On 23 May 2012 06:13, Ewan MacMahon <[log in to unmask]> wrote:
>> -----Original Message-----
>> From: GRIDPP2: Deployment and support of SRM and local storage management
>> [mailto:[log in to unmask]] On Behalf Of Sam Skipsey
>>
>> What do we mean by exotic? At the very not-exotic end, but new for SEs,
>> DPM has an HDFS backend in beta which supports block distribution.
>>
>
> How near to usable is 'in beta' in DPM's case?
In this case, 'in beta' means: the plugin works, but is not offically
released yet (and doesn't do block-parallel copies yet).
> I've been vaguely musing
> about the possibilities of DPM over HDFS over spare WN disk (since ATLAS
> apparently only want a maximum of 20GB per slot rather than the 50GB we
> thought they did, it turns out that we have quite a lot of spare WN disk).
>
So, the intent for the DMLite backends is to have them supported as
multiple backends at once - you should be able to have an "HDFS pool"
and a "bunch of disk servers pool". I think this would probably meet
your requirements.
(In case mention of DMLite is worrying, I can also confirm that DMLite
is fully backwards compatible with the DPM databases etc - it's more
of a refactoring of the code to make it a) cleaner to develop on, b)
allow easy writing of plugins for backends, transports etc.)
Sam
> Ewan
|