Multi-Level Cluster Promises More Effective Systems

What is Multi-Level Cluster?

Ant Media Server provides highly scalable solutions with a cluster structure. We have a blog post and documentation about cluster structure. In this blog post, we will introduce a new feature which is the multi-level cluster.

A cluster that has different regions is called Multi-Level Cluster where each region has its own origin for a stream. Ant Media Server can be scaled in different physical locations when you create CDN like a cluster. In this structure, you can distribute the nodes to regions based on their physical location. This has some advantages in performance and data-transfer. Because each region will have its own origin node for a stream, the edges in a region will pull the stream from the origin of their region.

How does it work?

Let’s clarify the case with an example scenario. We have two scenarios

1. Publisher & Players in the same cluster
2. Publisher & Players in the different clusters

Publisher & Players in the same cluster

1. Publisher starts a stream and it is assigned to the Origin1 instance in the Region1
2. Player1 who is geographically closed to Region1 instance requests to play the stream.
3. Player1 is assigned to Edge11 instance in the Region1
4. Since the Origin1 is the origin of the stream, Edge11 will pull stream from Origin1.

Multi-Level Cluster

Publisher & Players in the different clusters

1. Player2 who is geographically closed to Region2 requests to play the stream.
2. Player2 is assigned to Edge21 instance in Region2
3. Edge21 checks the origin of the stream.
   • Since there is no local origin instance in Region2 to pull the stream. It assigns itself as the secondary origin for that stream in Region2.
   • Edge21 pulls the stream from the Origin1 (main origin) in Region1.
   • Edge21 sends the stream to Player2.
4. Player3 who is geographically closed to Region2 requests to play the stream.
5. Player3 is assigned to Edge22 in Region2.
6. Edge22 checks the origin of the stream. Since Edge21 is a secondary (local) origin for the stream in the Region2, it pulls the stream from Edge21 and serves to Player3.

In short, each stream is distributed from the main node that was the closest node to the publisher. Also, all regions will have their own secondary (local) origin for a stream if someone close to that region requests to play the stream.

How to configure Multi-Level Cluster?

Region for a node is determined with the nodeGroup parameter. You can set the node group (or region) of a server by adding nodeGroup=GROUP_NAME in conf/red5.properties file. Make sure that the instances running in the same region should have the same nodeGroup value.

Also, this parameter can be set on the fly with the REST method.

Keep in mind that you should configure the Load Balancer to forward Publish and Play requests to the best region. If you have a global cluster in different geolocations, you should have a Load Balancer like AWS Route53 Service.