[asterisk-users] asterisk and multiple cpus/cores

[Matthew Rubenstein]

	Yes, that is perfectly clear - thanks for the data. The problem with
discussing load capacity of hosts running codecs is just how many codecs
are running at a time, how many code/decode instances each "call"
comprises, without knowing how many codecs are running per "call".

	I'm disappointed that Digium has not published exhaustive benchmarks on
capacity planning on different HW configs for different running setups.
An major unanswered question in planning Asterisk deployments,
especially scalable biz apps, is "how many hosts with what specs will I
need for X, Y, Z scales and usage combos?" This question is asked
several times in different ways on the maillists every month. And of
course the answers aren't in terms that can be collated into a
consistent planning guide. Since Asterisk is free, and minutes are
cheap, the HW, though relatively (to proprietary) cheap, is still a
major cost fraction. And of course running out of capacity by surprise
is a crippling blow.


On Sat, 2007-02-10 at 15:57 -0500, Andres wrote:
> Matthew Rubenstein wrote:
> 
> >	Are there 45 G.729 instances for the 45 ZAP legs in addition to 45
> >G.729 instances for the 45 SIP legs? Or do the ZAP legs not get a codec
> >(HW instead)?
> >
> >  
> >
> Its a 4 Port T1 with 92  ZAP channels.  So we are talking about 90 SIP 
> Channels being fed into 90 ZAP channels (which means 180 people are 
> talking, 90 on SIP Phones and 90 on PSTN lines).  We are therefore 
> transcoding 90 G.729 calls into 90 G.711 Calls.   It eats up 90 G.729 
> licenses.  I hope that clears things up.
> 
> Digium has also reported 80 G729 calls on their own dual cpu 2.8Ghz 
> Xeon  boxes: http://www.digium.com/en/products/voice/g729codec.php
> 
> >On Sat, 2007-02-10 at 12:06 -0500, Andres wrote:
> >  
> >
> >>Hi Matthew,
> >>
> >>Yes, those are really 90 SIP-ZAP calls.  Which means the 4 port T1 is 
> >>pretty much full of calls.  All SIP endpoints are forced to G729.  And 
> >>as for your 125% question I really don't know why.  This is just what I 
> >>can see from our MRTG graphs.  We graph all CPU usage and SIP/ZAP 
> >>calls.  All our servers are running Asterisk 1.2.9.1.
> >>
> >>Andres.
> >>
> >>
> >>Matthew Rubenstein wrote:
> >>
> >>    
> >>
> >>>	Are those "90 calls" really 90 instances of the G.729 codec (+ other
> >>>processing), 90 "legs" (people at phones) for 45 2-party calls?
> >>>
> >>>	Also, how do you get 125% more CPU bandwidth by adding another CPU,
> >>>which usually gets less than 100% more power after its overhead to
> >>>function in the system?
> >>>
> >>>
> >>>On Sat, 2007-02-10 at 04:46 -0700,
> >>>asterisk-users-request at lists.digium.com wrote:
> >>> 
> >>>
> >>>      
> >>>
> >>>>Date: Fri, 09 Feb 2007 21:57:23 -0500
> >>>>From: Andres <andres at telesip.net>
> >>>>Subject: Re: [asterisk-users] asterisk and multiple cpus/cores
> >>>>To: Asterisk Users Mailing List - Non-Commercial Discussion
> >>>>       <asterisk-users at lists.digium.com>
> >>>>Message-ID: <45CD3493.6000507 at telesip.net>
> >>>>Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> >>>>
> >>>>Erick Perez wrote:
> >>>>
> >>>>   
> >>>>
> >>>>        
> >>>>
> >>>>>I have found a site that list the following (no date in the post, so
> >>>>>it may be old):
> >>>>>"since all transcoding and calls still go through one core in
> >>>>>     
> >>>>>
> >>>>>          
> >>>>>
> >>>>asterisk,
> >>>>   
> >>>>
> >>>>        
> >>>>
> >>>>>it doesn't make sense to buy a multi-core or hyperthreaded system
> >>>>>     
> >>>>>
> >>>>>          
> >>>>>
> >>>>that
> >>>>   
> >>>>
> >>>>        
> >>>>
> >>>>>will only slow you down"
> >>>>>
> >>>>>Does that still applies in asterisk 1.2.14/1.4.x ?
> >>>>>Or do we have to tweak source code to balance loads
> >>>>>     
> >>>>>
> >>>>>          
> >>>>>
> >>>>(transcoding,etc)
> >>>>   
> >>>>
> >>>>        
> >>>>
> >>>>>between cores?
> >>>>>
> >>>>>     
> >>>>>
> >>>>>          
> >>>>>
> >>>>I can tell you that statement is bogus.  We run a number of dual cpu
> >>>>and 
> >>>>single cpu systems on our network.  The dual ones (Xeon 3.6Ghz) can 
> >>>>easily handle 90 G729 calls at 50% CPU Usage.  The single ones will
> >>>>be 
> >>>>at 50% with only 40 calls.
> >>>>
> >>>>Andres 
> >>>>   
> >>>>
> >>>>        
> >>>>
> 
-- 

(C) Matthew Rubenstein