CDN Optimization: Measure What Matters Most

Many major online video publishers and CDNs are embracing a multi-CDN strategy to optimize video delivery.

Effective, multi-CDN workflows require proactive, intelligent decisioning driven by predictive analytics, relying on a very different approach to video streaming measurement.

Comcast CDN now includes DLVR’s multi-CDN optimization, a video delivery optimization service to automate intelligent, multi-CDN decisions to power peak streaming performance for every viewer on every device. Cloud-based solutions like DLVR use Direct Stream Measurement (DSM), an approach that has a number of advantages.

DSM does exactly what it says it does – it measures streams directly while tag-based measurement doesn’t. Instead, tag-based systems add a JavaScript tag to each web page (typically in the page footer); once the page has finished loading, the tag issues one to three test object requests from one or more CDNs.

So, why bother measuring actual streams vs measuring test object requests via tags? There’s actually a number of reasons:

  • DSM measures actual video delivery experience metrics. Direct stream measurement gathers data on video viewing metrics including buffering events, downshifts, and the bit rate of the stream delivered to the user by measuring actual video content delivered to every device. Using DSM along with real-time machine learning, rebuffering rates and bitrate downshifts can be continuously measured, updated, and predicted, enabling every viewer to be routed to the best-performing CDN for their particular ISP, geographic location, and playback device. DSM not only informs proactive optimizations, DSM is the only method that can tell you why buffering  downshifts and low quality streams happened.
  • Tags provide aggregate data. Tag-based services call this “crowd sourcing” or “community data,” averaging metrics across all publishers and all viewers, rather than the actual performance specific to any one publisher, let alone any particular viewer. To calculate a statistically meaningful average of aggregated test object performance for each of dozens of CDNs, a tag-based service needs very large numbers of small test object measurements, and that drives aggregation, not precision.
  • DSM captures the performance of a customer’s specific CDN resource “map.” In current CDN architecture, each customer is mapped to a particular subset of each CDN’s resources and each customer’s users are distributed differently – geographically, across access networks, and across devices. This CDN resource mapping contributes to mismeasurement by tag-based systems – the aggregate data collected from tag-based test object requests does not accurately align with the actual resource maps dedicated to each customer. DSM does not suffer the same flaw because it identifies the specific CDN resources and other factors that affect performance and tailors each individual user's CDN selection to their particular network, geographic location, specific device, and viewing experience.
  • DSM provides performance measurement throughout video playback. Data is captured continuously, typically performing 45-50 measurements or more during a 22-minute stream, providing the intelligence to switch a user to a better performing CDN at any time. Tag-based services measure a CDN’s test object performance only once for any one user – when the page load completes.
  • DSM can distinguish between network access performance and CDN performance. Cloud-based, direct stream measurement is able to break out metrics for network access providers and CDNs. Tag-based solutions can’t distinguish access network issues from CDN issues because they don’t measure CDNs and access networks separately – test object requests performed by a device-based tag collapse CDN performance and access network performance into a single measurement.
  • Cloud-based DSM requires no initial or ongoing client development and test cycles. Cloud-based direct stream measurement eliminates client or SDK integrations completely. Tag-based solutions, on the other hand, require player development and test cycles. Browser-based tags must be deployed on enough pages and properties to generate a minimum volume of traffic while matching video traffic patterns. Developers need to update all the pages, properties, and players where the tags are deployed to keep up with any changes from the tag-based service provider. Such deployments typically exclude mobile devices that could incur charges from sending test object data.
  • Cloud-based DSM eliminates client or SDK integrations which has two major benefits to video publishers:
  1. No development or QA resources needed for integration with apps, players or webpages. 
  2. Future-proofed for new devices. DSM measures streams played on new devices the day they hit the market.

Predict and Avoid CDN Outages with Intelligent Real-time Decisioning

Better data through DSM provides better decisioning which powers video delivery optimization. Cloud-based, direct stream measurement enables real-time decisioning helping publishers avoid the impact from CDN outages and congestion.