Understanding The Components of a Streaming Media Network – Part 1

Content Replication
One of the main paradigms of a CDN is to replicate content across the network for scalability, redundancy and an improved end-user experience. Scalability is accomplished through replication of the content across multiple servers each capable of handling a finite number of concurrent viewers. As additional capacity is required, additional streaming servers can be brought online, the network segmented to increase bandwidth and the content then replicated to these additional servers. By replicating the content across multiple servers redundancy is accomplished allowing the servers and/or the underlying network infrastructure to fail without the end user noticing. Redundancy requires the co-ordination of the CDN load-balancing component to direct viewers away from failed hardware.

Content replication can be accomplished in a number of manners. The simplest is a point-to-point transfer, or "Push" of content from the centralized server to the streaming servers spread across the network. Push based replication can be augmented by defining when, to where content should be replicated as well as how much bandwidth should be used for the replication (to not saturate the WAN). On a corporate network, multicast can be used to replicate the content across multiple servers conserving bandwidth.

"Pull" based content replication is an on-demand caching of content at the streaming server. Instead of pushing the content to the streaming server, the content is pulled by the streaming server into cache and the delivered to the viewer only when viewer requests the file. This allows for more efficient use of the disk space (or cache) on the streaming server since only content actually requested will be placed in cache. More advanced server cache chunks of the requested file rather than the complete file conserving disk space and network bandwidth. For example, if the viewers only request the first 15 seconds of a 3-hour movie then only the first 15 seconds are cached as opposed to the entire clip.

Is content replicated by Push, Pull or a combination of both mechanisms?

- Does the mechanism support Multicast?
- Does the underlying network infrastructure support Multicast?
- Can replication be controlled by time of day?
- Can content be replicated to a subset of servers?
- Can content be replicated in chunks or only in complete files?

Page 3: Load Balancing & Client Direction >>