Testbed Support for GridPP member institutes [mailto:[log in to unmask]] On Behalf Of Sam Skipsey said:
> What I'm looking for here is actual data for how these jobs perform on a system - how many high-pileup jobs do
> you need on a system, all at once, for it to start swapping with 2GB/core?
It's also going to be a matter of probabilities. As the number of jobs/node goes up the average fluctuations (relative to the total memory) will go down like root N, but the maximum will stay constant. Given the number of jobs in the Grid you don't need much probability to have a few nodes somewhere eating into the swap even if the vast majority don't.