Newsgroups: comp.parallel.pvm
From: mwolff@casa.Colorado.EDU (Micheal Wolff)
Subject: group library function on ALPHA
Organization: CU
Date: 13 Nov 1996 10:41:48 -0700
Message-ID: <uklloc551ur.fsf@casa.Colorado.EDU>


Hello,

I have encountered a strange (i think so anyway) problem using
pvm_reduce (and pvmfreduce) on the ALPHA architecture.  to elucidate
the matter quickly, when i run the sample program trsg (which included
as example in the pvm3 distribution), i get the following --

-adcaxp(wolff)-46->trsg

 *** Example use of PVM Reduce, Scatter, and Gather *** 
 Number of processors to use (1-16)? : 2
 Number of elements per row to use (1-20)? : 10
 INPUT values: nprocs = 2, count = 10 
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: gs_cachegroup: End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: gs_cachegroup: End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer
libpvm [t40020]: pvm_upkint(): End of buffer

i was running on a cluster of two alpha machines.  this problem was
first noted when i ported (or attempted to do so) a program written
in the Sun environment (works under both SUN4 and SUNMP).  i
received the same type of errors (i.e. pvm_upinkt(): End of buffer).
The above was compiled from PVM3.3.10

I have poured through my manual and the PVM website but haven't been
able to address the problem satisfactorily.

Does anyone have any suggestions?  I suspect I have missed something
obvious, but after a day looking for it, I am clearly not bright
enough to find it.

thanks in advance for any responses (either posts or email).

                        --mike wolff
                        wolff@colorado.edu

