Subject: Re: tunefs...
To: None <current-users@NetBSD.ORG>
From: der Mouse <mouse@Holo.Rodents.Montreal.QC.CA>
List: current-users
Date: 08/16/1996 08:44:50
> Is there any good guide to how to decide what numbers to feed
> tunefs/newfs?
Ummm...experience? :-)
> Is interleave good or bad? How do I know?
Interleave, for a modern disk, is bad. You want to set interleave
higher than 1 only when there is a bottleneck close to the disk, such
that bits come off the media fast enough to overrun something else.
For example, if you have a disk with a DMA engine that can't keep up
with the platter rotation speed, and no more than one sector of
buffering, you will need interleave > 1. (Or at least such is my
understanding of interleave; I welcome corrections if I'm wrong.)
> (A similar one: Does making fewer inodes make a file system have more
> space for files?
Yes; the space that would be occupied by unused inodes becomes
available for data blocks. However, each inode buys you only 128 bytes
of data space; you don't really gain a lot of data space unless you
have _lots_ of unused inodes. You can probably gain more by pulling
your frag size down, since that frees up, on the average, half a frag
per small file, which is probably more than the extra bitmap size
required by smaller frags. (Of course, you also can't have the
blk/frag ratio higher than 8, and the default 8k/1k achieves that, so
you probably can't shrink your frags without also shrinking your
blocks. I haven't thought about it enough to know what effect
shrinking blocks will have.)
> Make it faster?
Trading inodes for space shouldn't affect speed significantly.
Shrinking frags and/or blocks will impair performance.
> I can't tell, but my filesystems rarely get above 20% usage of
> inodes, so if it's good, I could easily do it.)
Might as well, then; you won't get a whole lot of space, but you'll get
some. How many inodes does the filesystem have? This, plus the
128-byte on-disk size of an inode, should allow you to calculate about
how much space you'd gain.
der Mouse
mouse@collatz.mcrcim.mcgill.edu
01 EE 31 F6 BB 0C 34 36 00 F3 7C 5A C1 A0 67 1D