[Users] Compiling Openser for best performance on Solaris SPARC

List overview All Threads
Download

newer

older

[Serusers] re: composing SIP...

[Serusers] Acc not writing to MySQL

Sergio Gutierrez

21 Mar 2007 21 Mar '07

4:52 p.m.

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.

Attachments:

attachment.html (text/html — 1.4 KB)

Show replies by date

Daniel-Constantin Mierla

21 Mar 21 Mar

5:07 p.m.

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:

...

Sergio Gutierrez

5:35 p.m.

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(a)voice-system.ro> wrote:

...

Sergio Gutierrez

8:14 p.m.

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(a)gmail.com> wrote:

...

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.

------------------------------------------------------------------------

_______________________________________________ Users mailing list Users(a)openser.org http://openser.org/cgi-bin/mailman/listinfo/users

Daniel-Constantin Mierla

9:34 p.m.

Hello, On 03/21/07 20:14, Sergio Gutierrez wrote:

...

yes, it was 64bit executable. OS is Debian amd64. Cheers, Daniel

...

Thanks. Kind regards. Sergio Gutierrez On 3/21/07, *Sergio Gutierrez* <saguti(a)gmail.com <mailto:saguti@gmail.com>> 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(a)voice-system.ro <mailto:daniel@voice-system.ro>> wrote: 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. 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

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. ------------------------------------------------------------------------ _______________________________________________ Users mailing list Users(a)openser.org <mailto:Users@openser.org> http://openser.org/cgi-bin/mailman/listinfo/users

Sergio Gutierrez

9:44 p.m.

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(a)voice-system.ro> wrote:

...

Hello, On 03/21/07 20:14, Sergio Gutierrez wrote:

yes, it was 64bit executable. OS is Debian amd64. Cheers, Daniel

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

openser processes has some impact as well. Cheers, Daniel On 03/21/07 16:52, Sergio Gutierrez wrote:

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,

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. > >

------------------------------------------------------------------------

_______________________________________________ Users mailing list Users(a)openser.org <mailto:Users@openser.org> http://openser.org/cgi-bin/mailman/listinfo/users

Daniel-Constantin Mierla

9:54 p.m.

Hello, On 03/21/07 21:44, Sergio Gutierrez wrote:

...

haven't tried myself, but as I could see, there were several fixes for libradiusclient on 64b. You should give another try. Cheers, Daniel

...

Kind regards. Sergio G. On 3/21/07, *Daniel-Constantin Mierla* <daniel(a)voice-system.ro <mailto:daniel@voice-system.ro>> wrote: Hello, On 03/21/07 20:14, Sergio Gutierrez 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?

yes, it was 64bit executable. OS is Debian amd64. Cheers, Daniel

Thanks. Kind regards. Sergio Gutierrez On 3/21/07, *Sergio Gutierrez* <saguti(a)gmail.com

<mailto:saguti@gmail.com>

<mailto:saguti@gmail.com <mailto:saguti@gmail.com>>> 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(a)voice-system.ro <mailto:daniel@voice-system.ro>

<mailto:daniel@voice-system.ro

<mailto:daniel@voice-system.ro>>> wrote:

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. > > 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,

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. > >

------------------------------------------------------------------------

> > _______________________________________________ > Users mailing list > Users(a)openser.org <mailto:Users@openser.org>

<mailto:Users@openser.org <mailto:Users@openser.org>>

http://openser.org/cgi-bin/mailman/listinfo/users

Sergio Gutierrez

10:04 p.m.

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(a)voice-system.ro> wrote:

...

Hello, On 03/21/07 21:44, Sergio Gutierrez wrote:

haven't tried myself, but as I could see, there were several fixes for libradiusclient on 64b. You should give another try. Cheers, Daniel

Kind regards. Sergio G. On 3/21/07, *Daniel-Constantin Mierla* <daniel(a)voice-system.ro <mailto: daniel(a)voice-system.ro>> wrote: Hello, On 03/21/07 20:14, Sergio Gutierrez wrote: > Hi Daniel. > > I am repeating the test using 6 instances of SIPp, and I am having

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?

yes, it was 64bit executable. OS is Debian amd64. Cheers, Daniel

Thanks. Kind regards. Sergio Gutierrez On 3/21/07, *Sergio Gutierrez* <saguti(a)gmail.com

<mailto: saguti(a)gmail.com>

<mailto:saguti@gmail.com <mailto:saguti@gmail.com>>> 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(a)voice-system.ro <mailto:daniel@voice-system.ro>

<mailto: daniel(a)voice-system.ro

<mailto:daniel@voice-system.ro>>> wrote:

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. > > 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. > >

------------------------------------------------------------------------

> > _______________________________________________ > Users mailing list > Users(a)openser.org <mailto:Users@openser.org>

<mailto:Users@openser.org <mailto: Users(a)openser.org>>

http://openser.org/cgi-bin/mailman/listinfo/users

6457

days inactive

6457

days old

sr-users@lists.kamailio.org

Manage subscription

7 comments

2 participants

tags (0)

participants (2)

Daniel-Constantin Mierla
Sergio Gutierrez