[Beowulf] about managment

[Sean Dilda]

On Fri, 2004-10-15 at 14:38, 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! 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.
> Would anyone please share your experience in maintaining the beowulf
> cluster? 

My cluster relies heavily on NIS and NFS.  NIS is used to share user
login information so that I only have to make an account once, then
create the NIS maps.  (I actually have pam configured to authenticate
off of the campus kerberos server on the head nodes, and using ssh's
host-based authentication across the cluster)  We also use NFS for users
home directories.  However, I make a point of trying to package up third
party software that's used and install the rpms on all of the machines
in the cluster.

As for an extra server, as RGB pointed out, NFS and NIS can easily run
on the same box.  Do you currently have a scheduler?  If so, you can run
NIS/NFS on that box.  If nothing else, you can just run them on whatever
box people use to login to the cluster.  The de facto standard for small
clusters is to have a single head node that users login to to launch
jobs, servers out home directories and possibly NIS, as well as any
scheduler you might have.