How users use

Sorry folks, this one is going to be a bit dense.

One of the arguments we always have with clients’ IT people about intra.tv prior to the event is about concurrent views of videos. “What if everyone views at once? The network will fall down!”, they say. And we calmly go about soothing their nerves and pointing out that a) people don’t all watch at once and b) even if they all watch on day one, videos are short and viewers distribute themselves across the time span and don’t watch concurrently.

I always feel on slightly thin ice at this point, so I was thrilled to read about an academic paper which shows that…

 

... obviously.

... obviously.

 
These boffins studied viewing patterns on YouTube and identified two types of videos: “endogenous” – chosen by the community (or more commonly, “virals”) – and “exogenous” which are selected by YouTube editors and featured on the homepage.

 

Clearly...

Clearly...

Endogenous is on the left, and as you’d expect the viewing numbers grow slowly as users recommend the videos to their friends and then tail off. Exogenous videos, on the right, burst onto the scene when recommended and then tail off quickly. [Tries to recall maths from dim past]. This is a “power law relaxation”, isn’t it?
  

The chart on the right is particularly interesting, because this is a pretty good analogue for an internally launched video. Employees get an email telling them to watch, and the keen ones do immediately, the keen busy ones do as soon as they can, and then the interest tails away as the long tail of people get round to watching.

The chart is a logarithmic scale, so using Photoshop and Excel to very, very roughly map the numbers onto a linear scale you get a curve something like this:

 

Conversely...

Conversely...

And because my brain is too small to handle power law relaxations, bodging on a rule of thumb that viewership declines by 75% per day we get a pretty good match:

 

Meanwhile...

Meanwhile...

Where this gets useful to us is that if we accept this viewing pattern, we can use the curve to plot a likely maximum viewership.

Let’s say that 10,000 eager employees are going to view the latest video from the CEO. Using a 75% decay rate, we can redraw the pink line so that the data points add up to 10,000. And you get this:

Therefore...

Therefore...

And we can see that the peak day has almost exactly a quarter of the total views: 2,508 to be exact.

That’s as far as this data can get us, but a client recently gave us the raw logs from their last couple of streamed videos and they showed peaks early in the morning and at lunch time,  like this:

 

On the other hand...

On the other hand...

19% of the peak day’s views fell in the peak hour, so applying that number to the peak day figure from our 10,000 audience – 2508 – you get 481 views.

And let’s say the video is ten minutes long, then divide the peak hour by six and you get a peak concurrency of… 80. That’s all.

Nothing to worry about…

Thanks to NewTeeVee for posting about this, btw.

Advertisements

Tags: , , , ,

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: