estimating VoIP data traffic size from VoIP signaling traffic size ?
John Todd
jtodd at loligo.com
Sun Oct 23 18:26:30 UTC 2005
>Hi,
>
>is there any statistics on aggregated VoIP signaling
>bandwidth and aggregated VoIP data bandwidth? eg. if
>we monitored there is 2Mbps(average) traffic on VoIP
>signaling protocol ports ( including SIP, H.323,
>MGCP), how could we estimate average VoIP data
>bandwidth?
>
>Joe
As mentioned in prior responses to this thread: there are several
ways to guess, but mostly the answer is "No, not easily." The good
news is that excepting proprietary protocols like Skype and efficient
trunking protocols like IAX2, RTP is standardized. This means one
VoIP protocol is pretty similar to the other as far as RTP size goes,
so at least that part of the equation isn't open-ended. (I'll assume
you're looking for end-user statistics, and not inter-nodal
statistics where some type of aggregated IP header compression or
trunking might make flows more IP-header-friendly.)
Looking just at protocols that use RTP, it's still not quite possible
to map RTP volume simply from signalling volume without opening up
the signalling to see what codec is being used. If you have a mix of
codecs, then your bitstreams for the RTP can range from (typically)
~24kbps for G.729 up to ~80kbps for G.711 (1). Each call can be
different, depending on the ability of the originating and
terminating gateway/useragent to accept or prefer each codec during
the call set-up. You'd need a clear understanding of what codecs
your user community was utilizing in order to build an assumption
table on number of streams using each codec and/or protocol.
The media stats in SIP BYE signalling Bill Woodcock mentioned in his
message (jitter, packets, loss, latency, etc.) are only available in
a few end devices at the moment, notably Cisco. The RTCP XR
(RFC3611) standards might be visible in signalling soon via SIP
NOTIFY messages (2), but I don't know of any equipment that supports
this right now.
I think the best way to do this would be to graph the signalling
volumes and the media volumes over a week or two, and then build
assumption charts for future use. It may not be a big win if the
effort to measure signalling is the same as the effort to measure the
media, since you have to sample at a point(s) where all this data
crosses your measurement instrumentation. If you're really a
masochist, or you can't see the media for architecture reasons, you
could write an extension to tethereal or ettercap or a similar
network monitoring and packet analysis tool which unfolded each
signalling message, extracted the codec descriptors, and calculated
flows. You'd then have to keep state on each call, etc. etc. etc. -
not simple, but not impossible. Lastly, I'm betting there are some
signalling analysis tools on the market that already do this, but I
would expect that they will not be cheap.
If you're looking at traffic generated by Skype or other
closed-protocol system, you're really hanging out in the wind but I'm
sure that can be averaged and extrapolated if you have access to a
number of media streams from your user population to examine. (Does
Skype use extensively variable bitrates depending on endpoint
capabilities?)
JT
(1) http://www.erlang.com/calculator/lipb/ (note: RTP for SIP and
H323 is identical)
http://www.voxgratia.org/docs/codecbw.html
Take all media flow estimates with a grain of salt; typically
numbers are higher than reported, like G.711 being just shy of
90kbps instead of 80kbps as noted in most charts.
(2)
http://www.ietf.org/internet-drafts/draft-johnston-sipping-rtcp-summary-08.txt
More information about the NANOG
mailing list