I believe that the general model of PP computing is that in general data
does not need backing up. Raw data should be copied so there is
effectively a backup somewhere else in the world and all other data are
derived, ie can be recreated either by re-processing the raw data or by
re-running a simulation from the same seeds. This means that there is no
need for backup for disaster recovery but it doesn't mean that all data
can be treated as volatile and just thrown away on a whim because this
creates big problems for the experiments who have to maintain catalogues
of where they think their data are. Perhaps they should have a more
fault-tolerant model that has multiple copies of everything and can cope
with unreliable catalogues but for now they don't.
I think it would be irresponsible of sysadmins (even collectively) to
define SEs as volatile without making sure experiments are aware of the
implications and that they are actually taking some action on the value.
John
> -----Original Message-----
> From: Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Henry Nebrensky
> Sent: 05 October 2005 00:38
> To: [log in to unmask]
> Subject: Re: Reminder of the next UKI meeting - Wednesday 5th
> October 11:00-12:00 via VRVS
>
> >
> > 3) In preparation for Service Challenge 4 all sites need to
> > initiate and progress two main areas: the deployment of an SRM
>
> Currently the user data in our (classic)SE is not backed up -
> hence the "volatile" (This was general LT2 policy at one
> point). What happens when we go to SRM? (Not sure of the
> forum for this)
>
|