LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Dimitris Zilaskos
said:
> Ok, now I remembered better, the particular case I have
> encountered is
> that the *same* lfc is announce as central by two *different* sites,
> each being part of a different grid. Thus, my site which was part of
> *both* grids, had the *same* host returned by lcg-infosites *twice*,
> since the top bdii was receiving it from two sites
That sounds like a general recipe for trouble, in such a situation the
grids should really be merging the information and not just adding it!
> I still think there might be cases that a "central" service like LFC
> might be have more than one node, for the LFC case in
> particular I have
> not seen that yet but that could happen in the future, with
> some sort of
> failover, in the fashion of the infosys variable. Central
> services are a
> very bad practice anyway, but this is a different issue.
Well, yes ... if you just have multiple front ends for load-balancing
that would normally be done with a DNS alias and you still just have one
published endpoint. Full service replication is a difficult problem for
the LFC, because it can have a very high transaction rate so
synchronising multiple copies of the database is hard - but it would be
nice to have, because even non-central LFCs are a single point of
failure, i.e. a given file is normally only catalogued in one server and
if that's down then all its files are unavailable.
Anyway from the publication point of view things will be a bit better
with glue 2.0, because you can have multiple endpoints for the same
service.
Stephen
--
Scanned by iCritical.
|