> Data servers: i.e., the amount of space required in SRM avaliable storage.
I meant manchester model with dcache on the WN or dataservers
specifically dedicated.
> Yes, it's all a bit vague. We don't really know what the VOs' models are, or
> how that's best served by T2 storage software. Certainly you would want a
> second copy of any precious files, i.e., ones which are produced at your site
> and are not yet copied up elsewhere.
Or you designate areas for this temporary data.
> However, I worry that with 1000 nodes your hardware failure rate will be
> ~1/day, and each time a disk goes phut with a single copy of a grid
> registered file on it, you'll have to manually delete the catalog entry, even
> if that file was not precious, because jobs will arrive and expect to be able
> to access that SURL. So some sort of resiliancy beyond 1/copy on an IDE disk
> would seem to be needed - either through RAID 5, or a second copy in the SRM.
this is a synchronisation problem between the catalogs and the local
storage. It is one of those things that data management developers never
wanted to listen to. I don't know how they foresee to solve it. Apparently
one of the VO agents that should go on the VO-box should be doing exactly
this.....
cheers
alessandra
>
>>
>> If other sites are evaluating to go the same way it might be interesting
>> to put together different scenarios.
>
> Bristol expressed an interest in this scenario, and so did Mark at Durham.
> It's a definately a good thing, but it's early days yet...
>
> cheers
>
> g
>
> PS. How's the Guinness?
>
>
--
********************************************
* Dr Alessandra Forti *
* Technical Coordinator - NorthGrid Tier2 *
* http://www.hep.man.ac.uk/u/aforti *
********************************************
|