I have just set up my disk-less compute servers yesterday. These compute servers were configured to boot from the network, pick up the operating system from a network server and save all their files on a network file-system. I was surprisingly easy to set up from a Debian base system. I am really glad that I have so many spare machines to toy around with.
I plan to put in a couple of verilog simulators. I wonder if distcc could be modified to accommodate these simulators as well. I will experiment with this and report my findings later.
So, these compute servers will boot up and run distcc that I mentioned earlier. I did not need to do much except to configure dd-wrt to provide the dhcp-boot option. It was just as simple configuring freenas to provide the necessary storage and network service. First time I got to fool around with PXE – fun!
Now that I have a number of machines, I need a way to keep administer and track them all. For administration, I am currently testing out dish, dsh, clusterssh, pssh and kanif distributed shells. There are advantages and disadvantages of each and I kind of think that Kanif (from INRIA) is more suitable for my use.
As for tracking them all, I have actually just set up a wiki on my local home network. I plan to use the wiki to not just track the servers but to document everything I do at home. This includes any sort of configuration and customisations that I perform on the home network so that I have a easy place to refer to in the future, in case I need to replicate the work at any time.
You know what I’ve started to realise? Computing takes on a whole new dimension when you have a dozen machines running on a home network. There is just so much more that can be done with that power of computing.
Sybreon Ones Nought 
I’m not sure I get it; if you do PXE-booting, why do you need to administer in more than one place? Can’t you run updates and modify scripts on the host only? (warning: n00b speaking here…)
If your simulations will be single-threaded individual instances (as per your post yesterday), you might consider using condor to automatically push them out to idle nodes on your cluster. I’ve never set this up, so I can’t promise that it’s easy, but I do use it a lot 🙂
What wiki do you use? I’m using the one that comes with Trac – which also has a bug/task tracker to keep your ideas in… before that I played around with TiddlyWiki, which is really not bad either (and serverless! can travel with you on a stick etc), but since I’m not well-versed in the AJAX stuff it’s hard to hack it… 😛
I meant administering all those virtual machines as well as the cluster. At the moment, I’m just using my fast fingers to do all the repetitive tasks. 🙂 From brief reading, condor seems to be generic enough to be used for the simulation runs. Cool stuff!
I’m testing mediawiki for the moment. I’m planning to use it more for documenting installation and configuration information but I’ve started to realise that I might need some way to attach files as well (such as config file customisations).
Maybe trac might be a better way to manage these things. It could easily do a diff of my customised config file against the default config file. I could also just easily ‘trac’ the entire /etc path if necessary. That’s rather appealing, actually.
For keeping tabs on /etc, I’m using etckeeper now (it’s in the Debian repos). That uses a version-control system of choice, which you could then indeed bind to trac…
Dude, you’re a good friend to have. 🙂