Hi, I have been experimenting with ConfBridge from the asterisk-11 stable SVN branch (and with 11.2.0 also) for the last 3 weeks and I see a problem, which what I believe is performance related. I just wanted to ask if someone else has made any tests and what is the maximum number of participants that they've seen in a conference. I was never able to get more than 8 participants (mixed G722 and G711a) on a conference (actually that's per server limit) with almost all settings on default, except for dsp_drop_silence and denoise which are enabled. I tested on Debian squeeze, 64-bit, quad-core Xeon server @2.4GHz and also on another virtual server with similar processor (just one core available to the VM). While this is not the latest and greatest CPU, I would certainly expect it to handle more than 8 calls. To be honest, I was in fact able to get it working for up to 20 participants (most with G711), when I switched from res_timing_timerfd to res_timing_dahdi and turned off denoise, but that's still not normal I believe, especially with most participants on mute and with dps_drop_silence enabled and nothing else running on the server. The problem itself is, that once I get over the "critical" number of participants, the voice starts to break up and it's impossible to understand the person who's talking. This is certainly not bandwidth related because all tests were made on the LAN and besides I could see that the CPU was sometime close to 100%. Did someone observe something similar? BTW, once the first participant enters the conference I start seeing probably over 50 messages per second saying: bridging.c:757 bridge_channel_join_multithreaded: Going into a multithreaded waitfor for bridge channel 0x292d708 of bridge 0x28f3658 Best, Hristo -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130206/72244270/attachment.htm>
Thorsten Göllner
2013-Feb-06 12:54 UTC
[asterisk-users] ConfBridge performance problem...?
Did you watch the cpu usage (for example with top)? You have a board installed which does use dahdi? Did you check the command "dahdi_test"? Maybe a (performance) problem of the software ec? Am 06.02.2013 11:13, schrieb Hristo Trendev:> Hi, > > I have been experimenting with ConfBridge from the asterisk-11 stable > SVN branch (and with 11.2.0 also) for the last 3 weeks and I see a > problem, which what I believe is performance related. I just wanted to > ask if someone else has made any tests and what is the maximum number > of participants that they've seen in a conference. > > I was never able to get more than 8 participants (mixed G722 and > G711a) on a conference (actually that's per server limit) with almost > all settings on default, except for dsp_drop_silence and denoise which > are enabled. > > I tested on Debian squeeze, 64-bit, quad-core Xeon server @2.4GHz and > also on another virtual server with similar processor (just one core > available to the VM). While this is not the latest and greatest CPU, I > would certainly expect it to handle more than 8 calls. > > To be honest, I was in fact able to get it working for up to 20 > participants (most with G711), when I switched from res_timing_timerfd > to res_timing_dahdi and turned off denoise, but that's still not > normal I believe, especially with most participants on mute and with > dps_drop_silence enabled and nothing else running on the server. > > The problem itself is, that once I get over the "critical" number of > participants, the voice starts to break up and it's impossible to > understand the person who's talking. This is certainly not bandwidth > related because all tests were made on the LAN and besides I could see > that the CPU was sometime close to 100%. > > Did someone observe something similar? > > BTW, once the first participant enters the conference I start seeing > probably over 50 messages per second saying: > > bridging.c:757 bridge_channel_join_multithreaded: Going into a > multithreaded waitfor for bridge channel 0x292d708 of bridge 0x28f3658 > > Best,
Thorsten Göllner
2013-Feb-08 10:41 UTC
[asterisk-users] ConfBridge performance problem...?
