Alexander Olekhnovich
2008-May-16 07:47 UTC
[asterisk-users] Asterisk concurrent calls count
Hi Asterisk Users, I'm interested in how many concurrent calls Asterisk can process without troubles. I mean 1 Asterisk server (software) like either proxy or media server (any numbers will be appropriate). 1. Is there any limitations by the software? What is this number? 2. What is the maximum count of concurrent calls you've ever seen/tested? -- Thanks in advance Alexander Olekhnovich -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.digium.com/pipermail/asterisk-users/attachments/20080516/76982d61/attachment.htm
Hello, Alexander. AO> Hi Asterisk Users, AO> I'm interested in how many concurrent calls Asterisk can process without AO> troubles. I mean 1 Asterisk server (software) like either proxy or media AO> server (any numbers will be appropriate). AO> 1. Is there any limitations by the software? What is this number? AO> 2. What is the maximum count of concurrent calls you've ever seen/tested? Look at this example http://www.transnexus.com/White%20Papers/asterisk_V1-4-11_performance.htm -- Alexey mailto:shimeshov at rontel.ru
> I'm interested in how many concurrent calls Asterisk can process without > troubles. I mean 1 Asterisk server (software) like either proxy or media > server (any numbers will be appropriate).Since one standard answer to this question is: "it depends on how you're using it", The ideal situation is that people could rattle off statistics of their eventual load, and be able to size their hardware purchase accordingly. The reality is that while that's hard, we can do the next best thing, which is once you have the hardware running asterisk, get historical data about your real-world asterisk load. We're running the open SNMP daemon, and we've configured the open software project Cacti to do SNMP polling against our cpu load. We now have a few months of data on how two systems running Zaptel cards, with no VoIP are holding up under load. Our business is amazingly seasonal, not quite as bad as H & R Block, but similar scenario where we're very busy parts of the year, and the rest of the year, not so much. Our results: at our US Eastern time zone business, load peaks a little after 2pm EST/EDT, most business days, and dramatically tails off most days. Once we have more months of data we'll also be able to more accurately profile the seasonality of our business, as well as make some predictions about next peaks from previous peaks, given the growth rate of our business. Hope this helps people!
On Fri, 16 May 2008, Steve Totaro wrote:> On Fri, May 16, 2008 at 9:47 AM, David Backeberg <dbackeberg at gmail.com> wrote: >> >> Has anybody ever tried to roll their own VoIP or Zaptel load >> simulator? How did they do it? >> > SIPP can help with benchmarking SIP calls and you can loop back T1 > calls if you have two machines with T1 cards or even one machine with > multiple T1 ports. > > Then just look at top. Make a few test calls and see if they are choppy....What value do you look at with top? (Especially with multiple processor/core servers.) I have an old 1.2.7 server with "custom features" hacked in that leaks memory. We know audio quality goes to hell when Asterisk hoards more than 100mb. How do you quantify "choppy?" Anybody volunteer to write "app-MOS?" Thanks in advance, ------------------------------------------------------------------------ Steve Edwards sedwards at sedwards.com Voice: +1-760-468-3867 PST Newline Fax: +1-760-731-3000
Alexander Olekhnovich wrote:> Hi Asterisk Users, > > I'm interested in how many concurrent calls Asterisk can process > without troubles. I mean 1 Asterisk server (software) like either > proxy or media server (any numbers will be appropriate). > > 1. Is there any limitations by the software? What is this number? > 2. What is the maximum count of concurrent calls you've ever seen/tested? > > -- > Thanks in advance > Alexander Olekhnovich > ------------------------------------------------------------------------ > > _______________________________________________ > -- Bandwidth and Colocation Provided by http://www.api-digital.com -- > > asterisk-users mailing list > To UNSUBSCRIBE or update options visit: > http://lists.digium.com/mailman/listinfo/asterisk-usersRather than jump into the heavy list of replies, in which there's some heated discussion, I thought I'd offer a quick $0.02: Asterisk's concurrent call capabilities is limited (as far as I know) only by the hardware you're using and the implementation. By this I mean that the amount of transcoding, meetme conferences, SIP/IAX/Zap channels, recording, CDR backend, etc...all take their toll on your hardware's capabilities. I'll give you two examples: 1. On a Dual 1.5Ghz XEON, 2GB RAM server running CentOS 4.5(unsure on this anymore) with only Asterisk 1.4 TRUNK in 1995 in a SIP only environment with ONLY ulaw encoding, I've seen 500+ concurrent calls with over 2K users on a single machine. All clients were set for canreinvite=no, and qualify=yes. This system did not show degradation of performance. 