Bugzilla – Bug 3090
CAMPAIGN: WS GRAM Study #4.1 MEJS max concurrency Condor
Last modified: 2005-08-03 17:20:38
You need to
before you can comment on or make changes to this bug.
An important metric for the GRAM Service is its capacity for
managing/monitoring a large number jobs
at the same time. For this test, globusrun-ws was used to submit -batch jobs
to the same GRAM (MJFS)
service. 4 concurrent client test program were started, each submitting 10,000
batch jobs in a loop.
The gram jobs were submitted to the condor scheduler. Jobs failed after 32,000
jobs. The error was
due to the MJFS trying to create directory in ~/.globus for the job. There is
a system limit of 32.000.
This can easily be solved, by using sub directories, but has not been done yet.
A bug will be created for
this, but the current limit is 32,000.
To duplicate the test:
1. start a globus service container
2. mkdir client1 client2 client3 client4
3. in each client dir run: ../consub.pl lucky0.mcs.anl.gov:8443 10000 99999
note: replace lucky.mcs.anl.gov:8443 with the container's contact info
4. test various jobs by running globusrun-ws -status -j <epr file>
Created an attachment (id=569) [details]
submission program for the max concurrency test