Mu App Quadrant #1 – Not all video streaming apps are created equal…

April 13, 2011 by Simon Berman

There’s been a lot of debate recently about the impact of video apps on the network. According to Nielsen, Netflix alone now accounts for 20% of downstream traffic during peak times in the United States. In a previous blog Kowsik explained the behind-the-scenes interactions that are happening unbeknownst to you when you watch a Netflix movie. So that got us thinking – are all the popular video apps as network-intensive as Netflix? Are some video apps more user-friendly with their bandwidth consumption than others? Are some video apps more operator-friendly with their consumption of networking resources?

The Mu App Quadrant
In this first edition of the Mu App Quadrant we dug deeper into this topic to look at the most popular desktop video apps – Netflix, YouTube, Hulu and Amazon – which we then compared across 2 dimensions:

  • Consumer-friendliness: The less bandwidth an app consumes, the more Consumer-friendly it is
  • Operator-friendliness: The fewer connections established by an app, the more Operator-friendly it is

We were surprised to see the results showing that – relative to each other – some video streaming apps were considerably more consumer- and/or operator-friendly than others. As in golf, a lower score is better in either dimension, so our quadrant is drawn with negative percentages in the upper right. 

Top video streaming apps are compared in terms of their Consumer-friendliness and Operator-friendliness

Top video streaming apps are compared in terms of their Consumer-friendliness and Operator-friendliness  

Can you see me now?
What can account for these differences? At a high level, application developers of video streaming services are probably not even aware that their choices can have such major consequences. For instance, a developer of a particular service might choose to use a very high-quality video CODEC because they want their customers to receive a sharp picture. Some CODECs will achieve higher quality by consuming more CPU on the receiver, while others will send more data. If the developer is even aware of the consequences resulting from this choice, they probably never connected the dots between their decisions resulting in the cost to the network, or to the consumer. Many of these costs do not manifest themselves during development when testing a few dozen clients. It’s only much later after a full scale rollout that the true impact of the service is understood.

Methodology
The traffic from diverse video streaming services was sampled multiple times, for different videos, and samples for both HD (High-Definition) and SD (Standard-Definition) were captured. Each sample was several minutes long. The comparisons were based on the rate of connections (per second) and the rate of bandwidth consumed (per second).

By comparing rates instead of absolute numbers, we established a level playing field to make meaningful apples-to-apples comparisons, even for video samples of different lengths. The center point of the graph is the average (i.e., un-weighted arithmetic mean) and each data point is depicted as a percentage better (or worse) than the mean on either axis.

To read the full report, click here.

Turns out that not all video apps are created equal.

Categories: Android, Cloud, DPI, iPhone, Netflix, Research, Scale, Testing, Video
12,246 Comments
Bookmark and Share