[linux-dvb] Re: we might be losing multicast packets

[Jeremy Hall <jhall@maoz.com>]

Hi Johannes,

In the new year, Johannes Stezenbach wrote:
> On Sun, Feb 06, 2005 at 08:51:45PM -0500, Jeremy Hall wrote:
> > I am using current dvb-kernel v2.4 branch with a dvb/s ff card.  I am 
> > using dvbnet that I found in some apps dir on my hd, no telling how old it 
> > is, but I run dvbnet -p pidnum to get a dvb0_0 and another one for dvb0_1.
> > 
> > The datarate for dvb0_1 is small, like 64k, so I don't see losing packets, 
> > but dvb0_0 is like 7 or 8 megabits per second.
> 
> What do you mean by "datarate for dvb0_0"? The bandwidth on that PID
> or just the multicast you want to receive?

It is hard to tell what the bandwidth is for each multicast group, since 
there are 3 groups transmitting, eventually I'd like to pull all three at 
once.  The srate is 7955, so I'm guessing maybe the whole stream is about 
10 or 11 meg a second.  This would mean each stream, if evenly divided, 
might be around 4mb each, although if one isn't transmitting, the others 
might spin up their datarate to use available bandwidth.  How would you 
recommend I determine this accurately?

> The DEBI bandwidth on FF cards is limited (don't have exact figures).
> If you just want a part of the packets on that PID you could try
> hw_sections=1 to reduce the used DEBI bandwidth (but there are reports
> that hw section filters stop on overflow conditions).
> 
My experience with hw_sections=1 has been (in the past) that the arm 
crashes because I'm trying to process too many sections at once or am not 
able to keep up or some other nonsense.

> Or the bottleneck is elsewhere: Does ifconfig report dropped packets?

no, it doesn't show dropping packets.

> Did you try to set SO_RCVBUF (see socket(7))?
> 
I just ran with SO_RCVBUF set to 2mb, 4mb, and now am running with 8mb.  I 
have had improving results with each size increase.  That still seems 
quite high to set the receive buffer.

_J

> Johannes
> 
>