Personal tools
Batch queue system (LSF)
System for running long programs
At the ICTP Linux is used for long calculation jobs. If you have a PC in an office assigned to you, you can run your programs there. Computers in the public areas must not be blocked for this purpose. Instead make use of our Linux cluster.
LSF
With this system, you specify on what type of computer your program should run, and (roughly) how much time and memory it needs. Based on these indications the least used computer with the requested capabilities will be chosen.
New: The system was previously offering a choice of 32-bit and 64-bit platforms. Now there are only the 64-bit computers left (named amd64-##
). That means that any program that you want to run must be compiled under the same architecture. For this purpose you can log in to either amd64-1
or amd64-2
where you will have 15 minutes of CPU time per login session.
Basic commands
LSF is installed in Linux. All LSF commands must be prefixed by
lsfrun
, except if used on amd64-1
and amd64-2
. The first time you use lsfrun
in a login
session you will have to provide your password. In order to use
the bsub
command, for example, you would write
lsfrun bsub -q amd64 ./program
To get detailed information about each of the commands described below, you can enter
man command-name
- The command
bqueues
shows you a list of available batch queues. What distinguishes a queue from another is mainly the run time limit for its jobs, reflected in the queue name. This time limit is not an absolute value, but is always taken in relation to the processing power of the computer a job is sent to. The purpose of this is to make the amount of work that can be done by a job independant on how fast the computer is on which it happens to be executed. A typical output would be:unix1.ictp.trieste.it> lsfrun bqueues QUEUE_NAME PRI STATUS MAX JL/U JL/P JL/H NJOBS PEND RUN SUSP amd64 99 Open:Active 16 10 1 - 6 0 6 0 15min 90 Open:Active - 5 1 - 2 0 2 0 1hour 80 Open:Active - 5 - - 4 3 1 0 10hours 70 Open:Active - 5 1 - 2 0 2 0 week 60 Open:Active - 5 1 - 0 0 0 0 month 50 Open:Active - 5 1 - 1 0 1 0 90min 1 Open:Active - 5 1 - 5 2 3 0 amd64_3hours 1 Open:Active - 10 1 - 0 0 0 0
Update: Please note that all queues except amd64 and amd64_3hours are obsolete because they were used for the 32-bit systems. Hence, for your calculation jb select between amd64 and amd64_3hours. lsfrun bhosts
shows you the status list for all computers under control of LSF.lsfrun bsub
sends a job to a batch queue. The program must take its input (if any) from files prepared in advance. By default, the output is sent by e-mail to the user:lsfrun bsub submit-options command command-options where the most important submit-options are: -q queuename (see above) -i input-file (stdin will read from this file) -o output-file (stdout will be redirected there) -e output-file (stderr will be redirected there)
At least the queue name must be specified, otherwise you will see an error message about a non-working default queue.
Matlab: To run Matlab, you can create an input file, e.g.sample.m
, and submit it in the following manner:bsub -q amd64 -o sample.output matlab -nodisplay -nojvm -r sample
lsfrun bjobs
shows the list of all your jobs in the queue with basic status information. For more details, use the-l
switch. To see the jobs of all users, writelsfrun bjobs -u all
.lsfrun bkill
jobnumber kills a job.lsfrun bhosts
shows you the list of hosts and their current workload. More information is obtained with the-l
switch.
For more detailed information, check the user guide of LSF .