Re: [Freeswitch-users] Load test - performance not even matching Asterisk
Hi all An update on the performance measurements: The measurements I have referred to earlier all involved an Asterisk as the call generator. Somehow this setup leads to extensive rtp bandwidth usage. Each channel used around 500 kbps. If a phone is entered into the loop, this is reduced to the expected 64 kbps. I have not found any reason for this, but it certainly fouls up the test, and I have changed the test setup. Further, and since the earlier tests, the network has been updated to a Gbits network. I have now made two new test: 1) Using WinSIP from Touchstone as a call generator. 2) Using the Asterisk as one component, and setting up a chain of calls which goes forth and back from the Asterisk and the FS. All call are started from a real phone, and after 100 loops, where the calls are answered and sent on by the dial plan, the calls are terminated by an tone (<action application="gentones" data="%(500000,0,400)"/>) in the FS. The two test show similar top-figures at similar loads. The first test would be my preferable, but it is limited to 50 calls due to the trial licence limitations. Using an external non-FS and non-Asterisk device will eliminate some uncertainties, that's why it would be preferred. The other test has been done with 600, 400 and 200 channels (300, 200 and 100 calls), and the results of the top command are: cpu sy ni id wa hi si total * 600 10 30 0 33 0 2 25 100 FS600 22 33 0 30 0 0 15 100 0 * 400 7 18 0 67 0 1 7 100 FS400 14 17 0 62 0 0 7 100 0 * 200 3 10 0 84 1 0 2 100 FS200 7 8 0 82 1 0 2 100 The results do not show significant differences between the capacity behaviour of the Asterisk (*) and the FS. The also show an expected interrupt load (si) proportional to the square of the call load. Still the FS does not really outperform the Asterisk - which I find disappointing. Any comments are welcome.
Try using something like SIPP for load testing you can load test to much higher numbers http://www.freeswitch.org/eg/load_test.tgz is what we use for testing so you can duplicate the results... Also, look at the configuration you are doing and determine if you really need all the features that are there... Things like presence tracking, certain CDR loggers, and a few other things under high CPS loads can cause more problems then you think... Hint... Mount freeswitch/db as a ram drive in linux this is a big performance booster (since it takes the load of sqlite of the hdd), also turn off presence tracking on all sip profiles that don't need it... Something doesn't sound right on the 500k of rtp... Also remember that asterisk RTP stack doesn't handle async rtp.. It depends on receiving a packet to transmit a packet
From: Jon Bruel <jbr@consiglia.dk> Reply-To: <freeswitch-users@lists.freeswitch.org> Date: Sat, 4 Oct 2008 11:03:40 +0200 To: <freeswitch-users@lists.freeswitch.org> Subject: Re: [Freeswitch-users] Load test - performance not even matching Asterisk
Hi all An update on the performance measurements: The measurements I have referred to earlier all involved an Asterisk as the call generator. Somehow this setup leads to extensive rtp bandwidth usage. Each channel used around 500 kbps. If a phone is entered into the loop, this is reduced to the expected 64 kbps. I have not found any reason for this, but it certainly fouls up the test, and I have changed the test setup. Further, and since the earlier tests, the network has been updated to a Gbits network. I have now made two new test: 1) Using WinSIP from Touchstone as a call generator. 2) Using the Asterisk as one component, and setting up a chain of calls which goes forth and back from the Asterisk and the FS. All call are started from a real phone, and after 100 loops, where the calls are answered and sent on by the dial plan, the calls are terminated by an tone (<action application="gentones" data="%(500000,0,400)"/>) in the FS. The two test show similar top-figures at similar loads. The first test would be my preferable, but it is limited to 50 calls due to the trial licence limitations. Using an external non-FS and non-Asterisk device will eliminate some uncertainties, that's why it would be preferred. The other test has been done with 600, 400 and 200 channels (300, 200 and 100 calls), and the results of the top command are: cpu sy ni id wa hi si total * 600 10 30 0 33 0 2 25 100 FS600 22 33 0 30 0 0 15 100 0 * 400 7 18 0 67 0 1 7 100 FS400 14 17 0 62 0 0 7 100 0 * 200 3 10 0 84 1 0 2 100 FS200 7 8 0 82 1 0 2 100 The results do not show significant differences between the capacity behaviour of the Asterisk (*) and the FS. The also show an expected interrupt load (si) proportional to the square of the call load. Still the FS does not really outperform the Asterisk - which I find disappointing. Any comments are welcome.
