4 Out of 5 Viewers Leave If a Stream Buffers Once

38 Comments

Whatever happened to patience? More than 81 percent of all online video viewers click away if they encounter a clip rebuffering, according to a new study by Tubemogul. The Emeryville-based video distribution and analytics startup took a close look at 192 million video streams over the course of 14 days to figure out how much rebuffers matter. The result: 6.81 percent of all streams rebuffer at some point, and around 2.5 percent rebuffer twice.

Tubemogul also measure how many times rebuffers occur across several popular CDNs. During its tests, Limelight (s LLNW) performed the best, while Bitgravity’s streams had to reload the most, with Akamai (s AKAM) being somewhere in the middle.

Tubemogul’s research is based on both short-form content as well as TV shows and other longer streams. The company told me that it wasn’t able to give any specifics as to how people interact with buffering in different situations, so we won’t know, for example, if users are more forgiving with Hulu streams than with YouTube videos.

However, Tubemogul was able to measure some significant differences between various popular CDNs. Only 3.84 percent of all Limelight streams sampled for the analysis rebuffered, it said. Streams served by Akamai were interrupted roughly twice as often (7.96 percent), while Bitgravity took the dubious honor of being the CDN with the most instances of rebuffering (12.48 percent).

Those numbers are pretty significant, especially now that we know that most viewers don’t bother to stick around for a video to start playing again. As Tubemogul’s marketing director David Burch noted, “To advertisers, this means many post-rolls are routinely never seen.”

38 Comments

ObviousJoe

This is great information as it arms us with more knowledge when looking for a reliable CDN. So the message here is that we should stay away from BitGravity when shopping around for CDN? It looks like their closest competitor in terms of cost is performing more than two – three times better when encountering slow load times.

David Brazeal

I’ll bet if you broke this down by category, you’d find the defining issue is not the buffering, but the content quality. I’ll put up with buffering in my 30-minute Hulu show. But when that 2-minute YouTube video about a dancing cat buffers, it snaps me out of my trance and makes me re-evaluate whether this is really a good use of my time.

The vast majority of streams (by raw number) are the latter kind of content, so that makes the number of click-aways seem high.

David

You make a great point. Stay tuned for a follow-up bit of research breaking out longer-form video.

Michael Kuperman

Fiber to the home will not help. The problem is not related to how big your pipe is but more to consistently being able to fill the buffer and not let it run out. *Unless you are doing HD to the desktop but, even than most people on cable today can sustain 1.X Mbps.

The issue is a combination of network interconnection between various providers and I/O problems on the server typical in long tail content.

The real interesting question is, was this HTTP Progressive (YouTube) or Steaming (Like FMS/WMS) and where were the clients and on what type of connection?
Was the content long tail or short tail (popular vs. not)?
How long was it?
What was the format and bitrate of the content (HD/SD)?

Jason

Frequently a video will spend more time buffering than the length of the presentation. Two seconds of play then three seconds of buffering. I seldom experience this with Hulu or other large content providers. One problem is that some providers stop the stream at the first reload and the user must start the presentation from the beginning to view it. It frequently takes many attempts to get all the way through it. Since I experienced this I just don’t bother waiting any more.

John Pecora

What we need is to offer faster internet service here in the U.S. We have the fiber cables underground already, but we’re not using it.

Rob Silver

This study is interesting, but I would definitely be curious to see more supporting data, and if viewers were more patient with specific types of video content or platforms. As a new Internet TV network, we are just learning the importance of making our streaming content as functional as possible, especially for viewers with slower broadband connections. To think we could lose 80% of our audience on a single quick buffer seems a bit frightening.

Charles

I think you have to consider what type of content as well:

Hulu buffers for me at least once per show (>10 seconds); yet I still watch because I highly value longform professional content.

For short form I believe the study is more accurate, myself and many of my friends will immediately leave a YouTube, Break.com or other video player experience if it starts buffering or fails to load.

Janko Roettgers

That was my gut feeling as well. I’d think people are more forgiving with long-form content. You really don’t walk away from a 30 minute show if it buffers 25 minutes into the stream.

Charles

TubeMogul from what I can tell has not published their methodology or any details on how this analysis was performed. As everyone in the video community knows, a huge number of factors can impact any test including:

  • Video access
  • Time of Day
  • Video Encode Type
  • End client machine profile
  • Last mile connection speed
  • Demand on a given network at a specific time

Unless these tests were conducted in parallel or with some control the results are dubious.

David

With the help of an engineer, I wrote this study and am happy to weigh in. A lot of the issues you raise were taken into account when we designed our video delivery metrics technology (InPlay), which tracks these metrics for dozens of top publishers/platforms.

Jack Vlyn

So TubeMogul was able to tell us big CDNs perform better than small? Was the sample size and length of content the same at each CDN?

Dan Rayburn

You sure these numbers are right? On TubeMoguls’ website it links to an article that says “192 streams over two weeks”, not 192 million streams.

Janko Roettgers

Yeah, looks like Mediapost got this one wrong. Here’s what the original Tubemogul paper says: “For a 14-day period, we recorded a sample of 192,268,561 streams from six top video sites
and platforms, tracking detailed data on video delivery quality.”

192 would have been a somewhat small sample :)

Sonja Tombling

I fully agree ! The most important requirement we had when building our bespoke Web TV Player was that it loaded instantly … did we achieve that ?? Yes we did !

In fact its what our viewers love about our player , no loading issues, no buffering, it is on demand in every sense of the word, try it out for yourselves. We have yet to find anything else like it on the web.

Comments are closed.