sPHENIX discussion of software

[pinkenburg <pinkenburg AT bnl.gov>]

Hi folks,

this is about the logs condor writes (the Log  = in your jobfile), not 
your job log (Output =) or error log (Error =).

writing the condor logs to gpfs slows gpfs down drastically up to the 
point where it becomes unresponsive and jobs don't even start because 
condor cannot create this file. It's a ton of short writes and gpfs is 
not made for this and with users now being able to run thousands of jobs 
at once it's really having an impact.

Normally you don't care about the content of this file anyway, it 
contains stuff like:

000 (1297725.000.000) 10/02 13:02:22 Job submitted from host: 
<130.199.48.83:9618?addrs=130.199.48.83-9618&noUDP&sock=8867_649e_3>
...
001 (1297725.000.000) 10/02 13:02:25 Job executing on host: 
<130.199.157.87:9618?addrs=130.199.157.87-9618&noUDP&sock=21355_aff9_3>

which is helpful for debugging condor.

Please tell condor in your job file to open this file on the local /tmp 
directory. So we do not have condor overwrite each others log files 
please create a directory under tmp with your username:

mkdir /tmp/<your rcf name>

and tell condor to write the logfile there:

Log = /tmp/<your rcf name>/<whatever name you give it>


As an additional benefit - /tmp gets cleaned up by a cron job of old 
files so you don't even have to delete them yourself. If you need to 
look at them, you just have to be on the machine you are running the job 
from.

It's not a big change so please do it asap and we'll all benefit 
immediately.

Thanks
Chris



...

--
*************************************************************

Christopher H. Pinkenburg       ;    pinkenburg AT bnl.gov
                                ;    http://www.phenix.bnl.gov/~pinkenbu

Brookhaven National Laboratory  ;    phone: (631) 344-5692
Physics Department Bldg 510 C   ;    fax:   (631) 344-3253
Upton, NY 11973-5000

*************************************************************