Just FYI and a big thanks to Anthm and his team I have just completed a mid duration test run with FreeSwitch. Using sipp and a single core Xeon 3.06ghz machine with 1 Gig of ram, FreeSwitch ran for 27 hours under a call load of 60calls per second, with each call lasting 5 seconds each. Calls were ran in ³no media mode² passing the calls to a second FreeSwitch box on identical hardware which Answered the Calls and played back media until sipp acting as a UAC terminated the call. Total Calls Competed: 6,008,187 Total FreeSwitch Session on the relay box: 12,019,199 Average Concurrent Calls: 327 Total Time Ran 27hours My intent was to leave the test running for 48 hours and try to reach the 10million calls mark, however, I screwed up hit the wrong key (^\) and forced a segfault into sipp pulling an early abort on the test. Currently I am working out a test plan to determine several performance metrics independent of Tony and his team. If anyone has any suggestions on performace metrics and test tools you would like to see here drop me a note and let me know. K
What I would like to see is full and complete information that would allow someone to repeat the test and get similar enough results to say its the same. Specifically I would like to see: what operating system (include kernel version and libs resolved via ldd) what hardware (motherboard, cpu, if disk matters disk type, network card - and driver versions if seperate from the kernel for that stuff) sipp configs (uac.xml etc) FS configs (dialplan, sofia.conf, etc) FS version (svn tag is fine) FS modules loaded (even if they arent used, just in case) Basically enough so that it can not only be duplicated but the test itself be questioned by anyone to say its either a valid fair test or its invalid and *why* so that it can be repeated with those corrections. Benchmarking often has some criticism of the test method used, and the more we are open about the test conditions, methods, and procedures the more we can correct or defend against attacks on the testing itself, and then start to get to a way to test different platforms to see which actually work better and maybe isolate why some dont. I think these can go a long way towards unbiased benchmarks. On 10/6/07, Ken Rice <krice@suspicious.org> wrote:
Currently I am working out a test plan to determine several performance metrics independent of Tony and his team. If anyone has any suggestions on performace metrics and test tools you would like to see here drop me a note and let me know.
-- Trixter http://www.0xdecafbad.com Bret McDanel Belfast +44 28 9099 6461 US +1 516 687 5200 http://www.trxtel.com the phone company that pays you!
participants (2)
-
Ken Rice
-
Trixter aka Bret McDanel