[Asterisk-Users] RE: 5,000 concurrent calls system rollout question

[Greg Boehnlein]

On Thu, 2 Feb 2006, John Todd wrote:

[SNIP]

> 3) Nobody else has thus far taken the bait and made any comments 
> about their systems. I appreciate Signate's comments; they seem to be 
> the only ones to publicly claim large-scale throughput using Asterisk 
> in a public forum.  Most other people who claim thousands or even 
> high hundreds of connections do so offhand, without responding to 
> second questions when I raise my figurative eyebrows.

John,
	Per our conversation in San Fransciso, I am starting to push a 
couple of my Asterisk boxes farther than I've gone before. I'm not yet 
anywhere near the 5,000 concurrent call level on my boxes, but I am 
starting to see 150-160 concurrent calls coming through the system. In 
this case, these are SIP to SIP where Asterisk is staying in the media 
stream, but rarely transcoding. Approximately 99% of the calls coming 
through are just pass-through g729, with the occasional gsm conversion. 
I'm running Asterisk 1.2.4-svn in a completely stock configuration. I.E. 
no patches whatsoever, and absolutely performance tweaks. In fact, the 
system is running using MALLOC_DEBUG to catch memory leaks and is built 
using "dont-optimize" so we can get backtraces if things go south.

	My Dial-Plan is highly optimized, with a focus on being as 
efficient as possible while offering failover options for call completion.

> 4) There are still no notes on other problems with scale here.  I've 
> had systems with several hundred simultaneous SIP connections, but 
> "sip show channels" sure does start to take a while.  What _other_ 
> problems crop up, but don't necessarily cause a "failure" condition?

Well, debugging anything on the console with 160 concurrent calls coming 
through the system (sometimes 4-5 calls / second) is nearly impossible. 
Most of the time, I don't even run the console, and simply execute 
commands from a bash prompt as "asterisk -rx 'sip show channels'". I 
ALWAYS, ALWAYS, ALWAYS issue a "set verbose 0" before I reload the box, as 
a reload causes the box to hiccup slightly while it is printing the data 
to the console.

I had originally opted to write CDRs to disk and then import them into a 
SQL database, but after I cleaned up my dial-plan, I opted to use 
cdr_odbc. I am concerned that this could cause a blocking condition if the 
SQL server is unavailable, but for now I'm taking the risk because I need 
to have real-time stats on call statistics.

> 5) I will agree that most SIP testing systems are currently too 
> pricey.  I would love to find a well-connected network that rents out 
> a few of the better-known SIP testing tools to beat on Asterisk 
> installations in remote places for short periods of time.   But this 
> has always been the case... test gear is a small market, and 
> expensive.  Just look at the MSRP of new high-end HP Oscilloscopes if 
> you want to get a picture of price-gouging.

I know that Olle spent some time at SipIt w/ Asterisk, and he's been 
interested in doing some additional compliance and scalability testing. 
I'd like nothing better than to get a couple of key developers together 
for a weekend of scalability bashing somewhere, preferably outside of the 
regular conference circuit (too distracting) to push things to their 
limits.

-- 
    Vice President of N2Net, a New Age Consulting Service, Inc. Company
         http://www.n2net.net Where everything clicks into place!
                             KP-216-121-ST