The installation of an SGI workstation uses an install server, as we mentioned above. The install server acts as an SGI server for a Diskless station.
The server needs a MINIROOT and the information telling it which named station boots which MINIROOT. To avoid enumerating all station which are allowed to boot the MINIROOT (1.2.5) we use wildcarding in the bootparams file. This makes for easy maintenance of the bootparams file but requires the install server to be isolated on a subnet, away from other diskless Sun (real diskless Suns which would be confused otherwise) (1.2.6). In addition the SGI installer needs a bootpd. The one used on the SGI is the one that is used on the DEC bootp server which is based on the BSD 4.3 bootpd.
As with the Sun the MINIROOT is a work of art. But unlike the Sun the SGI does not use a MINIROOT for their CDs. They use a Stand Alone SHell (SASH) so we do not have that to follow. So we follow the instructions for creating a diskless workstation server in the adminguide, we enter all the hardware options we expect to need at this point. We then start finding all the files the machine needs to boot and make them symlinks to /tmp and /tmp is a symlink to /root/tmp. This dictates that we have to grab control of the machine very early on in the process - init. /root/dev/ is populated and symlinks from /dev are made. and /root/dev/dsk,rdsk is a symlink to /etc/mit/dev/dsk,rdsk. This is so that once we have root partition we can mount it with impunity and then have devices that can be read and written.
Copy in programs that are needed in the install that reside in /usr/sbin into /sbin or /etc. Copy in programs that are not available on the SGI gtar.