17 Mar
2010
17 Mar
'10
8:04 p.m.
Daniel-
On 03/16/2010 10:33 PM, Vikram Ragukumar wrote:
Yes SEMS looks good. We've been talking with Stefan Sayer.
Hello,
We're making initial modifications to rtpproxy to support high channel
capacity transcoding and encryption.
At this point we want to get some general idea of the scope of changes
needed for rtpproxy and Kamailio... so we're starting with small steps.
We've been studying rtpproxy source and our current thinking is to add
a sub-structure to the existing rtpp_session structure (defined in
rtpp_session.h). The new sub-structure would include:
-encryption options (type, key length,
salt size, type of key mgt protocol, etc)
-encode / decode options (type, VAD/CNG,
VIF size, etc)
Any comments or advice on this approach appreciated.
Not being a rtpproxy developer at all, I do not see a problem with the
approach. Also note that rtpproxy is a single process application (or
used to be in case last version changed), take that in consideration
when designing.
Not sure whether to start a separate thread, but
also there is the
issue of what changes are necessary to Kamailio to support sending
updated commands to rtpproxy. Would modifying Nathelper alone be
sufficient?
Just updating the nathelper is sufficient in kamailio.
Another idea I was playing with in the past, but time was limited, was
to enhance sems (sip express media server) to support communication via
nathelper-rtpproxy protocol. sems is lightweight sip media server,
supporting already transcoding. Not being a rtp/audio guy, my plan was
to use an existing audio mixer. Right now i use routing via a media server when
needing transconding, so
call flow is:
[caller] ---- [kamailio] ------ [media server: asterisk/freeswitch/sems]
----- [kamailio] ------ [caller]
With the above scenario, the issue for us is channel capacity -- Asterisk can't do a
lot of transcoding, and even if the TC400B card is used, it turns into a "PCIe bus
slugfest" as all speech/packet data has to go back and forth. Asterisk + Linux
kernel still have to "touch" every RTP packet. Also the TC400B can't do
encryption,
conferencing, etc. Other hardware can do 100s or even 1000s of channels, so it seems
to make sense to enhance rtpproxy, at least at this point.
For an Asterisk-centric approach, one way may be to enhance native bridging
(canreinvite=yes). We're looking at ways to spoof SDP negotiation so both ends think
they have the same media (codec) capabilities, even if not actually the case. Then
exchange RTP through a card with its own GbE that does the transcoding (or other
required functionality). The advantage in this case would be "keep it simple":
add
a card to Asterisk, some external software, and capacity is enhanced. Advantages
such as "included with the chip" Texas Inst licensing might also be useful.
-Jeff