LAM/MPI logo

LAM/MPI General User's Mailing List Archives

  |   Home   |   Download   |   Documentation   |   FAQ   |   all just in this list

From: Brian Barrett (brbarret_at_[hidden])
Date: 2005-03-30 09:38:43


On Mar 30, 2005, at 6:21 AM, Marcin wrote:

> Can I start lam cluster without
> using module for example ssh.
>
> I want to start demon lamd on node
> and after that add it to cluster.
> How do I have to start lamd on
> nodes and master node.

I'm afraid I don't exactly understand what you are trying to do or what
you are asking. So I'll go into some detail about how LAM does it's
booting, and if I don't answer your question, please let me know and
maybe try to add some more detail.

LAM daemons have to be "wired up" when they start - including
determining how to contact their peers. The lamboot application is
used to start the daemons on all the nodes and coordinate this "wire
up" phase. lamboot uses a component architecture for booting (the boot
ssi - see the LAM/MPI web page for more information). The "wire up" is
done via TCP sockets connecting back to the lamboot process from the
started LAM daemons.

Once the daemons are "wired up", the only way to expand the set of
connected daemons is to use the lamgrow command on a node that is
already running a LAM daemon. The lamgrow command uses the boot ssi
for starting up the new LAM daemon on the remote host, then uses a TCP
socket callback to find the connectivity information for the new
daemon, then sends it to all existing LAM daemons.

There are boot components for using rsh/ssh, the PBS batch scheduler's
TM interface, and the SLURM batch scheduler. The second two never use
rsh/ssh but instead use the process control interfaces available from
the batch scheduler.

There really isn't a way in LAM right now to have two independently
started LAM daemons to "join up". There are a number of issues in
doing so - each LAM daemon needs a unique id (a low integer from 0 to N
- 1 or so) - having daemons with the same id would be very bad. If
this problem could be overcome, it is possible that the interfaces into
the lam daemon would allow joining two LAM daemons that are already
running. But we certainly haven't tried it and it isn't something that
is on our schedule for future enhancements.

Hope this helps,

Brian

-- 
   Brian Barrett
   LAM/MPI developer and all around nice guy
   Have a LAM/MPI day: http://www.lam-mpi.org/