Newsgroups: comp.parallel.pvm
From: Ingo Reise <reise@informatik.uni-wuerzburg.de>
Subject: Re: PVM 3.3.9 / pvm_upk... End of buffer on SUNMP
Organization: University of Wuerzburg, Germany
Date: Mon, 16 Oct 1995 14:21:34 +0100
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <30825C5E.41C6@informatik.uni-wuerzburg.de>

Hello!

As Stefan described, we've got a similar problem. A few days ago I've got
the following answer on my posting:

In article <41v0bq$sj6@winx03.informatik.uni-wuerzburg.de>,
Ingo Reise  <reise@informatik.uni-wuerzburg.de> wrote:
>Hello!
>
>We've got a problem sending pvm-messages from our ALPHA-cluster to our ALPHAMP
>and the other way round.
>
>Out test-program works fine on the ALPHAMP and between the ALPHAs, but any
>packet bigger than 3 kilobyte sent from one architecture to the other results
>in:
>
>[t80040000] (time) [t80001] libpvm [t80001]: pvm_upkbyte(): End of buffer
>[t80040000] (time) [t80001] slave cork: pvm_upkbyte failed (7)
>(The second message is my one,
> all other used pvm-calls are ok.)
>
>Sending packets up to 3 kilobytes makes no errors and the packet is received
>complete and correct (tested).
>
>So only the transmission of packets bigger than 3 kilobytes and from one
>architecture to the other results in errors.
>
>We don't know, if this is a pvm-error or an error in out network-configuration,
>but we don't know where to adjust this 4 kilobyte-value to a bigger one.
>
>Any helpful hints are appreciated,
>Ingo
>

We had a similar problem with passing messages between SGI and
SGIMP machines with similar error messages.  I believe it was a
bug in the SGIMP message passing code.  Our workaround was to
modify the pvmgetarch to return SGI instead of SGIMP.  This
caused the SGIMP machines to avoid using their internal message
passing schemes and only use the external mechanisms and the
problem went away.  You might try a similar thing to at least
narrow down the problem.  Hope this helps.  You might also try
getting the latest and greatest version as some of these bugs
get fixed in the later releases.  Good luck.


John

john@tone.jpl.nasa.gov
---Signature under construction---
(I quail at the impossibility of capturing the
essence of myself in four lines in an ASCII file)


Perhaps you can try this workaround, since we haven't found a
solution yet.

Greetings
Ingo

-- 


                     __o           
                     \<,           
___________________()/ ()______________________________________________________
Ingo Reise                                      
Robert-Kirchhoff-Strasse 55    Lehrstuhl fuer Informatik II
97076 Wuerzburg                Universitaet Wuerzburg
Tel. 0931/273597               e-mail: reise@informatik.uni-wuerzburg.de

