Re: Re: libdvdread

Thu Jul 25 2002, Niels Provos wrote:
> On Thu, Jul 25, 2002 at 02:54:49PM -0500, Billy Biggs wrote:
> >   Also implementing POSIX aio methods would be useful too.  I keep
> > meaning to try this out to see.
> In the case of OpenBSD, as we are lacking UBC, the raw IO to the dvd
> drive is not being cached.  If I access the dvd via isofs I get a 
> throughput of about 6MB/s.  However, reading from the raw device
> maxes out at 900KB/s.
> 
> That's why a caching layer in libdvdread would be useful for me.
> I do not know if many other people are having similar problems.
> 
Wouldn't fix OpenBSD have an enven larger impact?  Lots of programs
would benefit from that.

Will there be any benefit from not using the raw device?
I.e split up the fd's that are used for the 'ioctl' calls and the
one used to read the actual data?

What is it that make the isofs access so much faster?

Ogle, at least, has a somewhat strange access patter in that the video
data is read like (in sectors) 1, 120, 1, 70, 1, 210, 1... and so on.
I.e. every other access in a single sector and the other are for one
whole VOBU (normaly 0.5 seconds of compressed video).

-- 
Håkan Hjort