Urgent: needed some specific info about installing lam-mpi

[valkyrite]

Hello,

I want to install lam-mpi on a cluster using "emerge".

I have choosen gentoo as the OS since I have been using it and am comfortable with it. At present I am already running lam-mpi successfully but have installed it manually. This is what I have done.

• 1. Installed lam-mpi (using ./configure --prefix=/usr/local/lam-mpi, make, make install).
  2. Exported the direcotry /home and /usr/local to all other nodes and have configured necessary services for it to run smoothly.

However, I wish to use the lam-mpi ebuild provided with the portge for the purpose. One way I know of is that I install lam-mpi on all the nodes (I can do it using a small script).

But, is it possible (and if yes how) that I install lam-mpi using the emerge script in a particular directory (of my choice) on my master node only such that all the stuff resides in that particular directory and I simply export it to other nodes. It will reduce my job and more importantly I would be able to use the features provided by emerge.

I tried doing:

Code: Select all

env ROOT=/usr/local/lam-mpi emerge -v lam-mpi

And it installed (with a few glitches) almost an entire gentoo system (similar to what I expected) in that directory along with mpi stuff.

The problem is how do I get to run it from such an installation.
What is the procedure to unmerge it. (I can simply delete the entire directory though).

Code: Select all

env ROOT=/usr/local/lam-mpi emerge search lam-mpi

tells lam-mpi is not installed but I can see all the mpi stuff in there.

Any suggestion will be highly appreciated.

--
Tiwaris

[Skywacker]

Sorry I'm not directly answering your question, but why are you doing this?

I have a cluster running lam-mpi with no problems, but lam-mpi is emerged locally on all nodes. I think you are going to take a performance hit running lam-mpi off an nfs share. You're already using the hard drive for the OS, so you're not aiming for diskless nodes, so I have to assume you're trying to ease administration of the cluster. In that case have your own local portage cache on your master for your nodes to update from. Search these forums, there's a couple threads on it.

-Sky

[valkyrite]

Thanks for the reply. Definitly I wanted to ease my administration and also wanted to keep the installation size to a minimum. At present I have installed lam-mpi on the master node (made a package of it) and then using a simple shell script, I installed the compiled package on all other nodes.

I have exported /usr/local/ and have installed all the custom stuff in that place (intel compilers, mpich etc.) It makes my job easy to do and takes less time.

Also isn't it redundant that you have to install it on all the nodes for it to work. Could you suggest me the best way to go about it (my main priority ofcourse is the peformance and second priority is ease of administration).

--Tiwaris