[prev in list] [next in list] [prev in thread] [next in thread] 

List:       asterisk-dev
Subject:    [Asterisk-Dev] Astricon - call for help
From:       joachim <zoachien () securax ! org>
Date:       2004-08-24 18:33:09
Message-ID: 6.1.1.1.2.20040824203255.03601cb0 () pop3s ! securax ! be
[Download RAW message or body]

Hi all,

I'm trying to make a paper/presentation for astricon with a lot of graphs 
and performance statistics for asterisk.

I'll try to handle:

- differences between the 2.6 kernel and the 2.4 series
- differences in hardware, ranging from slow embedded pc's to SMP setups.
- difference between opteron and xeon
- difference between icc and gcc (if i find a way to compile it). + maybe 
different optimization flags.
- differences between codecs.
- differences between iax/sip/h323

But my major problem is, how should i do the benchmarks ?  I need more 
suggestions.

My tests showed that the server load cannot be used to determine the 
quality of the calls. (i could make perfect sounding calls with a load of 
96 - no its not a typo!)

In general, i think that overall cpu usage should not be over 50% otherwise 
small clicks start to appear.

So far i think of trying to measure the maximum amount of calls (without 
audio) i can make to a server untill it refuses to accept new ones, by 
using a wait in the extensions.conf.
One way codec decoding i'd test by sending no audio to the server, but 
connecting using alaw and decode a prerecorded (gsm etc) file on the server.
Codec encoding i could maybe do by sending alaw to the server 
and  monitoring to /dev/null.

What else can you guys think of ?

I'd like to split up calls into smaller chunks. (decoding/ encoding / sip / 
zap / applications etc) to make it easier for people to say, hey i'll be 
using sip(x) to zap(y) with g729(z), what server do i need.... ooh its x + 
y + z

Feed me... :)

Zoa. 


[prev in list] [next in list] [prev in thread] [next in thread]