[MPICH] SMPD
Rajeev Thakur
thakur at mcs.anl.gov
Fri Jan 11 15:27:17 CST 2008
I would recommend ch3:nemesis instead of ch3:ssm. It is faster.
You can't mix 32 and 64-bit systems with MPICH2.
The shared library error is coming from somewhere outside MPICH2 since you
haven't configured MPICH2 to use shared libraries. Try compiling and running
a simple C program.
Set CFLAGS to the value you want.
Rajeev
_____
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Krishna Chaitanya
Sent: Thursday, January 10, 2008 2:48 PM
To: Dave Goodell
Cc: mpich-discuss at mcs.anl.gov
Subject: Re: [MPICH] SMPD
Hi,
I am basically having one INTEL(32 bit) laptop and temporary access to
another AMD(64 bit) machine and I am trying to trace and understand the
flow of MPICH2 code. I have worked on a couple of other implementations of
MPI, but I am a bit new to MPICH2. When I posted the question, I was trying
to get it to work on one node, by forcing it to use shared memory. I went
through the documentation again and I realised that I can select the device
ch3:ssm, when I run the config script and this gives me more flexibility as
it uses socket between nodes and shared memory for intra-node
communication.Right now, I have coupled up both the machines. However after
doing :
# ./configure --prefix=/home/kc/mpich-install/ --with-device=ch3:ssm CC=gcc
CFLAGS=-g
# make
# make install
on both the nodes, I tried out
# mpiexec -n 2 /bin/hostname and it worked, but, on the AMP machine, it
flashed :
ERROR: ld.so: object '/usr/lib/libtsocks.so' from LD_PRELOAD cannot be
preloaded: ignored.
And on trying another with another simple MPI code, I got the following
messages :
$ mpiexec -n 2 /home/kc/sample
32 bit INTEL node :
/home/kc/sample: error while loading shared libraries: libuuid.so.1: wrong
ELF class: ELFCLASS64
and
64 bit AMD node :
ERROR: ld.so: object '/usr/lib/libtsocks.so' from LD_PRELOAD cannot be
preloaded: ignored.
/home/kc/sample: error while loading shared libraries: libuuid.so.1: wrong
ELF class: ELFCLASS64
ERROR: ld.so: object '/usr/lib/libtsocks.so' from LD_PRELOAD cannot be
preloaded: ignored.
With respect to the ELF message, i guess the problem is due to the
compatibily issue which could be handled by using a CFLAG= -xarch=x86_64 on
the AMD machine. But, the configure script does not recognise this
CFLAG(correct me, if I am wrong) . Is there someway i can compile the MPICH
library with this flag?
Thanks,
Krishna Chaitanya K,
National Institute of Technology,Karnataka,Surathkal (NITK)
India
On Jan 10, 2008 12:06 PM, Dave Goodell <goodell at mcs.anl.gov> wrote:
Hi Krishna,
On my linux box, errno 98 means "Address already in use". I think
that the default port for smpd is 8676, so you might try looking for
something else that is already bound to that port:
% netstat -an | grep 8676
-Dave
On Jan 10, 2008, at 4:36 AM, Krishna Chaitanya wrote:
> Hi,
> I am trying to launch mpiexec on a single node and I ran
> the configure script with the " --with-pm=smpd" and the smpd file
> was installed in the bin subdir. However, when i try to launch the
> smp daemon, I get the following :
>
> ./smpd -s
> Please specify an authentication passphrase for this smpd:
> Would you like to save this passphrase in '/home/kc/.smpd'? yes
> [00:2915]..ERROR:MPIDU_Sock_listen failed,
> sock error: generic socket failure, error stack:
> MPIDU_Sock_listen(369): unable to bind socket to port
> (port=135207912,errno=98:(strerror() not found))
>
> It would be of help if someone could let me know how to
> troubleshoot this.
>
> Thanks,
> Krishna Chaitanya K
> --
> In the middle of difficulty, lies opportunity
--
In the middle of difficulty, lies opportunity
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20080111/e958f9be/attachment.htm>
More information about the mpich-discuss
mailing list