Platform LSF Administrator's Primer Version 6.2

Common LSF Problems
Platform LSF Administrator’s Primer
60
❖ The messages in LSF_LOGDIR/res.log.host_name on the execution
host.
res is responsible for authenticating users in LSF
❖ The setting of LSF authentication (LSF_AUTH in
LSF_CONFDIR/lsf.conf):
✧ LSF default authentication is eauth (LSF_AUTH is not or is defined as
eauth in lsf.conf)
✧ If LSF_AUTH is defined as identd in lsf.conf, identd must be
installed and running properly on each LSF host
❖ That the eauth executable in LSF_SERVERDIR is implemented correctly
❖ That all LSF commands are owned by root and have the file permission mode
to
-rwsr-xr-x so that the user ID bit for the owner is setuid
See Administering Platform LSF for information about LSF authentication and the
eauth executable.
6 Submitted jobs
do not start
Many things can prevent jobs from starting. Check:
◆ Wait for up to a minute for LSF to dispatch your job. Some LSF activities happen
periodically. You may be able to speed up job dispatch by changing the following
parameters:
❖ JOB_ACCEPT_INTERVAL in
LSB_CONFDIR/cluster_name/configdir/lsb.queues
❖ SBD_SLEEP_TIME in
LSB_CONFDIR/cluster_name/configdir/lsb.params
See the Platform LSF Reference for information about these parameters.
◆ Use bjobs -p and check the job pending reason.
◆ Examine the output of bhosts -l for the load thresholds on the hosts and
lsload -E for the effective run queue lengths on the hosts.
◆ Use bqueues -l to check the queue thresholds, host limits, and user limits.
◆ If you submitted the jobs from a host that is the only host of its type in the cluster,
by default, LSF tries to run the job on this host only. To run jobs on any available
host type, submit the job with a command like:
% bsub -R "type==any" myjob
If none of these applies to your situation, contact [e-mail].
7 LSF jobs go to a slow host while faster hosts are idle
Host processor speed is not the only factor in host selection:
◆ LSF uses the effective run queue lengths (displayed by lsload -E) to choose the
least loaded execution hosts. Two hosts with the same effective run queue length are
ordered according to their CPU factors.
◆ By default, LSF runs jobs on hosts of the same type as the submission host. To run
jobs on any available host type, submit the job with a command like:
% bsub -R "type==any" myjob
◆ The faster hosts may not be included in the queue where the job was submitted.
◆ The faster hosts may not have the resources required by the job.
◆ The faster hosts may be closed or have closed dispatch windows or closed run
windows.