[Live-devel] NIC overruns on IBM stbx25xx embedded platform

[Richard Flude]

> I would look at modifying the calculation of "fNextSendTime" (and/or 
> "uSecondsToGo") in "liveMedia/MultiFramedRTPSink.cpp".

I modified uSecondsToGo to be a small positive when negative with some 
success. It significantly reduced the number of overruns (sometimes 
close to zero).

Whilst I was doing this I wrote a simple multicast receiver to analyse 
the changes to the RTP streams produced by 
testMPEG1or2AudioVideoStreamer. I noticed the  application produces 
many small udp packets for the video stream. This seems a little 
strange to me given the large amount of data to be sent for a video 
stream.

30% of the video RTP packets are smaller than 500 bytes, and 19% 
smaller than 100.

Is this typical or expected? Could I see an improvement in MVP 
throughput if these packets were consolidated? Is this consolidation 
practical?

I'm starting to enjoy this, which is a worry.

Richard

PS Here is the output for just over 100,000 packets received 
(1m14.870s) from the streaming of the coyote ugly song from that DVD 
(it's what's on-hand alright;-).

audio - packet cnt: 518, total bytes: 605024, min packet size: 1168 max 
packet size: 1168, cnt < 500: 0, cnt < 100: 0, avg packet size: 1168

video - packet cnt: 99832, total bytes: 85335576, min packet size: 17 
max packet size: 1448, cnt < 500: 31435, cnt < 100: 19095, avg packet 
size: 854