[asterisk-users] where the bottleneck lies ? (was: Serverredundancy)

[Simone Cittadini]

Douglas Garstang ha scritto:

>>-----Original Message-----
>>From: Simone Cittadini [mailto:mymailforlists at gmail.com]
>>Sent: Wednesday, July 12, 2006 10:02 AM
>>To: Asterisk Users Mailing List - Non-Commercial Discussion
>>Subject: [asterisk-users] where the bottleneck lies ? (was:
>>Serverredundancy)
>>
>>
>>unplug ha scritto:
>>
>>    
>>
>>>I feel interested about you can support 16,000 users of your system.
>>>As I have tested using sipp in a dual CPU Xeon with 2G Ram, the
>>>maximum number of current call is about 160.  In some 
>>>      
>>>
>>forums, most of
>>    
>>
>>>ppl claim the maximum current call is about 100-200.  What do you
>>>expect the number of current call to handle in 16,000 users?
>>>
>>>      
>>>
>>I'm curious about what was limiting the number of calls in your tests.
>>
>>For every system I have in production/testing I see the only 
>>bottleneck
>>is system load, cpu and memory usage is well beyond limits 
>>when "things
>>starts to fall apart". The unexplicable (at least by me) thing is that
>>system load seems to be only partially influenced by the number of
>>calls, for example sometimes there are 100/150 calls and the load is
>>around 0.70, sometimes it skyrockets to  2.00 / 2.50 (when it is > 2
>>calls quality is crippled, I think because of too many 
>>dropped packets).
>>I see this behaviour no matter how simple/complex the system is, from
>>just a terminator with a couple of digium in it and a five-lines
>>extension to the central server with fastagi doing mysql queries and
>>taking hundreds of concurrent calls in both sip and iax.
>>Can it be something related to asterisk itself ? I'm thinking about
>>installing oprofile on the various servers, someone by chance already
>>did it ?
>>    
>>
>
>Another consideration is if the phones have performed reinvites, and removed Asterisk from the RTP stream. If you can live without call recording, and other features where Asterisk has to remain in the RTP path, then I imagine that this would significanlty reduce load on the Asterisk systems. Could some of your phones be reinviting? This may explain the variation in load.
>
>Doug.
>
>  
>
no, all the traffic has to pass from the machine (and all the codec is 
g711 so no differences in transcoding either)