Hi, perhaps it is a problem with your Host-Guest-Setup? Did you try the Asterisk-Setup on a dedicated server without virtualization? -Thorsten- Am 07.02.2013 11:42, schrieb Hristo Trendev:> Hi Thorsten, > > Thanks for your reply. I did check core show translations, but the > following > http://lists.digium.com/pipermail/asterisk-users/2012-November/276132.html > suggests that the values displayed are no longer representing the > computation cost only. However to answer your question: > > G722 to SLIN16 cost is 9000, reverse direction is 6000 > ALAW to SLN16 cost is 17000, reverse direction is 14500 > > G722 to SLN cost is 9600, reverse direction is 8250 > ALAW to SLN cost is 9000, reverse direction is 6000 > > With regards to the CPU usage per core - inside the VM, where only one > core is available, the CPU was close to 100% when the problem started > to apear, on the physical server with 4 cores, the cores were evenly > loaded at about 30-40%. A single call into the conference consumed > between 10-20% depending on whether I have denoise enabled or not. > > There is no dahdi board installed, I only use the dahdi module for > conference timer (note that the problem is also present with the > timerfd timing module). > > BR, > Hristo > > > On Wed, Feb 6, 2013 at 1:57 PM, Thorsten G?llner <tg at ovm-group.com > <mailto:tg at ovm-group.com>> wrote: > > Did you check > asterisk -rx "core show translation recalc 10" > > Am 06.02.2013 13:56, schrieb Thorsten G?llner: > > Sorry - I just read you alsways checked the cpu usage. Are all > cores at 100%? Is it the atserisk process which consumes it all? > > Am 06.02.2013 13:54, schrieb Thorsten G?llner: > > Did you watch the cpu usage (for example with top)? > You have a board installed which does use dahdi? Did you > check the command "dahdi_test"? > Maybe a (performance) problem of the software ec? > > Am 06.02.2013 11:13, schrieb Hristo Trendev: > > Hi, > > I have been experimenting with ConfBridge from the > asterisk-11 stable SVN branch (and with 11.2.0 also) > for the last 3 weeks and I see a problem, which what I > believe is performance related. I just wanted to ask > if someone else has made any tests and what is the > maximum number of participants that they've seen in a > conference. > > I was never able to get more than 8 participants > (mixed G722 and G711a) on a conference (actually > that's per server limit) with almost all settings on > default, except for dsp_drop_silence and denoise which > are enabled. > > I tested on Debian squeeze, 64-bit, quad-core Xeon > server @2.4GHz and also on another virtual server with > similar processor (just one core available to the VM). > While this is not the latest and greatest CPU, I would > certainly expect it to handle more than 8 calls. > > To be honest, I was in fact able to get it working for > up to 20 participants (most with G711), when I > switched from res_timing_timerfd to res_timing_dahdi > and turned off denoise, but that's still not normal I > believe, especially with most participants on mute and > with dps_drop_silence enabled and nothing else running > on the server. > > The problem itself is, that once I get over the > "critical" number of participants, the voice starts to > break up and it's impossible to understand the person > who's talking. This is certainly not bandwidth related > because all tests were made on the LAN and besides I > could see that the CPU was sometime close to 100%. > > Did someone observe something similar? > > BTW, once the first participant enters the conference > I start seeing probably over 50 messages per second > saying: > > bridging.c:757 bridge_channel_join_multithreaded: > Going into a multithreaded waitfor for bridge channel > 0x292d708 of bridge 0x28f3658 > > >-- Thorsten G?llner OVM Office Voice Media GmbH Herderstrasse 68 40237 D?sseldorf Tel.: +49(0)211 / 618 57 53 Fax: +49(0)211 / 618 57 54 -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130208/4d265329/attachment.htm>
Thorsten Göllner
2013-Feb-08 10:41 UTC
[asterisk-users] ConfBridge performance problem...?
