Re: [rfc] aio-core for 2.5.29 (Re: async-io API registration for 2.5.29)

[Pavel Machek]

Hi!

> > For a lot of applications like multimedia you actually want a counting
> > of time not any relation to real time except that you can tell how many
> > ticks elapse a second.
> 
> Absolutely. I think "jiffies64" is fine (as long as is it converted to
> some "standard" time-measure like microseconds or nanoseconds so that
> people don't have to care about internal kernel state) per se.
> 
> The only thing that I think makes it less than wonderful is really the
> fact that we cannot give an accurate measure for it. We can _say_ that
> what we count in microseconds, but it might turn out that instead of the
> perfect 1000000 ticks a second ther would really be 983671 ticks.
> 
> A 2% error may not be a big problem for most people, of course. But it
> might be a huge problem for others. Those people would have to do their
> own re-calibration..

I don't think so.

Imagine DVD playback. If you have 2% error, your audio is going to get
1 second off each minute. It is going to be off by one minute at the
end of hour. 2% is probably not acceptable.

[I'm not sure how exactly video/audio synchronization works, besides
fact it does not; but 2% could be huge problem for something like
that.]
								Pavel
-- 
Casualities in World Trade Center: ~3k dead inside the building,
cryptography in U.S.A. and free speech in Czech Republic.