Hi Liviu,
Thank you for the response.
2) Regarding the feedback, this is what I observed till now.
1. Without HTTP -150 calls per sec for almost 100K calls => Looks good.
2. With HTTP being sync -
a. Children parameter =1 , only 4 cps, since depending on the query and
the response, it takes time to respond.
b. Children parameter = 50, 100K calls sent at 100 cps => OpenSIPS handles
95cps - 3000+ active calls
c. Children parameter = 20, 100K calls sent at 35 cps => OpenSIPS handles
35cps
3. With HTTP being async tested with the latest fix included on main branch
=> Children parameter =8, 10K calls sent at 25cps => OpenSIPS handles
23 cps, with 1200+ active calls.
The result with async may improve in a release version from the development
branch I hope.
Regards,
Agalya
From: Liviu Chircu [mailto:li...@opensips.org]
Sent: Monday, September 19, 2016 11:18 AM
To: Ramachandran, Agalya (Contractor) <agalya_ramachand...@comcast.com>;
OpenSIPS users mailling list <users@lists.opensips.org>
Subject: Re: FW:[OpenSIPS-Users] Asynchronous operation for REST queries
Hi, Agalya!
1) Regarding connection reuse optimization, I haven't done any research on this
topic. In other words, libcurl will be managing its connections through its
default settings / parameters, whichever they are. The module is mature enough,
maybe now is a good time to look into this feature.
2) Since you're doing lots of obviously blocking operations (HTTP transfers),
then you will definitely need more processes. However, I noticed lots of
transfers are quickly completed, without requiring async, so you don't want too
many processes if all requests work like that. Just tune according to your
needs until you find a sweet spot. You could even share the performance
results, I'd love to hear some feedback.
Liviu Chircu
OpenSIPS Developer
http://www.opensips-solutions.com
On 19.09.2016 17:57, Ramachandran, Agalya (Contractor) wrote:
Hi team/Liviu,
Couple of questions for clarification.
1) I see for every HTTP request(for both sync and async), TCP ports are
being opened and closed.
Are we not using same port and pooling the request in a
particular port if we are hitting simultaneous requests?
2) Children parameter in the config file - I understand the use case of
this is to spawn multiple threads.
If we want to handle more load in OpenSIPS, we should increase this
number correct?
So when increasing this number of children process we should take care of
CPU utilization and according to the no of CPU and RAM we used in our Linux
system.
Correct me if am wrong in my understanding of the above questions.
Regards,
Agalya
From: Liviu Chircu [mailto:li...@opensips.org]
Sent: Thursday, September 15, 2016 3:53 AM
To: Ramachandran, Agalya (Contractor)
<agalya_ramachand...@comcast.com><mailto:agalya_ramachand...@comcast.com>;
OpenSIPS users mailling list
<users@lists.opensips.org><mailto:users@lists.opensips.org>
Subject: Re: FW:[OpenSIPS-Users] Asynchronous operation for REST queries
On 14.09.2016 21:33, Ramachandran, Agalya (Contractor) wrote:
Are you going to add one more parameter "connect_polling_interval" and what
will be use case of it?
Is it replacement for 10% wait time logic?
Yes.
Also, I observed in the case of async call, it takes ~90 to 100ms to open the
port itself. Is this expected behavior?
Yes.
Liviu Chircu
OpenSIPS Developer
http://www.opensips-solutions.com
_______________________________________________
Users mailing list
Users@lists.opensips.org
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
