Users have recently noticed a problem with pre-WS GRAM and the change
introduced in Globus 4.0 to 
use a single Condor job log for all jobs. The pre-WS GRAM poll function reads
the entire condor log 
each time it determines the state of a job. That is, the entire log is read all
the way through once for 
every job several times a minute. The log contains entries for all previous
jobs and cannot be easily 
rotated. After several thousand jobs have been run, the work needed to poll
each current job grows 
significantly, as entries for all previous jobs are reread.

We have developed a patch that restores the old configuration of a separate
Condor job log file for each 
job. This is only done for pre-WS GRAM. If the gram script for condor detects
that it's being called by 
WS GRAM, it still uses the single condor log file read by the SEG.

I understand that Globus plans to eventually have pre-WS GRAM use the SEG,
which will also solve this 
problem. I believe this patch serves as an important interim solution.