How does Netflix work?

More specifically, what are the logistics of handling thousands and thousands of DVDs at Neflix HQ? I realize that there are multiple regional centers, let’s just talk about what’s going on at a particular center, and leave aside the occasional need to ship something from a different center.

I am mailing back my current DVD today. They will receive it tomorrow. A machine will open the envelope and extract the disk in its bar-coded sleeve. A scanner will read the bar code and identify the disk, then the computer will say “twicks had this, let’s assume she’s the one returning it.” The computer then credits my account with the return and cues another program to send me the receipt confirmation email.

Then what?

The physical disk goes … where? Does the stream bifurcate immediately, into “this is at the top of someone else’s queue” and “this can go back onto the shelf”? Do they actually check queues at that point or do they have an algorithm indicating relative popularity? (Recent releases of popular movies obviously get a lot of action – some of my beloved '40s musicals presumably don’t.) For DVDs that are “shelved,” I assume that’s also done by machine – is there a way of storing them so that the bar codes can be read easily to find a particular disk when the next request for it is made?

For now, let’s assume that the disk I’m returning is a popular one that goes into the “going out today” stream. Say there are 25 people who have this disk at the tops of their queues. How does the computer pick which one of those people gets it? Is it based on which of their returned movies gets processed first? (e.g., Member A’s return was processed at 7:00, Member B’s at 7:02, Member C’s at 7:04, so they are assigned the specific disks that reenter the stream at 7:01, 7:03, and 7:05 respectively.)

Leaving aside how the system finds my next DVD, once it’s found it, what happens? It addresses an envelope to me, but the only coding on the exterior looks like USPS coding, not Netflix coding. There’s the window in the back through which it can read the bar code, but how do they know that the DVD has gone into the right envelope? Presumably it’s about collating two queues, one of disks and one of envelopes, into each other – but if there’s a hiccup and the queues get out of sync (two envelopes stick together, e.g.), you could fuck up a whole lot of orders before that got caught. Does the USPS address code provide enough info to be specific to a particular member?

At that point, it’s all pretty straightforward – the “we’re shipping X today” email goes out and the physical envelope goes into standard mail sorting machinery that reads the USPS codes, and off it goes, to be delivered to me 48 hours after I’d sent back the first disk.

I’ve read before that the disks are processed by hand. I recall a quote of the founder who said he tried to work it for a day, but he couldn’t keep up with those that did it day in and day out.

No shit, by hand? Just the packing for mailing, or all of it?

The top half of this page answers a lot of your questions:

Thanks, Santo, cool link. (I’d googled “Netflix operation,” which didn’t give any helpful results.)

I still have a few questions, though – the fact that the packing and unpacking is done by people rather than machines doesn’t really affect what the streams of disks are – does everything go back on the shelf? or what?

The automated circulation software libraries use checks to see if an item has been requested as part of the check-in process. The software we use currently performs the following at a check-in: clear item from previous user’s account, scan for check-in notes (staff-added flags for when something happens like disk 2 of a set gets separated from disk 1), check hold list for unfilled requests, if none are found, mark item as a recent check-in.

I can’t imagine Netflix reinventing the wheel on that process, which has worked more or less like that for over 20 years. Letting an item with an outstanding request go back on a shelf only to be pulled later by a separate procedure would be horrifyingly inefficient.

I hadn’t thought about it in library terms, but I’m sure KneadToKnow is correct on their software system. How it works for the worker here in my library - the check in clerk goes through a stack of items, scanning them. It will either make no noise and just check in, taking itself off the previous patron’s record, make a noise and have a check-in note, or make a noise and say “Hold Requested ___ Branch” which means I toss it in the bin for that branch. Seems like that’d be the best thing to do for a business like Netflix, although of course I don’t know that that’s what they do.

Now we’re going to RFID, but that seems too expensive for Netflix’s business model, which seems to work quite well the way it is.

Which is what?

http://en.wikipedia.org/wiki/Rfid

Here’s a good and thorough first-hand account of someone who worked there: http://www.hackingnetflix.com/2007/02/working_for_net.html

It’s not clear how useful that would be. I don’t see how you’d put an RFID transponder on a DVD, and putting it on the sleeve would not be too useful - and also subject it to the tender mercies of the Post Office.