 16 processes. #$ -pe mpi 16 # Select the project that this job will run under. (Only if you have # access to paid resources). #$ -P <your_project_id> # Set the working directory to somewhere in your scratch space. #$ -wd /home/<your_UCL_id>/Scratch/output
You can then follow these directives with the commands your script would execute. Legion supports a wide variety of job types and we would strongly recommend you study the example scripts.
Jobs can be controlled with
qsub (submit job),
qstat (list jobs) and
qdel (delete job). See the Introduction to batch processing page for more details.
Testing jobs using Interactive Jobs§
As well as batch access to the system, it is possible to run short, small jobs with
interactive access through the scheduler. These can be
requested though the
qrsh command. You need to provide qrsh with the
same options you would include in your job submission script, so:
qrsh -pe mpi 8 -l mem=512M -l h_rt=2:0:0
Is functionally equivalent to:
#!/bin/bash #$ -S /bin/bash #$ -pe mpi 8 #$ -l mem=512M #$ -l h_rt=2:0:0
Except, of course, that the result of qrsh is an interactive shell. For more details, see the Interactive Jobs page.