2. I'm currently working with a client that has a Dual 2.5 Ghz, 2GB RAM server, running Debian Etch. They are running two EM Wink T1 Trunks, and 51 Zap phones locally running through Adtran Total Access Channel Banks, 12 POTS lines running through a Rhino channel bank, and 27 SIP Phones. Concurrent calls only run at around 43 calls currently, although I've seen it as high as 53, and ALL calls are recorded other than local spying on channels and inter-extension calls. Additionally, this server has PostgreSQL and Apache running on it to allow administration to review CDRs and pull recordings, and a Zabbix monitoring agent daemon sending data to a local network Zabbix server. This server showed little or no degradation in call quality or service (even with Sox and Speexmix running in the background converting recordings via a background script) until just recently when we changed T1 providers and got EM Wink instead of the requested PRI. Before we had 99.999% of all calls complete from dial to hangup with no issues. Now we're at 98.8%, with calls being dropped in midconversation. I have not found the answer to what is causing the server to drop calls, other than after the switchover to EM_W our Zaptel accuracy started degrading. We are in the process of figuring out how we can resolve this, including possible hardware upgrades (which were already planned for handling recordings better) I hope these two examples help show you how two similar machines can vary drastically in performance with similar hardware. Differences in implementation make a BIG difference. Slainte, Sherwood McGowan
On Mon, May 19, 2008 at 3:03 AM, Tzafrir Cohen <tzafrir.cohen at xorcom.com> wrote:> On Mon, May 19, 2008 at 12:13:05AM -0400, Steve Totaro wrote: > >> I am complaining that they should be provided by Digium. I have an >> early source of some funding for benchmarking, so it certainly will >> not be free. To the vendors it will. I will do their jobs for them. > > So far you have not come up with a description of a benchmark. > > You have not even described clearly what it is that you want to test. > > -- > Tzafrir Cohen > icq#16849755 jabber:tzafrir.cohen at xorcom.com > +972-50-7952406 mailto:tzafrir.cohen at xorcom.com > http://www.xorcom.com iax:guest at local.xorcom.com/tzafrir >Yes, that is right "so far". Very observant, although the thread title may be a clue...... Thanks, Steve Totaro
On Mon, May 19, 2008 at 07:27:00AM -0400, Steve Totaro wrote:> On Mon, May 19, 2008 at 3:03 AM, Tzafrir Cohen <tzafrir.cohen at xorcom.com> wrote: > > On Mon, May 19, 2008 at 12:13:05AM -0400, Steve Totaro wrote: > > > >> I am complaining that they should be provided by Digium. I have an > >> early source of some funding for benchmarking, so it certainly will > >> not be free. To the vendors it will. I will do their jobs for them. > > > > So far you have not come up with a description of a benchmark. > > > > You have not even described clearly what it is that you want to test.> Yes, that is right "so far". Very observant, although the thread > title may be a clue......As others have noted, this is mostly mmeaningless. I think I can easily get some 1000-s of channels running on this Asteirsk instance on my desktop. Yeah, those would be Local channels and will push no frames at all. But who cares: my great PBX has many concurrent calls. -- Tzafrir Cohen icq#16849755 jabber:tzafrir.cohen at xorcom.com +972-50-7952406 mailto:tzafrir.cohen at xorcom.com http://www.xorcom.com iax:guest at local.xorcom.com/tzafrir
On Mon, May 19, 2008 at 7:39 AM, Tzafrir Cohen <tzafrir.cohen at xorcom.com> wrote:> On Mon, May 19, 2008 at 07:27:00AM -0400, Steve Totaro wrote: >> On Mon, May 19, 2008 at 3:03 AM, Tzafrir Cohen <tzafrir.cohen at xorcom.com> wrote: >> > On Mon, May 19, 2008 at 12:13:05AM -0400, Steve Totaro wrote: >> > >> >> I am complaining that they should be provided by Digium. I have an >> >> early source of some funding for benchmarking, so it certainly will >> >> not be free. To the vendors it will. I will do their jobs for them. >> > >> > So far you have not come up with a description of a benchmark. >> > >> > You have not even described clearly what it is that you want to test. > >> Yes, that is right "so far". Very observant, although the thread >> title may be a clue...... > > As others have noted, this is mostly mmeaningless. > > I think I can easily get some 1000-s of channels running on this > Asteirsk instance on my desktop. > > Yeah, those would be Local channels and will push no frames at all. But > who cares: my great PBX has many concurrent calls. > > -- > Tzafrir Cohen > icq#16849755 jabber:tzafrir.cohen at xorcom.com > +972-50-7952406 mailto:tzafrir.cohen at xorcom.com > http://www.xorcom.com iax:guest at local.xorcom.com/tzafrir > > _______________________________________________ > -- Bandwidth and Colocation Provided by http://www.api-digital.com -- >Whatever, vendor....