What is a reasonable update interval for video analytics (percent watched)?

I'm developing a custom analytics solution for videos on my site... the functions are mapped to both a Flash player and a HTML5 <video> player.

General design advice would be greatly appreciated, but I'm specifically curious how often I should update the percent watched value... I was thinking I'd set a certain percentage threshold rather than according to a time (say, every 5 or 10 percentage points).

Then, as far as sending the data back to the server, I was thinking I'd just use regular Ajax POST requests, which would update a MySQL database.

Thing is, when I researched the Vimeo and YouTube HTML5 players, I couldn't detect how they were handling analytics — so I'm wondering if they have a better implementation (?), as I can't imagine that they don't have any analytics.

Sending 20 or 10 Ajax requests for each video play seems somewhat inefficient (and that's quite a few MySQL UPDATEs, too) — and the fidelity isn't that high even.

(Sorry for the somewhat scattershot question — I can't find much info on the subject.)

POST-ANSWER EDIT

How can Wistia boast the开发者_StackOverflow中文版y they track "second-by-second viewing statistics for individual views"? (That is, without sending a huge number of requests.)

I would send update at times 50% 25% 12.5% 6.125% ... and so on (in reverse order) until time resolution gets too high (from a few seconds up to minute maybe.. your choice). That way you'll know which videos are watched fully/largely, and which ones are quickly dismissed.

There are of course many variations possible, but the key idea is to have best resolution in the beginning.

Track at percentages that will be useful for your reporting needs. For clients that need to know the general "health" of their videos, tracking at 50% and 100% completion rates is enough for most. Beyond that, tracking at 25%, 50%, 75% and 100% gives a really good amount of information.

Think about why you'd need to know that somebody saw 5% of a video vs 10% of a video. If you can think of some way to make improvements at that level, then by all means, track to that level. Otherwise, do what will get you the best reporting impact with the least load impact.