On Tue, 6 Aug 2002, Greywolf wrote:

> On Tue, 6 Aug 2002, David Brownlee wrote:
>
> # 	   In 1.5.2 and earlier the default FFS blocksize is 8K, and
> # 	   fragments 16K.
>
> Really??  Fragment sizes are larger than block sizes??
>
> This should have some interesting results on my filesystems.
>
	Thanks for picking up the typo :) -  should have been 1K.

> # 	This is in addition to the obvious softdep and other improvements,
> # 	and in 1.6 UBC and other wins :)
>
> Heh.  I was happy to see Chuq still whittling away at UBC, noting that it
> still has a few small problems.  So often, I've seen an attitude (on other
> projects) with new subsystems or whatnot that has been "Well, there it is.
> It's not bug-ridden, that's the way it's meant to work.  Deal with it," even
> when its performance is horrid.  UBC will likely be a work in progress for
> quite some time...

	The biggest problem I've seen with UBC is that it lets the
	machine efficiently use a large quantity of memory for
	filesystem caching. This is only a problem because it shows
	up a deficiency in the disk sorting algorithm. You could
	see the problem in earlier NetBSD versions, but it was
	harder to trigger. I believe work is under way on this issue.

-- 
		David/absolute		-- www.netbsd.org: No hype required --