Slashdot Mirror
Netflix's Secrets to Success: Six Cell Towers, Dubbing and More (variety.com)
Variety gets access to the people at Netflix who take care of the tech: Netflix has its own cell towers. Netflix wants to test its app running on mobile devices under a variety of conditions available around the world, so the company decided to bring the operating equipment of six cell towers to its Los Gatos offices. "Minus the towers," quipped Scott Ryder, the company's director of mobile streaming. The cell tower equipment is housed in the company's mobile device lab, where they are joined by a number of cabinets that look like fancy Netflix-themed fridges, but in reality are Faraday cage-like boxes to suppress any outside interference, and also make sure that those experimental cell towers don't mess up phone reception on the rest of the campus. Each of these boxes can house dozens of devices, and emulate certain mobile or Wi-Fi conditions. "We can make a box look like India, we can make a box look like the Netherlands," Ryder said. Altogether, Netflix runs over 125,000 tests in its mobile lab every single day.[...]
Netflix just re-encoded its entire catalog, again. To optimize videos for mobile viewing, Netflix recently re-encoded its entire catalog on a per-scene basis. "We segment the videos into shots, we analyze the video per shot," said the company's director of video algorithms Anne Aaron. Now, an action scene in a show may stream at a higher bit rate than a scene featuring a slow monologue -- and users with limited bandwidth are set to save a lot of data. A few years back, 4 GB of mobile data would get you just about 10 hours of Netflix video, said Aaron. Now, members can watch up to 26 hours while consuming the same amount of data. Netflix previously re-encoded its entire catalog on a per-title basis, which already allowed it to stream animated shows at much lower bitrates than action movies with a lot of visual complexity. The next step for the company will be to adopt AV1, an advanced video codec developed by an alliance of companies that also includes Apple, Amazon, and Google. Aaron said Netflix could start streaming in AV1 before the end of this year, with Chrome browsers likely being first in line to receive AV1 streams.
Netflix just re-encoded its entire catalog, again. To optimize videos for mobile viewing, Netflix recently re-encoded its entire catalog on a per-scene basis. "We segment the videos into shots, we analyze the video per shot," said the company's director of video algorithms Anne Aaron. Now, an action scene in a show may stream at a higher bit rate than a scene featuring a slow monologue -- and users with limited bandwidth are set to save a lot of data. A few years back, 4 GB of mobile data would get you just about 10 hours of Netflix video, said Aaron. Now, members can watch up to 26 hours while consuming the same amount of data. Netflix previously re-encoded its entire catalog on a per-title basis, which already allowed it to stream animated shows at much lower bitrates than action movies with a lot of visual complexity. The next step for the company will be to adopt AV1, an advanced video codec developed by an alliance of companies that also includes Apple, Amazon, and Google. Aaron said Netflix could start streaming in AV1 before the end of this year, with Chrome browsers likely being first in line to receive AV1 streams.
So they have cell towers, but they don't have cell towers. Nice.
I predict around 30%.
You assume Apple and Google will allow software decoding of AV1, which is extremely bad for battery life.
#DeleteFacebook
I do assume that, but I am a moran.
How many people will actually be using AV1?
Everyone eventually. AV1 will be the codec of choice for all web video. It outperforms the other options and doesn't have the licensing hassles of H.264 or the licensing mess of H.265. The licensing of H.265 is so bad that even the founder and chairman of MPEG, Leonardo Chiariglione, thinks MPEG probably doesn't have a future.
People will encode to H.264 for legacy devices, VP9 for current devices, and encode to AV1 for current desktops and future devices.
The problem is that only high-end devices are just getting H.265 support and even then it's pretty spotty as to actual implementation and performance. If the industry keeps coming out with new codecs at this rate, we are going to have to adapt our chips to be more flexible, perhaps even programmable in the hardware decoder area which brings with it a whole slew of issues, both physical (heat, size and cost) as well as in software (does every piece of software get to upload its own decoders, if so, how, what if we need more or different versions).
H.264 and VP8 is easy to find these days, VP9 and H265 is slowly but surely coming, releasing a brand new codec today will take 5 years to get it in the majority of high-end chip fabs and another 5-or-so years to go mainstream with at least 15-20y more years of having to have both available.
Custom electronics and digital signage for your business: www.evcircuits.com
The H265 rollout has been slow because its licensing is such a disaster. Why would you be in a rush to ship H265 when you have no idea how much you will have to pay patent holders for the units you are shipping?
AV1 does not have this problem. It's a no-brainer to implement and ship it ASAP. Within two years all new chips will have it.
Soon it will become clear that H265 will see very little usage outside broadcasting. The H265 patent pools will drop any pretense of encouraging broad H265 adoption and focus on extracting maximum revenue from those vendors foolish enough to have shipped H265 in advance of a clear licensing story. As soon as the inevitability aura around H265 dissipates, the non-TV vendors will drop it like a hot potato.
I've heard a mixed bag coming out of Netflix re: developer experience, but one thing I admire is their effort toward a reliable user experience.
From testing everything down to minutiae, to designing things so that failure is simply another regular and expected state to move forward from... most companies do not commit the time/funds to do this sort of thing.
This practical engineering is much cooler to me than Facebook/Google's latest me-too Javascript libraries that iteratively steal the next good established idea from desktop coding and call it innovation.
"We segment the videos into shots, we analyze the video per shot," said the company's director of video algorithms Anne Aaron. Now, an action scene in a show may stream at a higher bit rate than a scene featuring a slow monologue
Video encoders have supported constant quality modes for quite a while that already do this very effectively. I'm guessing they don't do this out of some need for precise control or for hardware compatibility. It's obviously not a wasted effort, but it's unfortunate that Netflix is needing to reinvent the wheel here.
they don't need to care if apple or google "allow" it on mobiles where they deploy their own app and can use whatever sw or hw codecs they want if they want...
what they would need to worry would be if there's enough power to do it and about the user experience. ...
and the bit about the mobile towers? pssh. standard thing if you have the money for it and can't think of some other way to simulate a crappy network with high ping times, high ploss but sometimes high bandwidth.
world was created 5 seconds before this post as it is.
I don't have the "approved by Netflix" devices
Here are the resolutions Netflix delivers various viewer software (seems to vary depending on the DRM schemes available in different browsers and platforms) .
And here's a Firefox add-on to get 1080p Netflix video in Firefox.
Qualcomm includes VP9 support. They'll include AV1 support as well.
Apple don't seem to be a member
To quote from the page you linked (emphasis added): "The Alliance for Open Media is governed by founding member companies: Amazon, Apple, ARM, Cisco, Facebook, Google, IBM, Intel, Microsoft, Mozilla, Netflix and NVIDIA."
The logo isn't on the page but Apple has joined the Alliance for Open Media at the highest membership level which is "founding member" (even though they weren't a member from the start). The membership terminology is poor.