How to build a hardware support system for MCU?

[Somnus_Chen]

Hi,

  I have integrated intel-webrtc into my project successfully.Maybe tens to hundreds of my customers will use it. 

  I'm worried about the pressures of MCU. How to build a powerful hardware support for MCU? How many MCU servers needed(1 for MongoDB,1 for RabbitMQ, manay for video/audio)? What's the configuration of every server hardware requriements(CPU,Bandwidth,GPU,RAM)? How many concurrent users  supported in this case? How much performance can improve by using intel-machine?

  Thanks.

 

[Lei_Z_Intel1]

Thanks for choosing Intel CS for WebRTC. Regarding to the system capability. this really depends your usage. E.g. what's your usage, 1 to N broadcasting, or N:N conference? Do you expect media processing in MCU or just media forwarding? If involve media processing, what's the typical usage scenario, how many users in one room and what's the resolution/bitrate choice for participants, what's the media codec? All these parameters will affect the system capacity. You can also contact webrtc_support@intel.com to share your information which is not suitable for publicly post.

Generally Intel CS for WebRTC MCU is a distributed system and can be easily expand the capacity by simply adding computing resource for those busy components.

[Somnus_Chen]

Lei Zhai (Intel) wrote:

Thanks for choosing Intel CS for WebRTC. Regarding to the system capability. this really depends your usage. E.g. what's your usage, 1 to N broadcasting, or N:N conference? Do you expect media processing in MCU or just media forwarding? If involve media processing, what's the typical usage scenario, how many users in one room and what's the resolution/bitrate choice for participants, what's the media codec? All these parameters will affect the system capacity. You can also contact webrtc_support@intel.com to share your information which is not suitable for publicly post.

Generally Intel CS for WebRTC MCU is a distributed system and can be easily expand the capacity by simply adding computing resource for those busy components.

Thanks for your reply.My usage contains 1 to N broadcasting and N to N conference, but (N) only up to 16 users in one room by mix mode.Resolution is vga(640x480), bitrate is the default value(15?), video codec is VP8, audio codec is the default value(PCMU?).

By the way,there are maybe dozens of rooms.

[Lei_Z_Intel1]

According to the condition you described, suggested deployment topology is like following. Just for your reference.

Firstly each MCU component configuration is no less than 8 logic cores and 8G memory.

1. One machine install RabbitMQ, MongoDB.

2. One machine install Nuve, Cluster Manager, Portal, Access Agent. This can well support hundreds of connections.

3.  1~many install video agent + audio agent. How many is decided  by your capacity.

      Usually the 8 logic core (4th - 6th generation Intel Core) + 8G memory machine can support 5 rooms' media processing (16 user in each room) .