[spacko]

> When you hover your mouse over a title, you will now see a slideshow of images from that movie or show. We hope this slideshow will give you a better feel for what the show is about than reading the description alone.

Well, this is a standard feature of any average Joe's porn site for about 5 years now

[harlanlewis]

There may be different expectations regarding spoilers.

[wesleytodd]

Built this for our site (stream.me) in like a day and a half. Makes you feel good about your work when big companies with much bigger teams "announce" features you have already tackled :)

[duaneb]

I would imagine it's much harder to roll out at scale.

Netflix's ui has been pretty terrible for a while, though, I wish they would have an alternative to the grid view. Sometimes I can't find the genre I'm looking for.

[bdcravens]

I imagine there's also licensing issues that aren't present in user or publisher submitted content.

[mmahemoff]

Shouldn't be a scale issue. The number of previews required is O(N) where N is the number of videos, not the number of users, and for Netflix, that's very small compared to a user-generated video site like YouTube. With 10,000 videos (what seems to be an estimate), you could spin up a bunch of EC2 instances and render them in under an hour.

As for serving them, it's also trivial to host on a CDN and serve them on demand (on rollover).

Possibly they'd need to work it into their licensing agreements.

[duaneb]

I'm more talking about the data migrations necessary to support the thumbnail fetching. At that scale (including all the bureaucracy), it's slow.

[ryaso]

And (likely) tooling for altering/fixing samples, and (likely) updating their transcoding process to include thumbnail generation at all the various sizes and qualities they encode for, and backfilling all the existing titles to fit into all such tooling, and coordinating such a project across what are likely multiple teams with different roadmaps and reporting structures, and doing it all without service interruption (lockstep deployments of all the various moving parts).

So much slow in large organizations. So much.

[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.

[taytus]

How do you know?