Stream Multiplexing for FME Failover in Flash Media Server
As some may wonder, I decided write my posts in English from now, so, anyone can follow the ideas that are discussed here. To start this new era, I decided to talk a bit about how we can do the failover of live streams in Flash Media Server through internal multiplexing of signals, that is, how we can use a single live stream as a backup of several others, without burdening bandwidth available between the Flash Media Encoder and the Flash Media Server.
To get the goal of this idea a little clearer, let’s suppose that we want to perform a live transmission, where we have 3 servers running an instance of Flash Media Encoder each, and we have 3 different signals (eg cameras), linked to a each encoder. If one of these servers crashes, the stream will no longer be published on Flash Media Server, and inevitably users connected to this content will have their experience interrupted. In order to avoid this problem, we can detect that a stream is no longer running, and then send to users the signal of another encoder, in a transparent way, without playback interruption (lost of connection or rebuffering). This approach dispenses the use of a backup server for each active server, which significantly reduces the investment in hardware and bandwidth, in environments with high availability needs.
The main concept behind this approach is the separation between the publishing streams and the viewing streams, which means that the users should not connect directly into the stream published for Flash Media Encoder. In fact, users should connect to a fake stream, and the published content will be dynamically linked to it via a server side Action Script. This idea is very well explained in the article “Building a Live Video Switcher with Flash Communication Server MX“, which was the base of multiplexing for failover concept. To automate this association between publisher stream and viewers, we can use the application.onPublish and application.onUnpublish events, which are triggered when a signal is, or ceases to be published in Flash Media Server.
The challenge now is how to control, automatically, which is the backup stream and which is the main, that is, not simply store a list of active streams, we need to know which stream will be displayed to users if a failure occurs in main stream. One approach is to isolate each stream in a different instance of application, and, in addition, create two different types of applications, one for the main streams (master) and one for the backups. Each master instance will have only one stream available so that users can connect, which means that we will need to have as many instances as the number of live signals.
The main application should store/remove from a list all the active/inactive live streams (via events), so that the first stream of the list always will be associated with the stream being viewed by users. If this stream is unpublished, the application should join the users stream to the next item, which means that the backup will always be the next stream on the list.
The backup application should receive a live signal and publish it in the main applications, performing an internal multiplexing. These internal streams will be stored on main instances as backups, so, all main applications will have the same backup streams. This internal multiplexing can be implemented in a very simple way using the NetConnection and NetStream classes, also available for server side scripting.
To make clear the idea presented above, let’s take a look at this picture:
As we can see, this approach is a simple and efficient way to implement the failover of streams published by FME, which is very important in environments where high availability is a imperative requirement, and where the investment resources for bandwidth and hardware are limited.
More information about Flash Media Server Development can be found in Adobe Developer Connection