How can I increase the number of concurrent subscribers?

sangsoo

Member
Hello~
I have a question.
How can I increase the number of concurrent subscribers?

Attachment 1 (htop + c4.2xlarge.200.png) is htop screen during stress test using two c4.2xlarge units.
-Both server's java heap have xmx8g and xms8g set up,
-More than 4000 media freee ports. (media_port_to=39999)

1. Server A played 250 streams from Server B. (using console-stress play stream)
2. Played streams in A and B separately for server comparison.
3. Streaming 220 or so slows the B server. (timer)

Load Average, CPU Avg, and Memory don't look high on htop. Why does streaming delay occur? Attachment 2 (play + c4.2xlarge.200.png)
What specifications do you need to increase to play more (300+) servers?
ex) AWS EC2 m2 / m4 instances are more appropriate.

Can I get performance data by server HW (including EC2) specifications?
I want to estimate the number of publishers & subscribers on the server.
please answer me.

Best Regards.
 

Attachments

Max

Administrator
Staff member
Good day.
1. Server A played 250 streams from Server B. (using console-stress play stream)
2. Played streams in A and B separately for server comparison.
3. Streaming 220 or so slows the B server. (timer)
In this configuration (server A and server B) you cannot test more subscribers. On server В you hit size of the Java Heap (we see in attach, 7010Mb)
Try server A (c4.4xlarge) with large memory size and specify heap size in wcs-core.properties file. For example 16 Gb where physical RAM is 32 Gb:
Code:
-Xmx16g -Xms16g
Can I get performance data by server HW (including EC2) specifications?
I want to estimate the number of publishers & subscribers on the server.
Performance depends on many factors (bitrate, transcoding, network load, e.g.). For additional help, please run test with another server A and attach output:
Code:
http://<server A>:8081/?action=stat
http://<server B>:8081/?action=stat
 

sangsoo

Member
Thank you for reply.
We will test after upgrading the A server specification.

I would like to ask more about this symptom.
In order to compare the status of the two servers, the video from A (for Subscriber) and B (for Publisher) was played in advance before the stress test. I assumed the server was stable if it streamed seamlessly. (Timer also matches almost.)

When A server Java Heap reached the limit of 7010, I could expect A server to slow down.
But rather, B server streaming was delayed. What is the reason? (see play + c4.2xlarge.200.png)
At this point, I think server B is stable (htop + c4.2xlarge.200.png, 8:36:29.11, 2 seconds slower than server A). Often slower than 10 seconds.
If the degrade streams field is looked up in the REST API "http://~?action=stat", Is this the maximum performance of the server?
I'm curious about this symptom.

Best Regards.
 

Attachments

Last edited:

Max

Administrator
Staff member
Good day.
When A server Java Heap reached the limit of 7010, I could expect A server to slow down.
But rather, B server streaming was delayed. What is the reason? (see play + c4.2xlarge.200.png)
When server A ran out of heap, it could not subscribe for new stream. For more investigations we need logs from server B (directory /usr/local/FlashphonerWebCallServer/logs/*)
If the degrade streams field is looked up in the REST API "http://~?action=stat", Is this the maximum performance of the server?
We will check and let you know.
 

sangsoo

Member
Thank you for reply.
All logs are deleted now (daily delete).
I'll try some more tests.
If you have any questions, I will attach the log.
Best Regards.
 

Max

Administrator
Staff member
Good day.
Performance tips in your case:
1. Select more performance Amazon instance (with large RAM and possibly multiple CPUs).
2. Configure the Z Garbage Collector (Java 12 required) according to the recommendations.
3. You have enough media ports (media_port_from, media_port_to). But by default, no more than 1000 streams are pulled in WCS load server. To expand this limit (wcs_agent_port_from, wcs_agent_port_to) see these documentations.
 
Last edited:
Top