Re: ssh update and parallel jobs execution
Le Jeudi, 12 Décembre 2002 12.25, Junichi Uekawa a écrit :
> At Thu, 12 Dec 2002 09:25:09 +0100,
>
> Fabrice Yerly wrote:
> > Dear happy Debian users,
> >
> > After the update of ssh to version OpenSSH_3.4p1 Debian 1:3.4p1-1, SSH
> > protocols 1.5/2.0, with passwordless connection through the cluster's
> > nodes, (see archives around 20 Nov 2002) I have already some problems...
> >
> > Jobs running on 1 processor works fine. But jobs running in parallel
> > stops abnormally returning me an error:
> > pc001: Connection refused (where pc001 is one of the nodes)
> >
> > I use OpenPBS and I don't know what do do... Before the update everything
> > worked fine. If I connect manually on any node it works. I don't see in
> > my ssh configuration files what forbids a multiple connection.
>
> What is it that you mean by parallel invocation ?
I compiled a program (AMBER) using the Message Passing Interface (MPI)
through the Mpich 1.2.1-16 library.
> It might be that ssh is refusing to start up without a valid
> pty, or valid standard input that is interactive.
OK, but I wonder why before ssh agreed to start up... And also the same
program works if I tell him to use only one processor. Why sould it be
different?
Thanks for help
Fabrice
Reply to: