On Wed, 10 Jun 2009, Burke, S (Stephen) wrote:
> Testbed Support for GridPP member institutes
>> [mailto:[log in to unmask]] On Behalf Of Henry Nebrensky said:
>> IIRC one of the worries back then was that if the SE storage
>> was directly
>> mounted (e.g. NFS) then users could modify the datafiles
>> without the SE
>> noticing, thus completely messing up replica management.
>>
>> Do current invocations of "file" access get round this somehow?
>
> I don't know what storm does, but the problem you describe potentially
> exists for any SE, e.g. you could write files directly with gridftp. The
> basic answer is not to do it! The LHC VOs at least are now doing checks
> for dark data and checksum mimatches to try to catch inconsistencies.
The related concern from sites was that if pool users could create files
on the mounted SE space, they could use this as an illicit persistent
storage space (so that's what "hot files" are :) ).
Both these activities are indeed still possible by using gridftp, but then
at least you have the gridftp logs to provide a trail of who did what;
with the NFS route you might not have more than a timestamp
(group-writable files could have been modified by any pool user).
Checking for dark data is an important activity, but with ~40Tb per server
it seems to me it will start to take up significant resources if run too
often.
The simplest solution might simply be to only mount the SE space
read-only, but that's yet another thing that would have to be published
correctly (and then parsed by the user).
Thanks
Henry
--
Dr. Henry Nebrensky [log in to unmask]
http://people.brunel.ac.uk/~eesrjjn
"The opossum is a very sophisticated animal.
It doesn't even get up until 5 or 6 p.m."
|