[linux-dvb] Re: Any need for gszap?

[Holger Waechtler <holger@convergence.de>]

Roberto Ragusa wrote:
> On Mon, 05 Jan 2004 09:53:51 +0100
> Holger Waechtler <holger@convergence.de> wrote:
>
>
> > Niklas Peinecke wrote:
> >
> > > I've seen the EOF problem also. I got the impression that it is speed 
> > > related, i.e. if there is heavy load on the machine EOF occurs more 
> > > frequently. Nevertheless if you pipe the stream to a file and then 
> > > replay, there are no problems. Maybe the driver replays values <0 
> > > through /dev/dvb/adapter0/dvr0 (don't know if this is possible)?
> > no, it's not. You get the raw unmodified data from air through the dvr 
> > device. Data is only lost in case of buffer overflows, but the MPEG 
> > decoder should be able to tolerate corrupted MPEG packets silently.
>
> Are you saying that a buffer overflow would only cause skipped packets?
yes, sure.

> I'm quite sure I used to get an error number 75 (EOVERFLOW) when the
> buffer overflows.
> Just grep for EOVERFLOW in dmxdev.c.
When a buffer overflows the ringbuffer pointers are reset, errno is set 
to EOVERFLOW. The data that has been previously in the ringbuffer will 
get lost.


> In the same file, function dvb_dvr_do_ioctl() there is a disappointing
>
>         switch (cmd) {
>         case DMX_SET_BUFFER_SIZE:
>                 // FIXME: implement
>                 ret=0;
>                 break;
>
> so the limit can't be even raised. The default is 10*1024*188 bytes
> (which is only one second with HDTV at 15Mbit/s).
You don't want to write a program that reads and processes the stream 
only every second, right? This would cause pretty heavy latencies...

Anyway, a patch implementing this is highly welcome as well as one which 
ports all ringbuffer structs to the general dvb_ringbuffer implementation.
:)

Holger



--
Info:
To unsubscribe send a mail to ecartis@linuxtv.org with "unsubscribe linux-dvb" as subject.