Desktop version

Home arrow Economics arrow Content Delivery Networks. Fundamentals, Design, and Evolution

Adaptive Bitrate - Enterprise

The first of these concerned enterprise networks. HTTP traffic is Worldwide Web traffic. To “block” this type of streaming from a corporate network requires intelligent firewalling - and so suddenly the policy decision to block streaming in the enterprise moved from “opt-in” to “opt-out.” Accordingly the argument behind the business case for streaming in the enterprise moved from “what is the business case for streaming” to “what is the business case for turning it off?”

The second advantage was also a critical issue for enterprises. Windows Media's own protocol (MMS), Flash's RTMP, and the “standard” RTSP were all complex protocols to forward into the enterprise. Each essentially required a proxy server between the LAN and the WAN, which would acquire the stream from the Internet source and then handle all subsequent requests from within the LAN.

For video-on-demand systems this caching required custom technology to set up, but this was not expressly complex.

However, when the CEO of the enterprise made a live announcement to all the staff, it became critical that the live stream was pre-configured on the proxy/gateway technology so that only a single stream would be delivered over the WAN to the thousands of recipients potentially wanting to “tune in” - and this severely limited the ability of enterprises to deploy live streaming in an ad hoc fashion.

Commercial solutions were almost invariably developed from the basic Windows or Adobe media server SDKs, and either came as part of wider proxy server solutions (often simply as a virtual machine image running on board the proxy using a common interface) or was delivered as custom-built implementations of the same SDKs. This meant there were no cheap options, and again, this inhibited enterprise live streaming adoption.

It is worth noting here that live traffic in enterprises is often concentrated around live “events” rather than 24/7 television-like streaming, and this causes very specific congestion issues on the networks involved. In a corporate LAN, the typical network is 100Mbps or even 1Gbps, so 100 users streaming at 500 kbps use a significant amount of the overall network capacity. While 30 minute on-demand files may take hours or event days or weeks to circulate an office, the number of simultaneous users will always be limited and so congestion on the network will be limited.

When the CEO makes a live announcement about forthcoming redundancies (for example) on the corporate network, the entire community may want to watch the stream, and this may saturate the networks in many ways - particularly in the absence of an IP multicast (which, again, always takes a specific and manual configuration). In the case of a modern 1Gbps LAN, it is possible that the internal network can handle most users' requests, but it is unusual for a corporate office to have a 1Gbps WAN connection between sites. This WAN connection, when it becomes saturated by the CEO's live stream, also prevents all other traffic from using the inter-site WAN resources, and given these other reasons are usually the main reason for the network existing in the first place, this makes streaming unpopular with network administrators. (The author had to help a bank pick up the pieces after a badly configured “multicast” became an unplanned unicast and the 2000 or so viewers saturated the corporate network, which also supplied the trading floor, resulting in several millions pounds of lost trades).

Adaptive bitrate HTTP streaming not only found its way through the firewall by essentially being web traffic but also critically added another benefit. Even in the case of a live stream, adaptive bitrate HTTP video is broken into small individual fragments of video transported on HTTP. These fragments can be cached by standard web proxy servers and served again to any number of users who are also requesting those streams.

While it can mean that these proxy server caches become full of chunks of video data very quickly for the duration of a live stream, this actually significantly addresses the WAN link saturation problem - only one copy of each video fragment (for each bitrate) will be copied into the proxy server. The proxy server may still serve many separate copies of that stream out over the 1Gbps

LAN, but if that LAN is only connected to a 10Mbps WAN, then most of that WAN link will still be available for all the other applications that that network is used for - so banking/web access/database referencing/VoIP, and so on, would all be relatively unaffected by the CEO's live stream, since the proxy server would be doing all the serving within the LAN and requesting just a single stream over the smaller WAN link.

< Prev   CONTENTS   Source   Next >

Related topics