It looks like Apple has netinfo'ized bootp so I can only give vague help.
The first step is bootp. It is a protocol. Machines usually have a small amount of firmware set up where you specify the bootp host's ip address. Usually you specify the client's ip address, netmask, and default route. Technically, this is not required.
bootp can work just off of the ethernet hardware address. It is broadcast out (this introduces limitations because broadcasts do not go across routers). The bootpd sees the bootp packet with the ethernet hardware address and replies back: this is your ip address, etc, etc (much like dhcp). It also replies back: you need to boot from file BLAH. BLAH can actually be on a different host I believe.
Step two is for the client to ask for BLAH. This is done with tftp. So, tftp needs to be set up. Again, all the setup files, etc, appears to have been plopped into netinfo. "man bootpd" gives you some pretty good information.
After that, the network style install I've seen uses nfs to mount the images to install but Sun may have another approach.
Hope this helps