When it Comes to Bitrates, Less is Always More
Most streaming professionals know that bandwidth usage has spiked as millions of workers are furloughed or working from home due to COVID-19. What can streaming producers do in the short term to reduce the load?
Netflix jumped in first by removing the highest-bandwidth streams from its manifest files, so even if you had the bandwidth to retrieve that stream, your player wouldn’t know it was there. This is an option that all streaming producers should consider because it’s easy to implement and, unlike more drastic changes I will discuss later, has very few testing implications.
Beyond this, producers should consider pre-processing, some form of per-title encoding, and deploying a new codec, in that order. By pre-processing, I mean using technologies like ZPEG or BitSave that pre-process your video before transcoding. Your encoding and packaging pipeline should remain the same, but you should produce higher-quality files at lower bitrates with minimal compatibility risk during playback.
Next to consider is some form of per-title encoding, if you haven’t done this already. (See go2sm.com/scene for a review of the evolution of per-title technologies.) Per-title takes many forms, such as Capped CRF you can implement yourself, Beamr’s frame-by-frame optimization, and features in cloud-based services like Bitmovin and Mux or products/services like AWS Elemental’s Quality-Defined Variable Bitrate (QVBR) encoding, Harmonic’s EyeQ, and Capella Systems’ Source Adaptive Bitrate Ladder. Note that AWS Elemental and Harmonic offer per-title for both video on demand and live.
Netflix dropped the top bitrate of its encoding ladder with minimal perceptible quality loss because it uses per-scene adaptation, an advanced form of per-title encoding. Whereas a fixed ladder might always deploy a 1080p stream at the top quality followed by 720p, the most advanced forms of per-title might deploy two 1080p streams for some clips so the viewer still gets a 1080p stream if the top quality stream is dropped.
How much benefit you’ll get from per-title depends on the type of footage you stream and the data rates deployed in your current ladder. If you’re streaming soccer matches at a maximum data rate of 4.5Mbps, per-title encoding won’t deliver much savings. However, if you stream a range of content and your top quality 1080p stream is 8Mbps, you should see significant bandwidth reductions on many of your streams.
Compared to your next option, changing codecs, per-title is fairly low risk because you’re still working with H.264. It’s also cheaper since you’re still only producing and distributing a single codec. Adding any codec to the mix increases encoding cost and storage cost, decreases cache efficiency, and involves a development cost to ensure playback compatibility.
Bitrate savings depend on the codec and codec implementation. According to Moscow State University’s Nov. 1, 2019, “Video Codecs Comparison,” savings for HEVC and VP9 averaged around 40% compared to H.264, with the proprietary WZAurora AV1 Encoder delivering about 20% more efficiency than the highest-performing HEVC codec. This is roughly similar to the 20% savings AV1 delivered over VP9 that Netflix announced earlier this year. However, encoding AV1 is still glacial, which translates to much higher encoding costs. Whereas AWS Elemental charges $0.024 per minute of 1080p30 H.264 encoding and $0.048 per minute of HEVC, the price skyrockets to $0.864 per minute of AV1 encoding, a premium of 36x over H.264. This likely makes AV1 impractical for all but the highest-volume producers, like Netflix, YouTube, Hulu, Facebook, and Amazon.
Where should these codecs play? HEVC should play seamlessly on most iOS devices with minimal muss or fuss, plus OTT devices and smart TVs. VP9 should play in Chrome and Firefox on all platforms, plus Android and most newer OTT devices and smart TVs.
What about the 2020 crop of MPEG codecs—VVC, EVC, and LCEVC? The first two will require hardware support for playback and therefore won’t really become relevant until 2022 or so. LCEVC is backward-compatible with existing H.264 and HEVC hardware playback and is implemented now by V-Nova as P+. In its submission to MPEG, V-Nova claimed that LCEVC delivered the same quality as H.264 at a bandwidth savings of 45%, making it a technology that could have an impact in 2020.
No one knows how long the COVID-19-related usage spikes will continue, but the situation presents a stark reminder of what we all knew all along: When it comes to streaming video, less is always more. At the same or better visual quality, of course.
Streaming Learning Center's Jan Ozer makes the case for custom, per-title encoding on the Netflix model, focusing on bitrate issues in part 1 of this 2-part series from his Streaming Media West presentation.
Akamai's Will Law, