Arch Linux as Thin Client, PXE, LTSP

Question

Hey there all, been using Arch for about a year now, from Gentoo through Zenwalk, gotta say that Arch is bar none the best so far. I think it's the way that it doesn't try to hide the system internals through abstraction (where every distro has it's own abstraction methodology).
I'm thinking in the future that I'd like to do a network-wide deployment of "terminals" for a non-profit. I say "terminals" in because unlike a true thin client, the idea was to have PXE booting to NFS and using the local hardware more. I'm fond of the idea of centralizing system management to a single system installation, and using solid-state machines.
VNC, NX, RDP seem to be out of the question because I need clients to have native multimedia applications like the Flash plugin, and Skype. I could see audio and video redirection becoming a pain. But having the full installation on the client doesn't seem like the right thing to do because then I've got specific machines with specific problems that I'd have to diagnose. Virtualizing an entire OS session for each user would require a powerhouse server, and then we are stuck with a graphical redirection like VNC, NX, RDP. Linux Terminal Server Project really seems to be on track, they've got a great customization method, but again, they're using XDMCP and sending raw X data over the network.
I envision the clients as being mostly web browser-centric to interface with the necessary applications, and using software like Sun Global Desktop or NoMachine NX to connect to individual Windows applications on a Windows server when needed.
I'm looking at the Fit-PC2 as a possible thick-client solution. It's an Intel Atom based PC with an AMD Geode graphics solution, 512MB of soldered RAM, really low energy consumption.
The wiki doesn't seem to address having more than one client using an NFS-mounted root. If you have more than one client writing .pid and other /var files, this doesn't seem like a feasible idea.
So I suppose that the / has to be mounted in RAM, and only the static dirs like /usr /bin /boot swap are NFS imports?
At first the network would be homogenous, but if we add computers with differing graphics cards
So, in summary:
* Any clues on multiple clients using one NFS root?
* How to autoconfigure X.org and other differing hardware?
It sounds like a fun challenge, wondering if anyone else has tried anything like it.
Last edited by VitaminJ (2009-10-19 00:42:31)

Answer 1

I don't really know enough about this to offer any real advice, but as far as multiple machines using a single NFS root goes, you could look into mounting it on all your clients read-only, and using a union filesystem (eg aufs2, unionfs), such that you can use another partition (say on a ramdisk) to handle writes...
Just a thought