Computing Resources
Lengthy calculations ( >> few minutes), in particular treatment planning,
should be run in background (batch) mode. This allows fair and more efficient
use of computing resources, because a scheduling system called
LoadLeveler
automatically locates eligible idle CPUs and assigns work to them.
This way "overcrowded" and "unemployed" machines can be avoided.
You use the LoadLeveler by submitting a command file
describing your requirements (in memory and CPU time by selecting a
job class
)
as well as specifying
the program to run and the destination of the
stdin, stdout, stderr
streams. Essentially it is not much different from working
with the command line and nohup, but a lot more efficient.
A commented example for a command file is
here
You may use either the Motif-based graphical tool xloadl
or a handful of shell commands to control job execution, e.g.:
llsubmit ll.cmd
submits a job defined in command file ll.cmd
llq
lists all active jobs
and
llcancel biors6a.4287.0
llcancel -u bioguest
cancels the specified job or all jobs of user bioguest,
respectively.
A documentation of LoadLeveler can be found as
HTML
and
PDF.
Job Classes
Here's the current definition of job classes on the bio*
AIX machines, with their upper memory and CPU time limit.
| Classes |
bioshort |
biomedium |
biolong |
biolarge |
biovlarge |
biohuge |
| Memory |
128 MByte |
256 MByte |
256 MByte |
2 GByte |
3.25 GByte |
8 GByte |
| Time |
15 m |
1 h |
24 h |
72 h |
72 h |
72 h |
Note that these times are calibrated against the workhorse Power4+
machines (biops6a,b,z). Jobs on slower machines are allotted
more time according to their relative speed.
Last updated: 20-Apr-2005, M.Kraemer
Impressum Data privacy protection
Impressum Data privacy protection