Tuesday, September 18, 2007

Asterisk Performance Testing

Over the last couple of months we have been stress testing Asterisk V1.4 to understand its performance. We started using V1.4.4 and advanced to V1.4.11 as new releases were made available. We have been stress testing Asterisk configured as a SIP Back to Back User Agent (B2BUA) installed on a server with two, dual core, Xeon 5140 CPUs running at 2.33 GHz.

The performance results look very good. When there is no transcoding, Asterisk can handle approximately 1500 simultaneous calls. When using a $3000 server to host Asterisk, this works out to be $2 per port which is a very low investment. When calls are transcoded from the G.711 codec to G.729, the maximum number of simultaneous calls is approximately 400. The economics are quite different when transcoding is required since there is a $10 per port patent royalty for using the G.711 to G.729 transcoding algorithm. The per port cost for using Asterisk with G.711 to G.729 transcoding is $17.50 per port. (Note, the $10 per port patent royalty is the amount charged if you purchase a transcoding license from Digium. OEM vendors could negotiate a lower royalty fee directly from the patent holders, but this is not practical for end users.)

More importantly, we have seen a big improvement is stability between V1.4.4 and V1.4.11. We have found that V1.4.11 has been very stable in prolonged, high volume stress tests. We expect that V1.4.11 will be the foundation of Asterisk Business Edition Rev C which should be released later this year. We see Asterisk Business Edition Rev C to be a quantum leap forward compared to earlier versions of Asterisk Business Edition.

We will publish all the details of our test plan, a summary of the test results and all the raw data on the TransNexus web site (www.TransNexus.com) in the coming weeks.

If you have any comments or experience sizing server hardware for Asterisk in carrier operations please post your information here.

Thank you

10 comments:

Anonymous said...

Jim - We would be very interested in seeing your test bed setup and results. Any idea when you'll be publishing this?

Jim Dalton said...

The test results are now available at http://www.transnexus.com/White%20Papers/asterisk_V1-4-11_performance.htm

Anonymous said...

hello jim, I'm a software tester. I have seen your blog while searching about testing of asterisk. I saw the results of the test on your web site. that's really interesting. I have some questions to ask you:
- did you use any tools, such as performance test tools to conduct the test?
- how did you measure the using of the resources system during the test ?

thanks in advance for your response

Jim Dalton said...

We used SIP-P to generate the test load and measured system resources using TOP. You can find all the details at: http://www.transnexus.com/White%20Papers/asterisk_V1-4-11_performance.htm

We have improved our performance test plan based on feedback from the Asterisk community and we will publish an updated test, with all the details, this Summer.

cheawoh said...

Thanks jim for your response. I will expect the update of your test plan.
I have already token a look on SIPp and it's very useful. but i got some problem to use correctly TOP to measure the resources systems. I have got an other tool SAR which can record in a file, the using of resources system during the test. Please, take a few moment to have a look on this tool and let me know, if possible, your opinion about it. Thanks

Jim Dalton said...

Hi Cheawoh,

Thank you for your suggestion.

We also use SAR to collect server CPU , memory and I/O utilization. This information will be included in the performance test when it is published.

Cheawoh said...

Hello Jim,
As you know, we are really interested to see the update of your performance test plan. Please, can you tell us when it will be published?

Jim Dalton said...

Hello Cheowah,

Thanks for your interest and follow-up. The Asterisk performance test is still on our roadmap, but several higher priority projects have forced us to delay this project for several months. I hope we will have results to publish by October 2008.

Morgan said...

Hey Jim, just came across your post about this, even though it was 4 years ago, this situation still applies since Asterisk just released version 10. If you're still into this, then i'd be very interested in helping out or getting some performance numbers on how 1.8 or v10 is coming along.
Thanks
Morgan

Jim Dalton said...

Hello Morgan - thank you for your interest. We have discussed performing a benchmark of Asterisk V1.8 compared to FreeSWITCH. We do not have a date set in our work plans for this project, but it is something we want to do in 2012.

BTW, this blog has moved to blog.transnexus.com