using pipelimit at CPS (1 second) - sr-users - kamailio.org

List overview All Threads
Download

using pipelimit at CPS (1 second)

Example kamailio.cfg...

Add Record-Route on 200 OK

Vik Killa

7 Jan 2016 7 Jan '16

6:13 p.m.

Hello, has anyone used pipelimit at a timer_interval of 1 second? We would like to do rate-limiting based on CPS? The doc says it can be CPU intensive. Testing with SIPP, we see the CPU go up to around 30% at 500 CPS (with pipelimit blocking at 300 CPS) Thanks /V

Attachments:

attachment.html (text/html — 353 bytes)

Reply

Show replies by date

Alex Balashov

7 Jan 7 Jan

7:22 p.m.

Vik, We have used pipelimit at a sampling interval of 1 second, and, on the basis of contact with reality, decided to shift to a 3-second sampling window. The major problem with the 1-second interval is not the CPU usage, but that it's not very useful. Such a sampling window is extremely responsive to fluctuations within a short time, and this manifests very well in SIPP testing. However, SIPP generates calls in an orderly and linear fashion. Real-world SIP traffic is highly bursty and volatile. The result is that a 1-second sampling window will not properly encapsulate the data so as to reliably enforce a CPS limit of X. For example, if you set a limit of 300 CPS, the short sampling window makes it quite possible to burst far beyond 300 CPS as long as it is done within < 50% of the duty cycle. 3 seconds is a more realistic compromise. In virtue of smoothing some peaks, it is less dynamic and responsive, but it far more accurately takes into account the bursty and chaotic nature of real-world high-volume call setups. These are not modeled well with SIPP testing. -- Alex -- Alex Balashov | Principal | Evariste Systems LLC 303 Perimeter Center North, Suite 300 Atlanta, GA 30346 United States Tel: +1-800-250-5920 (toll-free) / +1-678-954-0671 (direct) Web: http://www.evaristesys.com/, http://www.csrpswitch.com/

Reply

Vik Killa

9 p.m.

Do you operate Kamilio in a wholesale environment with many CPS? On Thu, Jan 7, 2016 at 12:22 PM, Alex Balashov <abalashov(a)evaristesys.com> wrote:

Vik, We have used pipelimit at a sampling interval of 1 second, and, on the basis of contact with reality, decided to shift to a 3-second sampling window. The major problem with the 1-second interval is not the CPU usage, but that it's not very useful. Such a sampling window is extremely responsive to fluctuations within a short time, and this manifests very well in SIPP testing. However, SIPP generates calls in an orderly and linear fashion. Real-world SIP traffic is highly bursty and volatile. The result is that a 1-second sampling window will not properly encapsulate the data so as to reliably enforce a CPS limit of X. For example, if you set a limit of 300 CPS, the short sampling window makes it quite possible to burst far beyond 300 CPS as long as it is done within < 50% of the duty cycle. 3 seconds is a more realistic compromise. In virtue of smoothing some peaks, it is less dynamic and responsive, but it far more accurately takes into account the bursty and chaotic nature of real-world high-volume call setups. These are not modeled well with SIPP testing. -- Alex -- Alex Balashov | Principal | Evariste Systems LLC 303 Perimeter Center North, Suite 300 Atlanta, GA 30346 United States Tel: +1-800-250-5920 (toll-free) / +1-678-954-0671 (direct) Web: http://www.evaristesys.com/, http://www.csrpswitch.com/ _______________________________________________ SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list sr-users(a)lists.sip-router.org http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users

Reply

Alex Balashov

9:04 p.m.

Reply

Daniel-Constantin Mierla

8 Jan 8 Jan

10:55 a.m.

Hello, besides Alex' relevant remarks on time interval for sampling, note that getting high performances require tuning other parameters, typically across many modules. For pipelimit, be sure you increase the value for hash_size http://www.kamailio.org/docs/modules/stable/modules/pipelimit.html#pipelimi… Look also to the readme of the modules you are using and if there is a hash_size parameter, you should increase those too. Then timer processes could be important for performances as well. In the next presentation I gave some hints about bits to be adjusted for good performances: http://www.slideshare.net/miconda/kamailio-surfing-big-waves-of-sip-with-st… Cheers, Daniel On 07/01/16 20:04, Alex Balashov wrote:

Not personally, but we sell a SIP platform built on top of Kamailio that is designed specifically for that purpose. -- Alex Balashov | Principal | Evariste Systems LLC 303 Perimeter Center North, Suite 300 Atlanta, GA 30346 United States Tel: +1-800-250-5920 (toll-free) / +1-678-954-0671 (direct) Web: http://www.evaristesys.com/, http://www.csrpswitch.com/ Sent from my BlackBerry. *From: *Vik Killa *Sent: *Thursday, January 7, 2016 14:01 *To: *Kamailio (SER) - Users Mailing List *Reply To: *Kamailio (SER) - Users Mailing List *Subject: *Re: [SR-Users] using pipelimit at CPS (1 second) Do you operate Kamilio in a wholesale environment with many CPS? On Thu, Jan 7, 2016 at 12:22 PM, Alex Balashov <abalashov(a)evaristesys.com <mailto:abalashov@evaristesys.com>> wrote: Vik, We have used pipelimit at a sampling interval of 1 second, and, on the basis of contact with reality, decided to shift to a 3-second sampling window. The major problem with the 1-second interval is not the CPU usage, but that it's not very useful. Such a sampling window is extremely responsive to fluctuations within a short time, and this manifests very well in SIPP testing. However, SIPP generates calls in an orderly and linear fashion. Real-world SIP traffic is highly bursty and volatile. The result is that a 1-second sampling window will not properly encapsulate the data so as to reliably enforce a CPS limit of X. For example, if you set a limit of 300 CPS, the short sampling window makes it quite possible to burst far beyond 300 CPS as long as it is done within < 50% of the duty cycle. 3 seconds is a more realistic compromise. In virtue of smoothing some peaks, it is less dynamic and responsive, but it far more accurately takes into account the bursty and chaotic nature of real-world high-volume call setups. These are not modeled well with SIPP testing. -- Alex -- Alex Balashov | Principal | Evariste Systems LLC 303 Perimeter Center North, Suite 300 Atlanta, GA 30346 United States Tel: +1-800-250-5920 <tel:%2B1-800-250-5920> (toll-free) / +1-678-954-0671 <tel:%2B1-678-954-0671> (direct) Web: http://www.evaristesys.com/, http://www.csrpswitch.com/ _______________________________________________ SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list sr-users(a)lists.sip-router.org <mailto:sr-users@lists.sip-router.org> http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users _______________________________________________ SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list sr-users(a)lists.sip-router.org http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users

-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda Book: SIP Routing With Kamailio - http://www.asipto.com http://miconda.eu

Reply

3244

days inactive

3245

days old

sr-users@lists.kamailio.org

Manage subscription

4 comments

3 participants

tags (0)

participants (3)

Alex Balashov
Daniel-Constantin Mierla
Vik Killa