David,
The short answer is "Not right now, but we're looking into it".
With proper SRM implementations (for instance the LCG DPM) we can
allocate both "volatile" and "permanent" space for a VO. The volatile
space would be cleaned up when the files run out of their pin time. The
permanent space would act like all space in LCG-2 is now - no automatic
cleanup, and it lives for ever. Also we would have information
providers that actually publish "non-reserved free space" so a job can
reserve space, and be guaranteed to get it when it needs it
Of course, all of this changes the model, and we need to work out what
that actually means for the tools - e.g right now every tool assumes
that once data is somewhere, it stays there unless explicitly deleted.
We are also looking at schemes where the SE would copy data off to a
"bigger" SE when it is full (SRM "durable" space).
As people have mentioned, our big limitation right now is the so-called
"Classic SE", and we are working hard on getting solutions to replace it
(dCache, LCG DPM). But this is a big change in the architecture (e.g.
how does the RLS deal with volatile files - who is responsible for the
update, what if it fails - which process must stay around to clean the
RLS up). I think we can't deploy anything until we all understand the
issues a bit better.
Of course, all input from sites and experiment data management experts
on how they might use these features is more that welcome.
Cheers,
James.
> -----Original Message-----
> From: LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Dr D J Colling
> Sent: 15 January 2005 17:37
> To: [log in to unmask]
> Subject: [LCG-ROLLOUT] An SE configuration question.
>
> Hi,
>
> Is it possible to have more than one area associated with an
> experiment?
> For example to have an area that has a cleanup algorithm
> where data can reside for (say) a week and another area that
> is permanent?
>
> Would anybody ever use the it?
>
> All the best,
> david
>
|