LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Leif Nixon
said:
> > If people take that view the grid will fail.
>
> Ehm, no. Thing is, the world is heterogenous. It is impossible to make
> all sites look identical.
In my view that's exactly what a grid is supposed to do - not absolutely
identical, but similar enough that a job can run anywhere without
needing special treatment for each site. Of course, it may indeed be
impossible, in which case we will fail.
> Heterogeneity will be there, whether you want it to or not. Find
> ways to deal with it, like writing portable (*gasp*!) software,
> and you will be able to run on a hugely larger set of resources.
Grids are supposed to help users, the complexity and heterogeneity
should be hidden by the middleware and not left to applications to deal
with - that's why it's called middleware. If the installations at
different sites are all different then big VOs like the LCG experiments
will just implement everything themselves (we already see this) and
small VOs will give up. What's the point of the middleware developers
writing (say) data management code if it's only installed at 20% of
sites and "portability" means that every application has to have its own
data management code to deal with the other 80%?
Stephen
|