Instructions and policy for the application of CPU-time on UPPMAX

Large scale applications

Large applications of more than 80000 core-hours per month should be applied for via SNAC,
The Swedish National Allocations Committee (http://www.snic.vr.se/snic-committees/snac).
National application rounds are held twice a year.  Applications for time can cover either half a
year or a full year, starting from January or July.
Application to all SNIC (The Swedish National Infrastructure for Computing) centers should be sent in at once.

For more information about large applications see: http://www.snic.vr.se/apply-for-resources/large-scale-applications

Medium scale applications

Medium size applications of up to 80000 core-hours per month are handled locally on
UPPMAX. Applications for time can cover either half a year or a full year, in both cases
ending each quarter (end of March, June, September, and December). Applications can
be sent in at any time and will be handled within two weeks of arrival. Should enough
time be left on the UPPMAX resources, your application may be granted time directly,
otherwise the grant will be postponed until the beginning of next quarter. If no time is
available, and if it is possible, UPPMAX may also forward the application to a similar
resource on one of the other SNIC centers. The applications are sent in by filling in a web
form on the SNAC/VR webserver: http://www.shpc.net/snac/medium/submit/

For more information about medium sized applications see:
http://www.snic.vr.se/apply-for-resources/medium-scale-applications

Local CPU time allocation

All users on UPPMAX are by default allocated 2000 high-priority core-hours per month.
To be a user on UPPMAX you must be member of a project headed by a senior scientist
in Swedish academia, at least at the level of assistant professor (forskarassistent). The
locally allocated time may be increased up to 4000 CPU hours upon request.
Should your demands increase above the default allocated time, it is recommended,
however, to send in a medium sized application (see above).

Local CPU time is applied for using the UPPMAX web forms.
Please see the support pages for instructions on how to create an UPPMAX project and
how to apply for an account connected to an UPPMAX project.

Emergency CPU time allocation

At the end of PhD studies or exam work, or when you have a research visitor, it may be
necessary to ask for more computer time for a shorter period of time. In this case you
are advised to send in a medium sized application (see above) and state that you want a
short time project, no longer than three months, and the reason for this. Note that if no
computer time is currently available, it might take up to three months before your application
can be granted time.

Computer time is distributed monthly

The computer time is distributed between the users once per month, starting on the first
every month. This means that if you do not use up all your allocated core-time every month,
the time is "lost". The lost time is shared among those users who have already used up their
time. It is also assumed that your load on the UPPMAX resources are fairly even. In other
words, you may not assume that all your allocated time is available to you the very last
week every month. For this reason it is advantageous for you (as a group) to send in
applications together with your colleagues, allowing you to perform calculations while your
other team members are developing and/or planning and vice versa.

How to specify which project should be billed the time on UPPMAX and how to get information about the time used so far the current month.

If you are a local user on UPPMAX, your default project number is registered when you first
apply for the creation of an UPPMAX project, using the UPPMAX web forms. If you are a
SNIC user (large or medium size) your project number is sent to you by SNAC when you
are granted time. If you are both a SNIC and UPPMAX user you will have at least two
different project numbers. The UPPMAX project numbers have the form pYYYYNNN where
YYYY is the year and NNN a numerical id. The SNIC project numbers have the form
sXYYMM-NN, where YY is the year, MM the month and X and NN numerical ids.

Each user must specify which project each job belongs to, by using the -A flag to
sbatch, for instance

#!/bin/bash -l
#SBATCH -J jobname
#SBATCH -A s00000-00
#
#SBATCH -t 24:00:00
#SBATCH -p node -n 16

module load intel/12.0 openmpi/1.5.0
mpirun -np 16 ./myprogram

To find out how much CPU time you have used in your project you can use the projinfo
command which is available if you have loaded the uppmax module

module load uppmax
projinfo

The projinfo command also lists the amount of core-time available to your projects, as well
as the amount of time all the members of your project have used. Note that only projects
where you have used time will be listed!

How the priority is implemented on the UPPMAX resources.

The priority of the jobs is different, depending on if they have high or low priority. High priority
is given to jobs belonging to projects who have not yet used up their per month allocated time
from SNAC or locally. Lower priority is given to jobs belonging to projects who has used up
their time.

Certain research groups have cofunded Kalkyl, they therefore have large CPU time allocations
corresponding to their part of the system.