Hi, perhaps it is a problem with your Host-Guest-Setup? Did you try the Asterisk-Setup on a dedicated server without virtualization? -Thorsten- Am 07.02.2013 11:42, schrieb Hristo Trendev:> Hi Thorsten, > > Thanks for your reply. I did check core show translations, but the > following > http://lists.digium.com/pipermail/asterisk-users/2012-November/276132.html > suggests that the values displayed are no longer representing the > computation cost only. However to answer your question: > > G722 to SLIN16 cost is 9000, reverse direction is 6000 > ALAW to SLN16 cost is 17000, reverse direction is 14500 > > G722 to SLN cost is 9600, reverse direction is 8250 > ALAW to SLN cost is 9000, reverse direction is 6000 > > With regards to the CPU usage per core - inside the VM, where only one > core is available, the CPU was close to 100% when the problem started > to apear, on the physical server with 4 cores, the cores were evenly > loaded at about 30-40%. A single call into the conference consumed > between 10-20% depending on whether I have denoise enabled or not. > > There is no dahdi board installed, I only use the dahdi module for > conference timer (note that the problem is also present with the > timerfd timing module). > > BR, > Hristo > > > On Wed, Feb 6, 2013 at 1:57 PM, Thorsten G?llner <tg at ovm-group.com > <mailto:tg at ovm-group.com>> wrote: > > Did you check > asterisk -rx "core show translation recalc 10" > > Am 06.02.2013 13:56, schrieb Thorsten G?llner: > > Sorry - I just read you alsways checked the cpu usage. Are all > cores at 100%? Is it the atserisk process which consumes it all? > > Am 06.02.2013 13:54, schrieb Thorsten G?llner: > > Did you watch the cpu usage (for example with top)? > You have a board installed which does use dahdi? Did you > check the command "dahdi_test"? > Maybe a (performance) problem of the software ec? > > Am 06.02.2013 11:13, schrieb Hristo Trendev: > > Hi, > > I have been experimenting with ConfBridge from the > asterisk-11 stable SVN branch (and with 11.2.0 also) > for the last 3 weeks and I see a problem, which what I > believe is performance related. I just wanted to ask > if someone else has made any tests and what is the > maximum number of participants that they've seen in a > conference. > > I was never able to get more than 8 participants > (mixed G722 and G711a) on a conference (actually > that's per server limit) with almost all settings on > default, except for dsp_drop_silence and denoise which > are enabled. > > I tested on Debian squeeze, 64-bit, quad-core Xeon > server @2.4GHz and also on another virtual server with > similar processor (just one core available to the VM). > While this is not the latest and greatest CPU, I would > certainly expect it to handle more than 8 calls. > > To be honest, I was in fact able to get it working for > up to 20 participants (most with G711), when I > switched from res_timing_timerfd to res_timing_dahdi > and turned off denoise, but that's still not normal I > believe, especially with most participants on mute and > with dps_drop_silence enabled and nothing else running > on the server. > > The problem itself is, that once I get over the > "critical" number of participants, the voice starts to > break up and it's impossible to understand the person > who's talking. This is certainly not bandwidth related > because all tests were made on the LAN and besides I > could see that the CPU was sometime close to 100%. > > Did someone observe something similar? > > BTW, once the first participant enters the conference > I start seeing probably over 50 messages per second > saying: > > bridging.c:757 bridge_channel_join_multithreaded: > Going into a multithreaded waitfor for bridge channel > 0x292d708 of bridge 0x28f3658 > > >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130208/2a9f04ce/attachment.htm>
Thorsten Göllner
2013-Feb-12 09:15 UTC
[asterisk-users] ConfBridge performance problem...?
