Subject: Re: Interactive responsiveness under heavy I/O load
To: Hauke Fath <hauke@Espresso.Rhein-Neckar.DE>
From: David Brownlee <abs@netbsd.org>
List: tech-kern
Date: 01/27/2004 19:37:28
On Tue, 27 Jan 2004, Hauke Fath wrote:

> At 23:43 Uhr +0100 26.1.2004, Johnny Billquist wrote:
> >>    Softdep is such a huge performance win, especially for tasks like
> >>    extracting a pkgsrc tree, because lots of metadata writes update
> >>    the same disk block repetitively (think about adding files one at a
> >>    time to the same directory). With softdep, each of these updates is
> >>    done in memory (without the sync disk wait) and the resulting final
> >>    directory blocks written to disk (again, speaking very broadly).
> >
> >softdep:s really sounds nice, but last I tried it (a few months ago) it
> >still crashed my VAX very predictably, so I'd guess it still isn't good
> >enough to turn on for people in general.
>
> Same here on a SparcStation 10. Both with current and 1.6 kernels, softdep
> mounts make sure the box locks up during the next /etc/daily run.

	I'm running softdeps on a 1.6.1_STABLE, alpha and sparc5 as
	reasonably loaded NFS servers, and on 1.6ZI sparc & sparc64 boxes
	as NFS clients (twenty or so users).

	Also on i386 boxes, but those are less interesting :)

	Alpha has been up for 165 days (I forget why it was rebooted),
	sparc5 tends to go down every few months. Client boxes running
	current are a little harder to benchmark as they keep getting
	updated :)

	I'm pretty sure there are still softdep problems, or problems
	elsewhere exposed by softdeps pushing things harder...

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