[mpich-discuss] Fatal error in MPI_Init
Jayesh Krishna
jayesh at mcs.anl.gov
Thu Feb 3 09:47:35 CST 2011
Hi,
You can run your MPI jobs across Windows and Linux machines. However you need to make sure that the machines are homogeneous (The same architecture - or more precisely follow the same data model).
The only process manager available on Windows is SMPD. The default process manager available with MPICH2 on Unix is Hydra.
Try reconfiguring MPICH2 on Unix with the SMPD process manager (./configure ... --with-pm=smpd ...) and let us know if it works for you.
(PS: Please also make sure that you have Windows and Linux binaries available on the respective nodes.)
Regards,
Jayesh
----- Original Message -----
From: "Gustavo Uribe" <gustfuribe at hotmail.com>
To: jayesh at mcs.anl.gov
Sent: Wednesday, February 2, 2011 3:01:21 PM
Subject: RE: [mpich-discuss] Fatal error in MPI_Init
Hello Jayesh:
now I'm trying to connect PC1 and PC2 with differents Operating Systems.
PC1: Windows XP
PC2: Centos 5.5
Is it possible?
At this moment I have configured the IPs:
PC1 (Windows) --> 126.1.20.168
PC2 (Centos) --> 126.1.20.169
I can PING the machines, so both PCs are connected.
But I can't use the command "mpiexec -hosts 2 126.1.20.168 126.1.20.169 helloworld.exe"
I get the following message:
abort: Unable to connect to '126.1.20.169:8676'
sock error: Error = -1
What is the problem?
Will I need to use the new version of MPICH?
Thanks,
Gustavo
> Date: Mon, 31 Jan 2011 15:30:44 -0600
> From: jayesh at mcs.anl.gov
> To: gustfuribe at hotmail.com
> CC: mpich-discuss at mcs.anl.gov
> Subject: Re: [mpich-discuss] Fatal error in MPI_Init
>
> Hi,
> Good to know MPICH2 is working for you now.
> Typically Infiniband adapters allow you to run TCP/IP apps above IB (They support TCP/IP sockets). Looking at the error message that you got it looked like the particular Infiniband adapter did not provide a service provider for TCP/IP apps. Anyway, good to know it is working for you now.
> The next release of MPICH2 will include a channel (nd) that supports IB (mpiexec -n 2 -channel nemesis:nd mympipgm.exe would work). You might want to try that out.
>
> Regards,
> Jayesh
>
> ----- Original Message -----
> From: "Gustavo Uribe" <gustfuribe at hotmail.com>
> To: jayesh at mcs.anl.gov
> Sent: Monday, January 31, 2011 3:12:20 PM
> Subject: RE: [mpich-discuss] Fatal error in MPI_Init
>
> Jayesh:
> I appreciate your quick response!
> Finally I could run the "HelloWorld" program!
> The problem was that I was not running smpd.exe in the 2nd PC.
> Maybe MPICH2 libraries supports the "Mellanox" InfiniBand adapters I have on both PCs.
> Now, I can run: mpiexec -hosts 126.1.20.168 126.1.20.169 HelloWorld.exe
> and it functions!!!
> I don't know how, but it functions!
>
> Thanks,
> Gustavo Uribe
>
>
> > Date: Mon, 31 Jan 2011 09:33:47 -0600
> > From: jayesh at mcs.anl.gov
> > To: gustfuribe at hotmail.com
> > CC: mpich-discuss at mcs.anl.gov
> > Subject: Re: [mpich-discuss] Fatal error in MPI_Init
> >
> > Hi,
> >
> > >> Can you provide a custom build for me? how much would it cost? or would it be free ?
> >
> > MPICH2 is open-source and available at no cost (free). I can provide a custom build for you, however we recommend all users to use stable releases for production environments. Also, the version that I provide you will be an experimental support for Network Direct (performance tuning is in the works... and I haven't had a chance to run it on a large system to weed out all the bugs.).
> > I would recommend buying "Using MPI" (http://www.mcs.anl.gov/research/projects/mpi/usingmpi/) or any other introductory book on getting started with MPI. There is also a lot of material on MPI if you google for it.
> > I will provide you a custom build in a couple of days.
> >
> > Regards,
> > Jayesh
> >
> > ----- Original Message -----
> > From: "Gustavo Uribe" <gustfuribe at hotmail.com>
> > To: jayesh at mcs.anl.gov
> > Sent: Sunday, January 30, 2011 5:26:47 PM
> > Subject: RE: [mpich-discuss] Fatal error in MPI_Init
> >
> > Jayesh:
> >
> > thanks for your quick response !
> > I'm a little complicated with time , I need it soon .
> > Can you provide a custom build for me? how much would it cost? or would it be free ?
> >
> > I'm just getting started with MPI, and InfiniBand networks . That's why I wanted to ask if you can send me a manual or tutorial where I can learn a little more about it .
> >
> > Thanks,
> > Gustavo Uribe
> >
> >
> > > Date: Fri, 28 Jan 2011 15:59:22 -0600
> > > From: jayesh at mcs.anl.gov
> > > To: mpich-discuss at mcs.anl.gov
> > > CC: gustfuribe at hotmail.com
> > > Subject: Re: [mpich-discuss] Fatal error in MPI_Init
> > >
> > > Hi,
> > > It looks like the adapter (driver) does not allow creating TCP/IP sockets (There is no service provided for TCP/IP).
> > > The next version of MPICH2 on Windows will have a Network Direct module that should allow you to run your application over Infiniband. I don't know when the next version will be out (~ 1-2 months), if you need it sooner please let me know and I can provide a custom build for you.
> > >
> > > (PS: Currently MPICH2 on Windows has support for only plain TCP/IP sockets. It does not support Winsock direct or Network Direct.)
> > > Regards,
> > > Jayesh
> > >
> > > ----- Original Message -----
> > > From: "Gustavo Uribe" <gustfuribe at hotmail.com>
> > > To: mpich-discuss at mcs.anl.gov
> > > Sent: Friday, January 28, 2011 10:52:41 AM
> > > Subject: [mpich-discuss] Fatal error in MPI_Init
> > >
> > >
> > > Hello everybody!
> > >
> > > I'm newbie with MPI, so I think I need your help!
> > > I want to tell you what is my problem:
> > > I have 2 computers connected with Infiniband adapters. On both PCs I'm running Windows XP, and I have installed the correspondent drivers. Then I configured the subnet, and it's possible for me to ping one PC with the other.
> > > After that, I installed MPICH2 (for Windows).
> > >
> > > I want to run a simple "Hello World" program (programmed in C), but in the moment when I want to run the program using MPIEXEC, I obtain the following message:
> > >
> > > Fatal error in MPI_Init: Other MPI error, error stack:
> > > MPIR_Init_thread<388>..........:
> > > MPID_Init<315>.................: channel initialization failed
> > > MPDIDI_CH3_Init<38>............:
> > > MPID_nem_init<234>.............:
> > > MPID_nem_newtcp_module_init<90>:
> > > MPID_nem_newtcp_module_init<89>: unable to create a socket. No se puede cargar o inicializar el proveedor de servicios solicitado.
> > > <errno 10106>
> > >
> > >
> > > Am I doing something wrong?
> > > Please! Help me! I'm newbie with MPI.
> > >
> > > Thanks!
> > >
> > > Gustavo
> > > >From Bariloche (Argentina)
> > >
> > >
> > >
> > > _______________________________________________
> > > mpich-discuss mailing list
> > > mpich-discuss at mcs.anl.gov
> > > https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> >
More information about the mpich-discuss
mailing list