| The Zen of Over-Subscription |
A few winters ago, I overheard an engineer pose this bit of bandwidth Zen to an audience of his peers: "If a channel is broadcast into a node, and nobody is watching, does it really exist?"
It got a pretty good chortle. The conversation then continued on switched digital video, and the focus of this week's translation: "Over-subscription," usually expressed with "rate" or "ratio" as a suffix.
It's a confusing term. It has nothing to do with whether or not a subscriber is getting too much of anything. Rather, it's a component in the math of video switching. It defines the number of video streams that can be allocated to the switch, before someone's channel change gets "blocked." (Think video equivalent of "please try your call again.")
Here's an example sentence, from a recent batch of notes: "One way I could deploy my next 25 HDTV streams is to switch them, knowing that I'm getting a 2x to 3x over-subscription rate -- so I can put six to nine, not three, per QAM."
What's over-subscribed in an over-subscription ratio is the number of streams packed into the pipe. A 2x over-subscription rate, or a 2:1 over-subscription ratio, means that twice the number of streams are available from the switch as could normally fit, if there wasn't a switch.
It'a confusing, because what drives the formula is the under-viewing of the video streams available through the switch. If no tuner, in any set-top, in any grouping of set-tops wants to watch that show at that time -- why send it?
Cable Needs a Video Erlang
In the olden days (late 1800s), the telephone industry was worried about having enough capacity for Mother's Day. From a traffic engineering perspective, that's their Super Bowl.
They came up with the "Erlang," eponymous with Danish mathematician Agner Krarup Erlang. It measured the probability that a given number of telephone lines would be occupied. It is still in use today.
The same problem exists with over-subscription rates, except there isn't an official formula yet. Today, the unofficial formula goes like this: Put in as many streams as the channel will hold, sans switching. Add a few more. See what happens. Repeat, until just on the safe edge of blocking.
This is why "lightly viewed" programs are the best candidates for switched distribution: If they're broadcast into a node, and nobody tunes them in, it greatly helps the math of digital video switching.
This column originally appeared in the Platforms section of Multichannel News.
