 |
 |
hi, i know this has been discussed at times but I haven't quite gotten FSL properly integrated with torque (pbs). For first level analysis, things work fine. Since pbs requires a script for submission, rather than just a command line, I am shuttling the fsl_sub arguements into a tmpfile and then submitting that. It seems to work fine. dependencies have also been changed over to the pbs way of doing things, but when it comes into arrays, things start breaking down. when fsl sends in these jobs, it holds some jobs on other jobs through the scheduler. Within SGE, you can have a job hold on array by specifying the jobid of the entire array. Within torque currently, you need to specify a hold on individual array elements. PBS returns the overall jobid when it is submitted, and whatever fsl application that is generating these submissions, tries to hold on it. here is an example: [naveed@hostname ~]$ qsub -t 1-1 testscript 1306.hostname.caltech.edu i then submit the same job that depends on the job id of the first: [naveed@hostname ~]$ qsub -W depend=afterany:1306.hostname.caltech.edu testscript 1308.hostname.caltech.edu this job holds indefinetaly because there is no jobid 1308.hostname.caltech.edu, but rather 1308-1.hostname.caltech.edu What is the best way to get around this? what application is actually sending the fsl_sub commands and is their an easy way to alter it to deal with the different dependency types? Does anyone have a clever idea on how to do this within fsl_sub itself?