[Angstrom-devel] [RFC] Using IOSCHED_NOOP as the default for linux-handhelds

[pHilipp Zabel]

On Feb 11, 2008 4:55 PM, Richard Purdie <rpurdie at rpsys.net> wrote:
> On Mon, 2008-02-11 at 12:39 +0100, pHilipp Zabel wrote:
> > On Feb 10, 2008 10:07 PM, Paul Sokolovsky <pmiscml at gmail.com> wrote:
> > > IMHO, noop's description is good enough for builtin flash and flash
> > > cards. I have no idea why IOSCHED_AS is used (pristine default is
> > > IOSCHED_CFQ), but it is the same as used by linux-rp, so I assume
> > > that's where it come from. So, I cc: Richard Purdie to ask if there're
> > > good grounds for that and to draw attention that it's a chance to
> > > recover some 5K for kernelspace-starving Z devices.
> >
> > That sounds sane enough to me.
> > Has anybody ever done benchmarks with the different schedulers on
> > MTD/MMC/CF flash storage? Without knowing much about how AS or CFQ
> > work, I'd even expect the noop scheduler to be the fastest here.
> > Although, isn't there one zaurus that can boot directly from the
> > internal microdrive?
>
> For flash based devices I'd agree that NOOP makes sense as long as it
> does some kind of block request merging which would be important to
> reduce partial read/writes.

Yes, it does that.

> Spitz is the zaurus with the microdrive and one of the other schedulers
> makes more sense for it since it really does have disk heads and will
> benefit from a 'proper' scheduler.

If it boots from microdrive it might speed up boot time to have as or
cfq as default. But maybe having the 'proper' scheduler as module and
enabling it early during boot would be good enough.

regards
Philipp