[wesleytodd]

Front-end features have nothing to do with "scale". Does it work in the browsers? That is all that matters.

EDIT: FWIW, we also implemented this feature on our last site which was the #63 most visited site in the world at its height. So we have done it "at scale"

[okbake]

Well the front-end feature is only trivial after you've pulled the frames from the 10k titles, ensured they're spoiler free and not useless shots, and allocated the resources to cache and store them. I think thats where the scale aspect comes in to play. Theres also a certain degree of testing and integration that are effected by scale.

[duaneb]

Well, you need to find those thumbnails somewhere. So that's an entire backend component right there.

Also, I seriously doubt you anywhere near approached the scale of netflix's data centers. #63 most visited site doesn't compete at all with porn or netflix in terms of sheer bytes.

Just pointing out it's easy to say you can roll something out in a day, but you can never implement a feature without context.

[wesleytodd]

I am certainly not saying have had near the traffic of netflix. But we are a live stream site, so certainly have to deal with more load for generating the thumbnail sprites we use because we do them for every new stream, and every few minutes during the stream

The backend part of it is just ffmpeg, which can do pretty much all of this for you. The scaling is just a matter of throwing more servers at it, which is relatively trivial, expensive but simple. Admittedly my 1.5 days was not including generating the thumbnail sprites. But if you include that we had 2 programmers work on this for a total of less than a week.

[duaneb]

> The backend part of it is just ffmpeg, which can do pretty much all of this for you.

Really? it takes in a movie and spits your thumbnails, sans spoilers, sans family un-friendly content, into your data store, already indexed? Across multiple data centers?

Seems like you're making it out to be much more trivial than it is.

EDIT: To be clear, the feature itself has trivial components. But rolling out a new feature requires far, far more than just the frontend code and an ffmpeg script.

[MichaelGG]

So you need a bit more time to run it through Mechanical Turk to check for porny scenes and perhaps spoilers. I'm not sure why this is a hard problem.

But perhaps I'm tainted because I find Netflix to be annoying, on the whole. Crappy playback options, iffy streaming of HD (even when torrents have no trouble), inconsistent audio and subtitles, and an atrocious recommendation system that seems aimed at wasting my time than recommending good content. More and more I find myself using Popcorn Time, even if the content is on Netflix.

[duaneb]

You misunderstand. I'm not saying it's a hard problem, I'm saying it's a slow problem.

[fineman]

Those are goalpost-moving requirements. Netflix didn't say anything about spoilers, etc.

[duaneb]

Well among other things, netflix DOES need to ensure that e.g. sex scenes don't pop up. This is not a legal requirement, but it will drive families that trusted their kids to not anymore. Additionally, you have to take into account things like "how to avoid getting terrible shots", which is something I'm sure porn sites have looked into.

These are goalpoast moving requirements, but I also wasn't the person making it out to be trivial.

[fineman]

They do? I imagine they already have a way to block shows with sex. So if you're past that and can see the show at all, it seems moot.

Also, all these things are "good ideas", but can you imagine going to your boss and saying. "I have three-9s of the cases covered, and for the vast majority we simply refuse to show slides and almost nobody notices ... But ... I need another six months of tweaking the facial detection to get to four-9s before we ship it. To make sure nobody seems a funny face when they pause the video."

Have you considered what the expected revenue differential between the two systems is (with and without funny-face detection) and have you compared that to the cost of the time spent in a meeting discussing it, let alone the work to implement it, and then to the opportunity cost of not doing other stuff?

These comments about potential difficulties are bike-shedding in action. Nobody has anything to add so they add nothing. I'm sure we could all imagine cases where there were complex restrictions, and that'd be totally cool if these was an article about rule-engines. But it's not.

[untog]

They're common sense requirements.

[fineman]

Sure, so is a pony. If we're wishing. But features cost money. So, in other words, no - not really.

What other product offers that? How is it central to their value proposition; what customer left because ...