[Olsr-dev] OBAMP real testbed results
ZioPRoTo (Saverio Proto)
Mon Oct 12 10:38:06 CEST 2009
friday we had the first working "real" testbed of the OBAMP plugin. I
mean that instead of trying the obamp protocol with Netkit (UML
virtualization) we configured a few routers and laptops to test.
The plugin works, but there is a major problem with the maximum
bandwidth of the multimedia stream we are able to handle.
We have a host streaming to a multicast address a mp3 file, we tried
both UDP and RTP. The UDP stream travels correctly within the network
encapsulated in the OBAMP tunnels and it gets decapsulated to the
receivers. We experienced no packet loss.
However, the problem is that the audio is very chunky with a stream of
5Kb/s. We used VLC to stream and to receive the media file. Looks like
the packets arrive with a eccessive jitter to the receiver and not at
a constant rate.
We believe that the problem is bounded to the fact that the OBAMP
capture socket, that intercepts the multicast traffic, is handled by
the default OLSR scheduler. This choice was done to avoid the use of
libpthread. Probably the OLSR scheduler is not fast enough and
introduces this jitter in the multimedia stream.
We are going to investigate further in the following days.
Is it possible to control if a packet arrives on a socket without
registering the socket descriptor to the OLSR scheduler and without
using libpthread ??
If we don't find a solution we are thinking of introducing the use of
libpthread to see if handling the media stream with a different thread
solves the problem.
More information about the Olsr-dev