On Tue, 14 Oct 2003, Erwan Velu wrote:
> > If you run "ping master.domcomp.com" from a compute node which IPs are
> > used ? There might be a name resolution problem...
>
> Don't care about that, it works fine.. the resolution is working.
>
> > > It is possible to ask lamboot to give the nodes the IP address of the
> > > NIC I want and not giving the IP Address that equals my hostname ?
> >
> > Sure. I do this all the time (although not with LAM 7.x yet).
>
> The same config was working on lam 6.5.9
>
> > >Some of the nodes have a Gigabit NIC which is used for computation and a FastEth one
> > used for file-serving and remote (ssh) access. The trick is to specify in
> > the LAM boot file the names coresponding to the NIC you want to use. In my
> > case I have nodexxx (corresponding to FastE NIC's IP) and gigaxxx
> > (corresponding to GigaE NIC's IP) names and use the gigaxxx ones in the
> > bootfiles - in this case also the rsh/ssh connections to start the LAM
> > daemons are made through these interfaces.
>
> I was doing that using 6.5.9 but it doesn't works for the reason i've
> told before.
Let me make sure I understand -- are you lambooting with the names that
correspond to the 10.x addresses, or the names that correspond to the
172.x addresses?
LAM should be using on the names (and corresponding resolved addresses)
that you gave in the boot schema file, and no others. So if you gave
hostname_for_10 in the boot schema file and LAM ended up using the 172.x,
then either something is wrong in your setup or we have a bug in LAM.
Can you send the entire output of "lamboot -d"?
-----
FYI, there is, however, a boot SSI parameter that can be used to allow
promiscious connections during lamboot. This is definitely a change from
6.5.x to 7.x -- in all versions prior to 7.x, we allowed promiscious
connections without question. Now we don't (unless you specifically
disable the checking). See page 59 in the 7.0.x user manual. The SSI
parameter to disable this stuff is:
boot_base_promisc
Set it to a value of 1. For example:
shell$ lamboot -ssi boot_base_promisc 1 boot_schema_file
or
shell$ setenv LAM_MPI_SSI_boot_base_promisc 1
shell$ lamboot boot_schema file
(remember that all SSI parameters actually get mapped down to environment
variables) The second method might be preferable for you -- you can set
that env variable in some system-level shell startup file, and then your
users don't ever need to know about it. :-)
Also FYI: in 7.1, there is a new SSI parameter called mpi_hostmap that
allows you to boot one one set of IP addresses and then use an entirely
different set of IP addresses for MPI communication (which is even better
than what you're doing here, potentially -- you can lamboot on the admin
network and automatically have all MPI traffic go on the high speed
network).
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
