Gordon, JC (John) wrote:
> Simon, I think you are too cautious. Has anyone seen network saturation yet?
> Put your whole farm on the grid. You can always throttle back if you see a
> problem but I suspect you won't. There are 6700 cpus on LCG now and the
> network is not a limiting factor in the current data challenges. Perhaps I
> should say that overall capacity isn't a problem. There could easily be
> problems caused by the low rates you report slowing jobs down but they don't
> fill up the network.
I would, in general, echo this view: bandwidth is *generally* not a
problem, however the LCHb data challenge has had two sites where
bandwidth has been an issue. One is in Spain where they simply didn't
have the bandwidth to get data out at any reasonable rate, so we just
removed them from our list of "allowed" sites. The other was Karlsrhue,
which is a large site, and our data transfers *appeared* to saturate
their network bandwidth to CERN. There were also various other problems
at FZK due to the size and certain specialisations of that site, so they
*may* have been to blame. I'm not sure bandwidth was every confirmed
100% as the cause of the data transfer backlog.
I'm sorry I don't have the actual (theoretical) bandwidth for the two
sites on hand.
Ian Stokes-Rees [log in to unmask]
Particle Physics, Oxford http://www-pnp.physics.ox.ac.uk/~stokes