Hi David,
David Smith wrote:
> On Thu, 8 Jul 2004, Sergio Fantinel wrote:
>
>
>> for a few weeks we, at LNL, have been observing a suspicious use/abuse of
>>globus-job-run on our CE that does "bacct" (LSF command for the accounting) on
>>each single job ID runned by our site. This method of operate is not
>>appropriate because it HEAVYLY loads our CE.
>
>
> Hi Sergio,
>
> Do you have a particular reason to believe they are run from
> globus-job-run? If not, I think it is more likely that these commands are
> being run by the lcglsf jobmanager after each job finishes, to place
> accounting information into the system log.
Thanks, this better clear the situation...
>
>
>>What's the use of this commands? Are they really needed? At least it could be
>>done in a smarter way eg. using a single bacct with all the list of job IDs.
>>Please try to use some other methods because otherwise we will have to disable
>>the access to grid users to the LSF commands other than submit.
>
>
> The bacct calls are only needed to populate the log with the accounting
> information - if the information isn't needed or used we can disable the
> command. (Let me know and I can help you do that). Do you find that this
> causes problems - apart from the observation that it keeps a load on the
> CE? The number of parallel calls is limited (3 as I recall, as in your
> output)
So I have some questions:
- in what (log) files are stored the info and what info are written/ is there
some documentation about this?
- is there some middleware components that needs this info?
- is there the same accounting from the jobmanager with PBS?
The main problem for us here, we are pretty sure, come from the fact that we
have a large accounting LSF log files and we do not know how to put safely "off
line" parts of these logs; maybe in the future we need to "re-attach" they to
the system...
So each bacct read the hole accounting from NFS... et voila the time in
operations explode!
Having this heavy load on the CE we think is not a good thing at all!
So we need to solve in some way the "problem".
If there is no other ways we would like to turn off the logging made by the
jobmanager.
Thanks,
Sergio
>
> Yours,
> David
>
> --
> -------------------------------------------------------------------------
> David Smith e-mail: [log in to unmask] tel: +41 22 76 74462
> Address: D. Smith, CERN G06610, Bat 28 R-007, 1211 Geneva 23, Switzerland
> -------------------------------------------------------------------------
--
---------------------------------------------------------------------
Sergio Fantinel EGEE Project
---------------------------------------------------------------------
INFN - Lab. Naz. di Legnaro phone: +39 049 8068 489
viale dell'Università n. 2,
35020 Legnaro (PD) ITALY [log in to unmask]
---------------------------------------------------------------------
|