I guess this question is the most frequently asked on earth! Here is
a summary of what worked for me and ANOTHER QUESTION (so please don't
move on just yet...:-)
Our configuration: A Sun 3/160, running 4.0.3, with a 7053 controller
and a 688 Mbyte SMD disk.
The problem: I needed more inodes and 2048 inodes/group is hardcoded into
kernel code. So more inodes must be milked in more sneaky ways.
What I tried:
(1) Just halving the number of cylinders / group -- using a -c 8 to
newfs. No go... because blocksize is 8k, cyl/group must be
multiples of 16 (because of computations involving the number of
sectors/track).
(2) Bump the blocksize to 4k and the cyl/group to 8. This made it
past newfs and mkfs, and the file system mounted, but the system
crashed when any use was made of the file system (e.g., a "cd").
Reason: on a Sun-3, page sizes are 8k, thus blocksizes must be 8k.
(3) SOLUTION: Trim the number of sectors the filesystem uses by 1. On
our disk, we have 67 sectors/track. I built the filesystem with
66 sectors/track. This allows the cyl / group to be set to 8 with
the blocksize still at 8k. And -- voila! -- we double our inodes!
We do indeed lose a few kilobytes, but diskspace was not the issue,
inodes were!
And now to my new question: I lied to mkfs about the number of sectors
in the partition. However, I DID NOT REPARTITION THE DISK. I simply
mkfs'd and everything seemed OK. Will something sneak up on me? (I'm
obviously trying to avoid extended down time here...). That is, is it pertinent
to do a disk repartition?
Thanks to everyone who responded to me with VOLUMES of material. These
include:
uswmrg2!james@boulder.Colorado.EDU
trinkle@cs.purdue.edu
bradshaw@qucis.queensu.ca
matt@oddjob.uchicago.edu
bit!markm
muon!baumann@ucrmath.ucr.edu
ensys!msm
love@Stars.Reston.Unisys.COM
dws@EBay.Sun.COM
scooper@NMSU.Edu
Mike Jipping Internet: jipping@cs.hope.edu
Hope College BITNET: JIPPING@HOPE
Department of Computer Science Voice: Hey!
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:05:56 CDT