[Beowulf] New HPCC results and the Myri viewpoint

[Greg Lindahl]

On Thu, Jul 21, 2005 at 12:59:55PM +0800, Stuart Midgley wrote:

> Actually, I tend to disagree with your comment here.  The curve tells  
> you one of the characteristics of the network,

In particular, the ping-pong bandwith curve is the lower bound, and
the streaming curve is the upper bound, of what you'll see in real
life.

> In my own experience, I tend to find that most codes are not latency  
> sensitive (that is, QsNetII, Infinipath, Myricom etc

... did I miss you signing up to run on our customer benchmark
cluster? ;-) In any case, I can point you to quite a few sets of
online performance numbers where scalability is being hurt by
short-message performance, which is the thing that everyone means when
they say "latency sensitive". For example, quantum chemistry is very
well known for being hard to scale. Climate involves running small
datasets with large numbers of timesteps, and that inevitably ends up
being latency sensitive. And so on.

-- greg