[Beowulf] about managment

[John Hearns]

On Fri, 2004-10-15 at 19:38, llwaeva at 21cn.com wrote:
> Hi all, 
>   I am running 8-node LAM/MPI parallel computing system. I found it's
> trouble to maintain the user accounts and software distribution on all
> the nodes. For example, whenever I install a new software , I have to
> repeat the job 8 times!
There are several answers to this question,
which you can learn about by staying on this group, and consulting
online resources.

A quick answer is that you could construct the cluster using one of the
toolkits, such as Rocks or Warewulf - many others.

And a very quick answer to your current dilemma. There are utilities
which allow parallel execution of commands on a set of machines,
or even to have a terminal session in parallel across a set of machines.

Once you have a server (below) you can rsync each node to that.



>  The most annoying thing is that the
> configuration or managment of the user accounts over the network is a
> heavy job. Someone suggests that I should utilize NFS and NIS. However,
> in my case, it's difficult to have an additional computer as a server.
Not meaning to be rude,  but you are wrong there.
Just use one of your compute nodes as the server. The additional CPU
load will not be great.
You should use some sort of centralised account management NIS or LDAP.
Even if you point blank refuse to do that, a cron job to rsync the
relevant files will help cut down your admin load.

And remember - eight machines may not seem a lot. But what happens if
you make a mistake on one machine, or one machine is down when you are
adding an account or software. Are you sure to run identical commands by
hand the next time it is up?