GeekForce

We all know that outside of the hardware, the OS is the most important component of a well behaved system. But this issue becomes more complex when you start to build a cluster environment because maintenance becomes more of a hassle.

The reason for this is that so much is shared in a clustered environment. Lets just talk about something simple like drivers, and the kernel: When I started to build my cluster, I built each machine separately, this means that i ran the redhat installer on each one, edited the /etc/hosts file. But it means more than that, I also made sure that all the right drivers were installed for each machine. But we already said that i started this whole project with a homogeneous network. Why should I install all of this on each machine? Why not set up a bootp server, and let each machine get it’s info from the network.

It turns out that this is alot harder to do than it seems at first blush. Network booting is easy, but you still have to install the OS on each node, so that it has a root filesystem, and all the other support tools and services it needs.

Bootp is 80% of the equation, but the other 20%is the more difficult to adminster part. This share of the work consists of creating a NFS root directory that is mountable by each node, and having a network boot kernel that is flexible enough to allow a new node to be added without having to redo it to include new drivers.

I had the beginnings of this situation with OpenSSI. It created a very robust, easy to administer environment that a sysadmin could walk into with no previous experience. I liked it. Alas, it is now a defunct project. This lead me to think that others had picked up the slack, and this was now a non-issue.

So, I’ve been trying Fedora for the last couple of weeks. I’ve given up. Fedora does not like the old hardware that I have, and when it does, I cant seem to get things to work really well.

I admit that part of the problem is that i have not been in the sysadmin business for some time now. and my chops are rusty. My fundamentals though, are really good!

So, I’ve been looking for a different distro to host the cluster on.

Here are the requirements:

• Easy setup of the OS on the master management node.
• Packages for DHCP, BOOTP, and NFS.
• Hopefully simple set up an NFS based root filesystem for all nodes.

If I can have those three things, I think I’ll be OK with the rest of it. Like I said, I tried Fedora, and now - drum roll - it’s on to Debian!

More later.