Hi again, I did a try on my asterisk 11.2.1 compiled on Ubuntu 12.04 (64 bit) with a simple Pentium 4 CPU (Intel(R) Pentium(R) D CPU 2.80GHz). I connected 5 SIP-Users with a ConfBridge. This is my picture: Please give a a hint where I can change "you parameters" like denoise etc. So I will try to change these settings on my box also. I have no experience with Confbridge. -Thorsten- Am 08.02.2013 19:34, schrieb Hristo Trendev:> Hi, the quad-core server is a dedicated asterisk server. I duplicated > the tests on a virtual server (running on another physical server) > only to rule out the possibility of hardware problem with the first > sever. > > Hristo > > > On Fri, Feb 8, 2013 at 11:41 AM, Thorsten G?llner <tg at ovm-group.com > <mailto:tg at ovm-group.com>> wrote: > > Hi, > > perhaps it is a problem with your Host-Guest-Setup? Did you try > the Asterisk-Setup on a dedicated server without virtualization? > > -Thorsten- > > Am 07.02.2013 11:42, schrieb Hristo Trendev: >> Hi Thorsten, >> >> Thanks for your reply. I did check core show translations, but >> the following >> http://lists.digium.com/pipermail/asterisk-users/2012-November/276132.html >> suggests that the values displayed are no longer representing the >> computation cost only. However to answer your question: >> >> G722 to SLIN16 cost is 9000, reverse direction is 6000 >> ALAW to SLN16 cost is 17000, reverse direction is 14500 >> >> G722 to SLN cost is 9600, reverse direction is 8250 >> ALAW to SLN cost is 9000, reverse direction is 6000 >> >> With regards to the CPU usage per core - inside the VM, where >> only one core is available, the CPU was close to 100% when the >> problem started to apear, on the physical server with 4 cores, >> the cores were evenly loaded at about 30-40%. A single call into >> the conference consumed between 10-20% depending on whether I >> have denoise enabled or not. >> >> There is no dahdi board installed, I only use the dahdi module >> for conference timer (note that the problem is also present with >> the timerfd timing module). >> >> BR, >> Hristo >> >> >> On Wed, Feb 6, 2013 at 1:57 PM, Thorsten G?llner >> <tg at ovm-group.com <mailto:tg at ovm-group.com>> wrote: >> >> Did you check >> asterisk -rx "core show translation recalc 10" >> >> Am 06.02.2013 13:56, schrieb Thorsten G?llner: >> >> Sorry - I just read you alsways checked the cpu usage. >> Are all cores at 100%? Is it the atserisk process which >> consumes it all? >> >> Am 06.02.2013 13:54, schrieb Thorsten G?llner: >> >> Did you watch the cpu usage (for example with top)? >> You have a board installed which does use dahdi? Did >> you check the command "dahdi_test"? >> Maybe a (performance) problem of the software ec? >> >> Am 06.02.2013 11:13, schrieb Hristo Trendev: >> >> Hi, >> >> I have been experimenting with ConfBridge from >> the asterisk-11 stable SVN branch (and with >> 11.2.0 also) for the last 3 weeks and I see a >> problem, which what I believe is performance >> related. I just wanted to ask if someone else has >> made any tests and what is the maximum number of >> participants that they've seen in a conference. >> >> I was never able to get more than 8 participants >> (mixed G722 and G711a) on a conference (actually >> that's per server limit) with almost all settings >> on default, except for dsp_drop_silence and >> denoise which are enabled. >> >> I tested on Debian squeeze, 64-bit, quad-core >> Xeon server @2.4GHz and also on another virtual >> server with similar processor (just one core >> available to the VM). While this is not the >> latest and greatest CPU, I would certainly expect >> it to handle more than 8 calls. >> >> To be honest, I was in fact able to get it >> working for up to 20 participants (most with >> G711), when I switched from res_timing_timerfd to >> res_timing_dahdi and turned off denoise, but >> that's still not normal I believe, especially >> with most participants on mute and with >> dps_drop_silence enabled and nothing else running >> on the server. >> >> The problem itself is, that once I get over the >> "critical" number of participants, the voice >> starts to break up and it's impossible to >> understand the person who's talking. This is >> certainly not bandwidth related because all tests >> were made on the LAN and besides I could see that >> the CPU was sometime close to 100%. >> >> Did someone observe something similar? >> >> BTW, once the first participant enters the >> conference I start seeing probably over 50 >> messages per second saying: >> >> bridging.c:757 bridge_channel_join_multithreaded: >> Going into a multithreaded waitfor for bridge >> channel 0x292d708 of bridge 0x28f3658 >> >>-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130212/2deda56f/attachment.htm> -------------- next part -------------- A non-text attachment was scrubbed... Name: hdagggha.png Type: image/png Size: 10236 bytes Desc: not available URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130212/2deda56f/attachment.png> -------------- next part -------------- A non-text attachment was scrubbed... Name: cjggbefa.png Type: image/png Size: 6359 bytes Desc: not available URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20130212/2deda56f/attachment-0001.png>