Diskless (network-booting) load-balanced cluster

[colinb]

I have a rack full of blade servers, all of which are booting over the network and run with a ramdisk root. (Permanent storage is on an NFS share.) These blades will be run targets during development. (A binary is placed in the NFS share from the development machine, and started running over a telnet connection to the blade.)

I'd like to be able to have a virtual host to which I can connect and be handed the address to the first available blade. I'd prefer to have only one developer per blade, so I don't really need any sophisticated load-balancing; each will be marked as either available or unavailable.

What's the best way to do this?

[Januszzz]

You could setup one loadbalancer in front of those machines with LVS kernel enabled and heartbeat installed (LVS is simply right here, in your kernel). This machine could also be doubled in case of troubles.

So you need machine, lvs enabled kernel and heartbeat. I tried this as small device running from cf, but I could not compile heartbeat against uclibc, so my installation was about 300MB.