Newsgroups: comp.parallel.pvm
Path: ukc!uknet!EU.net!howland.reston.ans.net!europa.eng.gtefsd.com!library.ucla.edu!csulb.edu!csus.edu!netcom.com!salekar
From: salekar@netcom.com (Sudhir Alekar)
Subject: PVM hangup
Message-ID: <salekarCpAMoJ.62v@netcom.com>
Organization: NETCOM On-line Communication Services (408 241-9760 guest)
X-Newsreader: TIN [version 1.2 PL1]
Date: Wed, 4 May 1994 19:41:07 GMT
Lines: 42

I would like to find out if anybody has
an experience with the hangup in the 'pvm_send'.
We are running pvm (3.0) on IBM RS/6000 cluster.
The virtual circuit consists of two nodes of
IBM 560, each running one child process.
The machines have 128MB memory and 400MB virtual memory.

The problem comes when the two child processes
try to exchange data with pvm_send/pvm_recv.

They do communicate for a while, but BOTH hang in
'pvm_send' at same place. At that point there is plenty of virtual
space left (200MB), but little real memory (0.5 MB)

Here is a trace:

read(0xffffffff, 0xdeadbeef, 0xdeadbeef) at 0xd000ef50
tread(d = 6, b = "", n = 16), line 869 in "lpvm.c"
mxinput(pcbp = 0x202196c8), line 907 in "lpvm.c"
mxfer(mid = 32, block = 0), line 1248 in "lpvm.c"
mroute(mid = 32, dtid = 262154, code = 11, block = 0), line 606 in "lpvm.c"
pvm_send(tid = 262154, code = 11), line 856 in "lpvmgen.c"


It is trying send a message of about 14700 real numbers.


We are using default UDP communication.


Any sharing of experiences will be appreciated.


-Sudhir Alekar
(sudhir@centric.com)

Centric Engg Systems
3801 East Bayshore Road
Palo Alto, CA 94303

(415)960-3600 Phone


