[asterisk-dev] rtp scalability improvement...

[alex at pilosoft.com]

On Mon, 20 Mar 2006, Anton wrote:

> I would just exoress my oppinion regardin the call volumes and the call
> switching core. 400 calls seems for me to be not so much good number and
> even "comodity" hardware with proper approach should be able to process
> more calls even WITHOUT sacrifing kerner routing/packet processing
> stuff.  For example We're operating commecial VOIP H323 softswitch,
> which is currently run on the ordinary P4 with 1Mb RAM - this one could
> switch more than 500 simultaneous calls per second (MVTS) and with
> Dual-head XEON it claims to be able to do more than 1500 calls/s. Not
> arguably this one does not such features as Asterisk does, but from one
> side it is not supposed to have. From another, the mentione tests with
> Asterisk used only PASSTHROUGH switching functions of the asterisk - so
> 400 calls not too bad, but I do thing that Single Xeon should be able to
> process more than 1000 concurrent calls with SIP. H323 is more CPU
> intensive.  One more point voting for looking for extra scalability - is
> that Asterisk potentially could be used even as switching core while
> running not even PBX but normal PSTN - but scalability is the stone on
> the way. Seems someone objects Asterisk growth to a potential competitor
> to a proprietary solutions...
Well, the answer is, Asterisk is not a softswitch. Asterisk is a PBX. 
Asterisk's performance doing thousands of concurrent calls is known to be 
crappy, and a *lot* of work is necessary to make it suck less.

-alex