Re: communication pattern causes corrupted data

Hi Josh,

On Thu, 2003-03-20 at 18:13, Joshua J. England wrote:
> To my understanding, with certain communication patterns (ie: buffer
> sizes sent in X-node runs), the last 4 bytes of the buffer is often not
> received.  When sending an array of integers then, the last integer
> becomes corrupted.  Changing the buffer sizes make the problem go away.
> 
> I've included a test program that demonstrates this.  Basically, it

I have tried your code on different configurations:

x86/Debian:
gm-1.6.3/gm-1.6.4-rc0, mpich-1.2.5..9/mpich-1.2.5..10-pre7

IA64/RH:
gm-1.6.3, mpich-1.2.5..9/mpich-1.2.5..10-pre7

It ran all night yesterday and all day today, and all of the runs
completed without errors. Is the probability of corruption large enough
to happen at least once in one run ?

I don't have enough alphas, so I used Itaniums to try a 64-bits
architecture, but it worked fine on them too.

I am out of options here. I suppose your machine is secure and
unaccessible, but I need to reproduce it somewhere to eventually fix it.

Does anybody on the list has been able to reproduce failures with Josh's
test ? It would be great if you can give a quick try. It's very easy to
build and to run, you just need something to run 16 processes.

Patrick
-- 

Patrick Geoffray, PhD
Myricom, Inc.
http://www.myri.com