[Beowulf] Re: Re: Home beowulf - NIC latencies
lindahl at pathscale.com
Thu Feb 17 22:44:34 PST 2005
On Tue, Feb 15, 2005 at 07:34:50AM -0700, Keith D. Underwood wrote:
> > c) ban of the ANY_SENDER wildcard: a world of optimization goes away
> > with this convenience.
> Um, our apps guys say this is more than a convenience. Apparently,
> sometimes you don't exactly know who you are going to receive from.
> Would you rather them post receives from 4000 nodes and cancel the ones
> that don't send to that node after a while?
That's a good reason to use it. The fundamental problem is that
supporting both ANY_TAG and ANY_SENDER efficiently is really
annoying. So most implementers would prefer to support ANY_TAG
efficiently and ANY_SENDER less efficiently.
And then those pesky users (life is much simpler when you only run
benchmarks that you write yourself, you know ;-) go use ANY_SENDER,
usually because they think it's faster when say 4 nodes are sending
you something at roughly the same time. Instead, we'd prefer that they
use Irecv for that.
More information about the Beowulf