Hi Sebastien,

it might really depend on the scalability versus the accuracy of
the solution you need to develop. I see you having pretty much two
ways to go:

a) Basic behavioural analysis. Assume all packets flying between
the known range of RTP ports are RTP packets. If you know in advance
the expected packet size, double check this assumption by dividing
the number bytes by the number of packets. You might proceed without
restricting such check to only the RTP ports, but might be tricky
depending on the scenario. A dumb variation of this approach is
possible if you know in advance the RTP payload size (and accuracy
is not a prime issue): using the Pre-Tagging infrastructure to tag
all, say, UDP packets matching a particular payload size.

b) Packet classification. The SIP pattern from the L7-filter project
recognizes SIP traffic; pmacct features a connection tracking module
for SIP which allows to expect upcoming RTP flows by looking into 
INVITE and 200 messages. This might turn into a far more accurate
solution (again, depending on the scenario) but with potential
drawbacks due to the required deep-packet inspection. 

On a side node, remember a stateful approach like the one relying
on packet classification is more sensible to outages compared to
behavioural analysis. Of course, nothing prevents any combination
of the two solutions. 

Here i'm assuming the easiest way to accomplish this, ie. every
host in the network is forced to speak RTP only passing through
a proxy and/or an AS does not hold.

Cheers,
Paolo

On Wed, Sep 03, 2008 at 02:26:29PM +0200, S?bastien CRAMATTE wrote:
> Hello
> 
> What is the best way to account SIP/RTP traffic with PMACCT ?
> Might we should use  L7-filter ? I mean essentialy for RTP traffic not 
> for SIP signalling
> 
> Regards

_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

Reply via email to