Hello, I am trying to write a program which runs using OpenMPI on our cluster and communicates the results using TCP/IP to another machine which has a network connection to the cluster. I am using Boost.Asio for the TCP/IP connection. If I remove the MPI code from the executable and compile it with g++, it connects to the remote machine without a problem. But when I add a simple MPI code to it and compile with mpicxx and execute using mpirun, it generates a connection time out in the call to socket.connect(). Any idea what might be going wrong ? Thanks. Your Mail works best with the New Yahoo Optimized IE8. Get it NOW! http://downloads.yahoo.com/in/internetexplorer/
On 14 Feb 2010, at 06:33, vibhor aggarwal wrote:
Hello,
I am trying to write a program which runs using OpenMPI on our cluster and communicates the results using TCP/IP to another machine which has a network connection to the cluster. I am using Boost.Asio for the TCP/IP connection. If I remove the MPI code from the executable and compile it with g++, it connects to the remote machine without a problem. But when I add a simple MPI code to it and compile with mpicxx and execute using mpirun, it generates a connection time out in the call to socket.connect(). Any idea what might be going wrong ?
Have you tried compiling it with mpicxx without adding any MPI code to it? Matthias
Hi, I have tried compiling with mpicxx and no MPI code in it. It compiles fine, but I still have to use mpirun to executed the generated binary and it does a connection time out. Kind Regards Vibhor ________________________________ From: Matthias Troyer <troyer@phys.ethz.ch> To: boost-users@lists.boost.org Sent: Mon, 15 February, 2010 12:44:57 AM Subject: Re: [Boost-users] [Boost.Asio] Compiling Boost.Asio with OpenMPI On 14 Feb 2010, at 06:33, vibhor aggarwal wrote:
Hello,
I am trying to write a program which runs using OpenMPI on our cluster and communicates the results using TCP/IP to another machine which has a network connection to the cluster. I am using Boost.Asio for the TCP/IP connection. If I remove the MPI code from the executable and compile it with g++, it connects to the remote machine without a problem. But when I add a simple MPI code to it and compile with mpicxx and execute using mpirun, it generates a connection time out in the call to socket.connect(). Any idea what might be going wrong ?
Have you tried compiling it with mpicxx without adding any MPI code to it? Matthias _______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users Your Mail works best with the New Yahoo Optimized IE8. Get it NOW! http://downloads.yahoo.com/in/internetexplorer/
2010/2/15 vibhor aggarwal <vibhor.aggarwal@yahoo.com>:
Hi, I have tried compiling with mpicxx and no MPI code in it. It compiles fine, but I still have to use mpirun to executed the generated binary and it does a connection time out. Kind Regards Vibhor
This is strange: with OpenMPI you should have to use mpirun to start a 1 process job. I'm using OpenMPI with Boost.ASIO and I never encountered your issue. Matthieu -- Information System Engineer, Ph.D. Blog: http://matt.eifelle.com LinkedIn: http://www.linkedin.com/in/matthieubrucher
Hi, Could you let me know which version of Boost.Asio and OpenMPI you are using ? Thank you. Vibhor ________________________________ From: Matthieu Brucher <matthieu.brucher@gmail.com> To: boost-users@lists.boost.org Sent: Mon, 15 February, 2010 10:15:17 AM Subject: Re: [Boost-users] [Boost.Asio] Compiling Boost.Asio with OpenMPI 2010/2/15 vibhor aggarwal <vibhor.aggarwal@yahoo.com>:
Hi, I have tried compiling with mpicxx and no MPI code in it. It compiles fine, but I still have to use mpirun to executed the generated binary and it does a connection time out. Kind Regards Vibhor
This is strange: with OpenMPI you should have to use mpirun to start a 1 process job. I'm using OpenMPI with Boost.ASIO and I never encountered your issue. Matthieu -- Information System Engineer, Ph.D. Blog: http://matt.eifelle.com LinkedIn: http://www.linkedin.com/in/matthieubrucher _______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users The INTERNET now has a personality. YOURS! See your Yahoo! Homepage. http://in.yahoo.com/
Have you checked which compiler, compiler and linker options the mpicxx script uses? Maybe one of those is causing your problems Matthias On 15 Feb 2010, at 01:41, vibhor aggarwal wrote:
Hi,
I have tried compiling with mpicxx and no MPI code in it. It compiles fine, but I still have to use mpirun to executed the generated binary and it does a connection time out.
