Hi Jeff,
I fully support Jeff point about increasing security by reducing
the number of unnecessary services running on our nodes.
However since the origin of this thread is the result of some
site test I have start during the last days let me motivate them.
During the last GDB there was a request to experiment to
increase the use of LCG resources. So far when we tried to do that in
the past we always reach a limit when few faulty sites (that remain
empty since they abort every job that reaches them) prevent jobs to
reach correctly configured sites.
This is not a static situation but with over 60 sites supporting
LHCb it changes every day, as you may imagine.
A typical source of problem is user authentication on the CE:
wrong grid-mapfile (might be OK for dteam, though), expired CRL,... all
these problems can be easily detected and diagnose with a simple
glbus-job-run and globus-url-copy.
Another source of problems is connectivity with RB's, thus I use
the above commands to upload a script to each CE supporting LHCb (and
publishing a queue in "Production") and checking the connectivity via
globus-url-copy commands.
This is why I have been trying to run globus-job-run and
globus-url-copy against CE's supporting LHCb, and why we have reported
the problem found at the new CE at NIKHEF.
The test suite it is still under development and further test
including real job submission still need to be added.
If we come to the conclusion that some of these services should
not run on the CE then I will have to change the test, but in the
current situation I think we did the right thing reporting to the
corresponding site admin.
I hope this helps to clarify the origin of the question.
Regards
Ricardo
=======================================================================
========
Ricardo Graciani Diaz
Dept. Estructura i Constituents de la Materia
Facultat de Fisica Tel: +34 93 403 9183
Universitat de Barcelona Fax: +34 93 402 1198
Diagonal, 647
E-08028 Barcelona
=======================================================================
========
> -----Mensaje original-----
> De: LHC Computer Grid - Rollout [mailto:[log in to unmask]]
En
> nombre de Jeff Templon
> Enviado el: lunes, 07 de marzo de 2005 15:21
> Para: [log in to unmask]
> Asunto: Re: [LCG-ROLLOUT] Does a CE need gsiftp?
>
> Hi,
>
> On Mon, 2005-03-07 at 15:12, owen maroney wrote:
> > Two questions:
> >
> > 1. If the ESM can change the tags through a grid job, do they
strictly
> > need to be able to change the tags from a UI? Or is this just
something
> > that makes their jobs easier?
>
> The usual case here is that if you can only do it from a job, you may
> have to wait *days* at some sites before your job actually runs.
> Suppose you want to turn off the ATLAS-OK tag at Imperial because of a
> problem ... you want to turn it off *now* ...
>
> Given the current system I am not sure you could do it any better.
The
> best idea is still to have this sort of info put in an info system
> component that's directly under the experiment control, so that tag
> setting on a CE is no longer necessary.
>
> J "kind of like using a train to put your
> number in the phone book" T
|