[asterisk-dev] Asterisk scalability (was: Improve scheduler performance under high load)
Chris Tooley
chris at tooley.com
Tue Feb 17 12:44:04 CST 2009
> Seems like I will restart these tests for a customer in March to get
> more data.
> My conclusion so far is the same as Stephen Uhler's at the Atlanta
> Astridevcon - the IP stack is our problem here.
>
I agree with this to a certain extent, but as I've pointed out
multiple times, I can get far more calls per server by splitting the
load between multiple asterisk installations on the same server. ie:
If I set up 4 installations and tell them to listen to 5060, 5061,
5062, and 5063 for SIP traffic, I get a growth curve of maximum
channels per server. The optimal maximum number of installations per
box for our particular hardware is actually about 8 installations, but
the management headache means that the 20% growth that we get compared
to running 4 installations isn't worth it. However, 4 installations
is right at 100% more effective than 1 installation.
In our test environment I can do 1000 concurrent channels before call
quality goes to down the drain on a single installation. By using
multiple installations I've been able to push that limit past 2000
channels. However, for "safety's sake" I usually leave them
configured to handle 400 channels per installation. This largely due
to some modifications I've made to call-limit to make it only affect
inbound calls, and not block outbound calls (ie: I could easily have
450 channels on an asterisk instance that is configured with a 400
channel call-limit).
My point is, based on what I've seen, I'd say there's definitely some
work that can be done to affect performance of Asterisk internally and
it's not just limited by the IP stack.
Chris Tooley
More information about the asterisk-dev
mailing list