Hi,
You should try some more tests with the newest code
there have been many recent performance changes in the last 2 weeks.
----- Original Message ----
From: Muhammad Shahzad <
shaheryarkh@googlemail.com>
To: freeswitch-users@lists.freeswitch.org
Sent: Friday, September 22, 2006 10:20:06 AM
Subject: Re: [Freeswitch-users] How many concurrent calls
Hi Everyone,
Sorry i was out-station and extremely busy for last couple of weeks or so, therefore, i couldn't reply earlier. Anyways, here is some data regarding our testing of FreeSWITCH conducted at a NOC in Lahore, Pakistan,
Setup:
1. We setup two servers of same hardware i.e. Fujitsu Siemens Xeon 3.2 GHz, each running Gentoo Linux 2006.0 with kernels 2.6.15-r8 and
2.6.17-r5 respectively.
2. These servers were then connected over cross-line cable to each other using Gigabit NICs.
3. We installed Asterisk version 1.2.10 on machine 'A' and FreeSWITCH
snapshot dated 2006-09-01 on machine 'B'.
4. Each machine had another Gigabit NIC through which it was connect to Internet, so that i can access them remotely.
5. I was connected to these servers over 100kbs CDMA wireless Internet link via ssh.
6. Date was Sept. 02, 2006, 0300hours PKST (GMT+5 zone).
Test Plan I: (Simple Playback)
We generated batches of 100, 200, 300, 400 and 500 calls, with steps 25, 50, 100 calls dialed simultaneously at an interval of 10 seconds. Each call plays a flat tone of one minute length, file format WAV, codec GSM.
Maximum concurrent calls should be,
X | 25 50 100 (steps)
------------------------------------------
100 | 100 100 100
200 |
150 200 200
300 | 150 250 300
400 | 150 250 400
500 | 150 250 500
(Test Matrix: Fixed interval of 10 seconds)
In this plan number of calls equal number of active IAX2 channels. Maximum time for maxium concurrent calls was between 9 and 10 seconds.
Results:
All batches successfully executed, however at 400 or above concurrent calls the load averge shoot to over 5.6 and FreeSWITCH console was showing call details so fast that my remote ssh link hanged. When the link resumed the FreeSWITCH was still up and number of calls were decreasing as the test was
moving towards end.
Test Plan II: (Forward Bridged Calls)
1. We generated same batches of calls except that now they dials out to Machine A (the one running Asterisk). That machine accepts the call and plays back a milliwatt for 1 minute, codec GSM.
2. Since calls were bridged so number of calls were half the number of channels.
Results:
At 436 active channels (218 concurrent calls) Asterisk crashes and therefore all calls dropped.
Test Plan III: (Reverse Bridged Calls)
1. Same as plan II, except now we generated calls from Machine A (i.e. Asterisk) and Machine B (i.e. FreeSWITCH) answers the call and plays back flat tone of 1 minute length (file format WAV, codec GSM)
2. Since calls were bridged so number of calls were half the number of channels.
Results:
At
394 active channels (197 concurrent calls) FreeSWITCH crashes and therefore all calls dropped.
Warning:
These tests may not exhibit same results in all environments, we did not follow any specific testing standard nor we endorse any software or hardware used in these tests.
I wanted to decrease the interval as well as call steps to get more exact information on how and when FreeSWITCH and asterisk crashes but i had to abandon further plans due to bad weather (i was using wireless link which went down).
Now i am planing for FreeSWITCH to FreeSWITCH bridged calls testing, i will send you the results whenever i got those servers available to me.
Thanks and best regards.
-shahzad
--
Muhammad
Shahzad
-------------------------------
Manager Support,
Convergence Business Systems,
Lahore, Pakistan.
Cell: 0092 334 422 40 88
Email:
shaheryarkh@googlemail.com