Sesame cluster

sesame_cluster.jpg Sesame is a 3584-processor beowulf cluster that can support applications requiring up to 3GB of memory per processor.

Production runs

to setup special events, mark your event in this calendar:

login: sesame.cluster@gmail.com
pwd: sesame4you

cal_sesame.cluster_gmail.com

Group schedule

default time slots for production runs:

Mon Tue Wed Thu Fri Sat Sun
user3.jpg user2.jpg user3.jpg user4.jpg user4.jpg user2.jpg user1.jpg
Daniel Hejun Daniel Ebru Ebru Hejun Yang

Queues

tromp_queue.jpg

risk interruptible default queue 3584 cores risk jobs can get killed by higher priority jobs. it should be used for non-critical jobs and testing
test non-interruptible test queue 3584 cores jobs can't get killed but have a run-time limit of 15 min. it should be used for testing and debugging purposes.
tromp non-interruptible priority queue 2118 / 1734 soft limit jobs have higher priority and can't get killed. it should be used for production runs

special: Quake for movie runs only (384 cores)


Tips

Job priorities are calculated on:

  • Group (tromp vs. the rest of the world)
  • User usage of the last 15/5 days
  • User wait time
  • number of cores job requests

To show priorities of your idle jobs:

diagnose -p

or

showq -i

To see total cores used in tromp queue:

make an alias

alias showq-totaltromp='showq -r | grep TR | awk "BEGIN{sum=0}{sum+=\$10}END{print \"tromp cores used:\",sum,\"of (2118 / 1734)\"}" '

then

showq-totaltromp

To see total usage of cluster:

pbstop
 
sesamecluster.txt · Last modified: 2011/03/28 17:25 by dpeter
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki