Formatting Sun disks with a Ciprico controller

[att!ihlpf!marselle at research.att.com]

I've got a question about formatting new disks and creating file systems
on them.  I have a Sun-3/140 with a Ciprico Rimfire 3200 disk controller
which serves one diskless Sun-3/50 client now, and will serve 4 more soon
(when we get them).  I have successfully formatted and partitioned the
disk (a Fujitsu 2344) following some local advice and some Ciprico and Sun
documentation.  However, there are a few places where my knowledge is
fuzzy and I can't seem to find any documentation to clear up the
confusion.  Perhaps this is in the realm of Unix "folklore" and someone in
net-land can help (pointers to articles, books, etc. would also be
appreciated).

My first question concerns the b: partition on the drive, which I
understand is the swap partition.  Is the b: partition always the swap
partition?  Is this a Unix convention?  How large should it be?  Mine is
18 Mbytes.  Is there a rule of thumb for estimating the size of the swap
partition?

Next, what about the space allocated for the /export file system?  Is
there a rule of thumb that says that for n diskless clients, you need
x=f(n) Mbytes of space in the partition which will be used for the /export
file system?  We use 16 Mbytes for the swap space for each client, which
lives in /export/swap/<hostname>.  There's also space required to hold
each client's root file system, in /export/root/<hostname>.  Is there an
estimate for this space?  On one of our other systems, I see this range
from 3500 blocks to 1700 blocks for various clients.  I'd rather not play
a guessing game and end up wasting a lot of space if I overestimate, or
have to re-partition if I underestimate.

Another thing I've noticed is that when I use "newfs" to make a file
system on a partition and then do a "df", the amount of space shown in the
"kbytes" column is less than than the amount I expected when I calculated
partition sizes.  I know that the the "used" plus "avail" amounts are less
than the "kbytes" amount to allow for a 10% (default) reserve.  For
example, I have a /usr file system on my disk which is 264114 512-byte
sectors (blocks), which is 135,226,368 bytes.  But when I do a "df", this
is what I see:

Filesystem            kbytes    used   avail capacity  Mounted on
/dev/rf0d             129320  107194    9194    92%    /usr

129320 * 1024 = 132,423,680 bytes, which is less than 135,226,368.  What's
going on here?  Does Unix snarf up some of the space for inodes and/or
other stuff?  Is there a rule that I can use to allow for this so I don't
get squeezed and end up with less space than I need (this happened to me
before and I had to re-partition the disk to make the /usr file system
larger; but I was just shooting in the dark when I did it).

I've been reading the Sun System & Network Administration manual but all I
find is an explanation of how to set up the partitions; not how to
estimate how large they should be.

Jim Marselle
AT&T Bell Labs
Naperville, IL
att!ihlpf!marselle
(312) 416-4108