Hi.
Currently I am installing Openser 1.2.0 on a Sun Fire T2000 running Solaris 10. Although I used an optimized version of GCC for SPARC Systems, and used the recommended settings to generate accurate code for my Processor (UltraSPARC T1), the results of the performance tests are just 50% of the presented ones on the web page.
My setup is the following:
Proxy and UAS: Sun Fire T2000, 8 GB RAM, 1 UltraSPARC T1 1.2 GHZ 6 Core. UAC's: Two domain of a Sun Fire 12K, 4 GB RAM, 4 UlraSPARC III+ 900 Mhz Software versions listed as on the webpage.
openser -V output: version: openser 1.2.0-notls (sparc/solaris) flags: STATS: Off, USE_IPV6, USE_TCP, DISABLE_NAGLE, USE_MCAST, SHM_MEM, SHM_MMAP, PKG_MALLOC, F_MALLOC, FAST_LOCK-ADAPTIVE_WAIT-NOSMP ADAPTIVE_WAIT_LOOPS=1024, MAX_RECV_BUFFER_SIZE 262144, MAX_LISTEN 16, MAX_URI_SIZE 1024, BUF_SIZE 65535 poll method support: poll, select, /dev/poll. svnrevision: unknown @(#) $Id: main.c 1827 2007-03-12 15:22:53Z bogdan_iancu $ main.c compiled on 21:43:51 Mar 19 2007 with gcc 4.0.3
I would thank any idea about how could I improve the performnace of my setup. I think that because of the hardware specifications, the result should be at least equal than listed.
Thanks in advance.
Kind regards.
Sergio Gutiérrez.
Hello,
do you use same configs as in published tests? The UAS is another openser with just sending a 200 ok reply or sipp with uas scenario?
First check what is the capacity of sipp -- make a very simple config with just sl_send_reply("200","ok") in main route and start sipp. See what results you get and that should be the reference for sipp capacity. Also, try to use multiple instances of sipp, from different servers, see if you get the cumulative value better.
The tests were done on Core2 processor, and parallel execution of openser processes has some impact as well.
Cheers, Daniel
On 03/21/07 16:52, Sergio Gutierrez wrote:
Hi Daniel.
Thanks for your answer.
My configs are as the same used in published tests, and my UAS, as described is another instance of Openser.
I will try what you sugest.
Thanks.
Sergio Gutiérrez.
On 3/21/07, Daniel-Constantin Mierla daniel@voice-system.ro wrote:
Hi Daniel.
I am repeating the test using 6 instances of SIPp, and I am having a better performance.
A question. In the page you have listed that the Proxy and UAS runs on a 64 bit machine. Was the compilation done to generate a 64 bit executable?
Thanks.
Kind regards.
Sergio Gutierrez
On 3/21/07, Sergio Gutierrez saguti@gmail.com wrote:
Hi Daniel.
Some time ago, I tried to compile Openser for 64 bits, but I had problems with libradiusclient, when It was compiled for 64 bits.
Is that problem solved? Could I run Openser + MySQL + Radius support compiled as 64 bit?
Kind regards.
Sergio G.
On 3/21/07, Daniel-Constantin Mierla daniel@voice-system.ro wrote:
Hi.
I will do it Daniel. Thanks for the information.
About the tests, I am currently using 8 sipp instances (4 in each machine) and I am pretty closer to the performance you gained.
I will try Openser 64 bit.
Thanks again.
Kind regards.
Sergio G.
On 3/21/07, Daniel-Constantin Mierla daniel@voice-system.ro wrote:
-
Daniel-Constantin Mierla -
Sergio Gutierrez