mark@hoist.nlcomm.com writes ("Re: binding a network card to a CPU?"):
> 
> Hmm; here is Jeremy Allison's quote, describing something
> he did with an SGI he was testing:
> 
> "Like Mindcraft, we also bound the interrupts from each network card
> to a processor. This allows concurrent interrupts not to cause a
> processor cache flush, which is an expensive operation."
> 
> I'm not a kernel internals genius; just trying to get a feel for
> what impact these things have on system performance.

This isn't quite the same scenario, but it's similar - using an MPI
ping-pong test between two machines, I get latencies of ~140
microseconds UP kernels and ~180 microseconds with SMP kernels. I
can't remember the bandwidth figures, but they were also higher with a
UP kernel unless the amount of data being transferred was very large.

Peter

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/