Configuring a network - with small local disks - Help !!

[Sean Marrett]

A configuration question:

We have recently installed a number of SGI systems, and are trying to put some
order into their configuration. Essentially, we have a SGI 210  (16 Mbyte) with
a 600 Mbyte SCSI disk, one PI 25S (no graphics) (16 Mbyte)  (with two 800 Mbyte
SCSI disk), and  4 PI/35 (with  24 Mbyte of memory each ,and 200 Mbyte disk
drives).  Our question is, how best can we configure these machines in a client
/server setup. Initially, we thought of setting up the 4 PI/35 systems as
diskless clients of the PI/25 (which was purchased as a server) and installing
local swap files on the individual machines, and using the remaining disk space
on the PI/35 as scratch space for users (3-D Brain Images -- lots of data).
However, we have been playing with the SGI software, trying to come up with
some other configurations to speed up local response time. 

The way we understand it, setting up a diskless client sets up two trees. A
client tree (one per machine) and a share tree (shared amongst all clients of
the server). What we though to do would be to move the client trees to each
machine, but keep the share tree on the server. This would seem to be the
optimal solution in terms of reducing redundant file storage across machines,
while keeping a minimum set of local files on each workstation. However, this
does not seem to be a supported solution, and the whole process of installing
new software via install would be a real problem, since NFS must be shut down
when using install.

So, my question is, how are other sites doing this ? Is it stupid to try to set
up small local disks like this (ie should we just pretend they are all
diskless) and just set up swap files on them or what. Please send me your
experiences or suggestions, and I will summarize them for anyone who is
interested.

Sean Marrett
sean at petvax.medcor.mcgill.ca
neuro at moe.mcrcim.mcgill.ca
McConnell Brain Imaging Center
Montreal Neurological Institute