25 Jun
2009
25 Jun
'09
12:45 a.m.
I am running SER/rtpproxy on some systems with
multiple cores and multiple ethernet interfaces.
I am running multiple copies of rtpproxy, each with
an IP address associated with a particular interface,
as in one rtpproxy is using em1 which is 208.33.44.3
and em2 is 208.33.44.4 which another rtpproxy listens
at, while ser listens in em0 at 208.33.44.2. These
three (actually ten) interfaces pass through ethernet
switches and then on to a router with a default gateway
of 208.33.44.1 and then off to the Internet or wherever.
(In my setup, each rtpproxy has a second interface pointing
into a private network, so only one side of a call appears
on any given interface and five face toward the Internet.)
SER+rtpproxy correctly divides RTP traffic for calls to
208.33.44.3 and 208.33.44.4 in the SDP payloads, and
callers send their RTP packets to the two rtproxies
IP addresses without problems. That part works just fine.
The problem is that rtpproxy is using the default route
for transmission, so everything that goes towards the
Internet goes out via em0, so em1 and em2 don't transmit
anything. This means that this one link is overwhelmed
by the traffic that should have been divided across
multiple links.
Rtpproxy is allowing the local route table and the
default route to handle the outgoing packets and so
the packets are all going out the same interface.
In my case, em0 was 100Mbit/sec (completely sufficient
for SIP messages, while em1 and em2 were gigabit links
to handle RTP, but they are getting used only in one
direction. Even if em0 was upgraded to Gbit, it would
still not be able to handle the potential for 2+ Gbit
of data from the rtproxies.
So, is there any way to force rtproxy to perform what
I believe they used to call "reflector mode", where
the answers go out the same interface they come in on?
(This used to be common on UDP-based NFS file servers.)
A number of communication programs also allow this
behavior as an option, but rtpproxy doesn't appear
to do it or it isn't turned on by default.
Because the destinations for the RTP are out on
the Internet somewhere, I don't believe I can solve this
via "route" or "ifconfig" settings, but would be
delighted if someone knows different or knows of
a setting change in rtpproxy that will do this
so that load balancing will work.
The OS in question is FreeBSD 7.2, but I have the same
issue on 6.2 and 6.3.
Thanks in advance!