Hi *,
A belated summary of the VO box discussion at the GDB in Bologna. This
isn't the official sumamry that you can get from the GDB, it's mine.
Ian Bird presented my slides, I was present for the discussion. I think
his slides are publicly available from the GDB agenda page, and he
presented a second set of slides which were basically listing all your
objections to my slides ;-) The stuff on CERN VO boxes at the end of
his slides was not actually presented, I think it was there for backup.
Afterwards the four experiments made presentations about what they
wanted to do with VO boxes. I already sent a heads-up about that page.
Here is my take on the ensuing discussion. It's not a complete summary.
I think Jeremy Coles is writing a detailed report.
- more or less everyone agreed that all attempts should be made to move
the "missing functionality" into the middleware stack. There was a wide
range of optimism/pessimism about whether this would actually be done in
a way that was useful to the experiments.
- most of the experiments seemed happy with the GSIssh approach, ATLAS
preferred talking about container technology (secure tomcat)
- CMS had a rather extreme position that sites were being out of line in
objecting to VO boxes, and that actually they had said they were going
to provide T1 service to CMS and hence had to deploy VO services as part
of the deal. Indeed, in the draft computing MoU there is a statement
(point ix in Annex 3.2) that could be read in this way. Might want to
have a look before signing anything (this goes for all experiments not
just CMS).
- my feeling was that for most of what the experiments want to do with
workload manglement on the VO boxes, we are almost there with the
generic LCG stuff, and there is no strong reason for deploying
VO-specific job management services. Each experiment was doing
something quite different here.
- on the other hand more or less ALL experiments had come up with
significant pieces of data management stuff, and they all looked pretty
much the same to me. Not having played with the newest round of gLite
FTS stuff I can't say whether it fits the bill, but it is certainly true
that *before* FTS and friends, there was a real hole in DM on LCG and
one can well understand why the experiments built something. Given that
it looks quite similar across VOs one might reasonably hope that a
generic solution would work.
- the experiments were not optimistic (me neither) that the need for
VO-specific site services will ever really disappear
- a couple of the 'solutions' appeared to be trying to address the
problem of WNs not having ANY outbound connectivity. Are there really
sites out there that are seriously considering this??
- the recommendation was that T1s deploy the services now, in order to
not hold up SC3. but that there was no real policy agreement yet. A
couple of sites mentioned that they would not deploy VO boxes given the
current situation. and a few more mentioned that they were sceptical
that the services would ever go away once deployed. call it
gruding and non-unanimous acceptance
There was also a recommendation (I think it was accepted, at least
people did not object) to carry on the discussion in a more limited
forum that would have more or less equal representation from sites and
experiments.
- experiments got the message loud and clear about the two docs
(security and operations) and did not have strong objections.
Those are the highlights as I remember them. Others who were there
please feel free to chime in here. People who are interested might also
contact the GDB chair and ask to be sent a copy of the official summary
when it is ready.
Thanks,
J "more in November" T
|