Subject: Re: LFS, unified vm/buffer cache..
To: None <current-users@NetBSD.ORG>
From: Chris Csanady <firstname.lastname@example.org>
Date: 08/30/1995 00:30:41
>john> so qualitatively, what would you say that these benchmarks show that
>john> netbsd lacks? What I picked up was that the page allocation scheme was
>john> perhaps not the best in the world and that netbsd seemed to overrun the
>john> data cache.
>david> Unified buffer & VM cache, and some way around the problem of sync
>david> writes slowing down the disk. (NT's metadata journalling is better
>david> at this, and shouldn't corrupt the disk no matter when you hit
>david> the power switch (reportedly) (allegedly) (or so its said :)
>LFS gives you much better performance time for file creates and removes.
>These are the classic UFS metadata performance-poor operations.
>Is anyone actively pursuing working LFS for -current?
After reading through the 95 usenix paper on LFS, etc.. Id also be interested
in any current work being done with it. BTW, is anyone working on
integrating the newer BSD-LFS sources into the tree? I have not even seen
them, but they supposedly support fragments and have some other nice fixes.
It seems like a nice FS, and id be willing to help with testing, etc. :)
Anyone else have any thoughts on LFS? Does the current implementation work
at all? (in -current)
Also, im curious if any work is being done to move toward a unified vm/buffer
cache. I installed freebsd over the summer(i like NetBSD more :), and these
mods seem to have had a noticeable effect on performance. would it be possible
to borrow(if its decent..) some of this code from freebsd to speed things up?