Kind Regards Vibhor
From: Matthias Troyer <troyer@phys.ethz.ch> To: boost-users@lists.boost.org Sent: Mon, 15 February, 2010 12:44:57 AM Subject: Re: [Boost-users] [Boost.Asio] Compiling Boost.Asio with OpenMPI
On 14 Feb 2010, at 06:33, vibhor aggarwal wrote:
Hello,
I am trying to write a program which runs using OpenMPI on our cluster and communicates the results using TCP/IP to another machine which has a network connection to the cluster. I am using Boost.Asio for the TCP/IP connection. If I remove the MPI code from the executable and compile it with g++, it connects to the remote machine without a problem. But when I add a simple MPI code to it and compile with mpicxx and execute using mpirun, it generates a connection time out in the call to socket.connect(). Any idea what might be going wrong ?
Have you tried compiling it with mpicxx without adding any MPI code to it?
Matthias
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
Your Mail works best with the New Yahoo Optimized IE8. Get it NOW!._______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
Hi, I have checked the OpenMPI settings using ompi_info and they seem to be correct. I did a bit more of debugging and found that it seems to be an issue related to multiple network interfaces on the headnode of our cluster. There are three interfaces on it, one ethernet for communication between nodes, second infiniband for the connection between the nodes and the third ethernet for connecting the cluster head node to rest of the network. When I pass the IP address of the local cluster networks, it gives me a connection error (because there is no server on those ip machines, but the connection does not time out and says connection refused). But, when I pass the IP address of a machine connected via the third interface, which runs the server, it does a connection time out. If I pass the hostname to the boost::asio::ip::tcp::resolver for the machine on third interface I get the following error: Host not found (non-authoritative), try again later. It seems to me that with the mpicxx/mpirun and Boost.Asio there seems to be some sort of confusion/conflict in handling multiple network interfaces, which might be the cause of my problem. Kind Regards Vibhor ________________________________ From: Matthias Troyer <troyer@phys.ethz.ch> To: boost-users@lists.boost.org Sent: Mon, 15 February, 2010 2:25:34 PM Subject: Re: [Boost-users] [Boost.Asio] Compiling Boost.Asio with OpenMPI Have you checked which compiler, compiler and linker options the mpicxx script uses? Maybe one of those is causing your problems Matthias On 15 Feb 2010, at 01:41, vibhor aggarwal wrote: Hi,
I have tried compiling with mpicxx and no MPI code in it. It compiles fine, but I still have to use mpirun to executed the generated binary and it does a connection time out.
Kind Regards Vibhor
________________________________ From: Matthias Troyer <troyer@phys.ethz.ch>
To: boost-users@lists.boost.org Sent: Mon, 15 February, 2010 12:44:57 AM Subject: Re: [Boost-users] [Boost.Asio] Compiling Boost.Asio with OpenMPI
On 14 Feb 2010, at 06:33, vibhor aggarwal wrote:
Hello,
I am trying to write a program which runs using OpenMPI on our cluster and communicates the results using TCP/IP to another machine which has a network connection to the cluster. I am using Boost.Asio for the TCP/IP connection. If I remove the MPI code from the executable and compile it with g++, it connects to the remote machine without a problem. But when I add a simple MPI code to it and compile with mpicxx and execute using mpirun, it generates a connection time out in the call to socket.connect(). Any idea what might be going wrong ?
Have you tried compiling it with mpicxx without adding any MPI code to it?
Matthias
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
________________________________ Your Mail works best with the New Yahoo Optimized IE8. Get it NOW!._______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
The INTERNET now has a personality. YOURS! See your Yahoo! Homepage. http://in.yahoo.com/
participants (2)
-
Matthias Troyer -
Matthieu Brucher -
vibhor aggarwal