_______________________________________________ Freeswitch-users mailing list Freeswitch-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
Like I told the other guy we are pretty much done supporting people doing load testing. I know for a fact that on my dual core wood crest that is now 4 years old that I can bring up 3000 channels playing a file out RTP at 20ms ptime, 5000 if i switch to 60 and 10000 if i use 120ms ptime. Ken can also attest to these numbers as well as provide you with his numbers from his own testing I'm sure. This is a fact. If you cannot reproduce it then it's not my problem. I know for a fact that you are clearly trying to develop a commercial application and I have subtly pointed out that you need to seek commercial support. Now I am making it crystal clear, Stop asking for help about load testing unless you are going to pay for our time to teach you how to set it up properly. On Sat, Oct 4, 2008 at 4:11 AM, Ken Rice <krice@suspicious.org> wrote:
Try using something like SIPP for load testing you can load test to much higher numbers http://www.freeswitch.org/eg/load_test.tgz is what we use for testing so you can duplicate the results...
Also, look at the configuration you are doing and determine if you really need all the features that are there... Things like presence tracking, certain CDR loggers, and a few other things under high CPS loads can cause more problems then you think...
Hint... Mount freeswitch/db as a ram drive in linux this is a big performance booster (since it takes the load of sqlite of the hdd), also turn off presence tracking on all sip profiles that don't need it...
Something doesn't sound right on the 500k of rtp... Also remember that asterisk RTP stack doesn't handle async rtp.. It depends on receiving a packet to transmit a packet
From: Jon Bruel <jbr@consiglia.dk> Reply-To: <freeswitch-users@lists.freeswitch.org> Date: Sat, 4 Oct 2008 11:03:40 +0200 To: <freeswitch-users@lists.freeswitch.org> Subject: Re: [Freeswitch-users] Load test - performance not even matching Asterisk
Hi all An update on the performance measurements: The measurements I have referred to earlier all involved an Asterisk as the call generator. Somehow this setup leads to extensive rtp bandwidth usage. Each channel used around 500 kbps. If a phone is entered into the loop, this is reduced to the expected 64 kbps. I have not found any reason for this, but it certainly fouls up the test, and I have changed the test setup. Further, and since the earlier tests, the network has been updated to a Gbits network. I have now made two new test: 1) Using WinSIP from Touchstone as a call generator. 2) Using the Asterisk as one component, and setting up a chain of calls which goes forth and back from the Asterisk and the FS. All call are started from a real phone, and after 100 loops, where the calls are answered and sent on by the dial plan, the calls are terminated by an tone (<action application="gentones" data="%(500000,0,400)"/>) in the FS. The two test show similar top-figures at similar loads. The first test would be my preferable, but it is limited to 50 calls due to the trial licence limitations. Using an external non-FS and non-Asterisk device will eliminate some uncertainties, that's why it would be preferred. The other test has been done with 600, 400 and 200 channels (300, 200 and 100 calls), and the results of the top command are: cpu sy ni id wa hi si total * 600 10 30 0 33 0 2 25 100 FS600 22 33 0 30 0 0 15 100 0 * 400 7 18 0 67 0 1 7 100 FS400 14 17 0 62 0 0 7 100 0 * 200 3 10 0 84 1 0 2 100 FS200 7 8 0 82 1 0 2 100 The results do not show significant differences between the capacity behaviour of the Asterisk (*) and the FS. The also show an expected interrupt load (si) proportional to the square of the call load. Still the FS does not really outperform the Asterisk - which I find disappointing. Any comments are welcome.
_______________________________________________ Freeswitch-users mailing list Freeswitch-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_______________________________________________ Freeswitch-users mailing list Freeswitch-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
-- Anthony Minessale II FreeSWITCH http://www.freeswitch.org/ ClueCon http://www.cluecon.com/ AIM: anthm MSN:anthony_minessale@hotmail.com <MSN%3Aanthony_minessale@hotmail.com> GTALK/JABBER/PAYPAL:anthony.minessale@gmail.com<PAYPAL%3Aanthony.minessale@gmail.com> IRC: irc.freenode.net #freeswitch FreeSWITCH Developer Conference sip:888@conference.freeswitch.org <sip%3A888@conference.freeswitch.org> iax:guest@conference.freeswitch.org/888 googletalk:conf+888@conference.freeswitch.org<googletalk%3Aconf%2B888@conference.freeswitch.org> pstn:213-799-1400
participants (3)
-
Anthony Minessale
-
Jon Bruel
-
Ken Rice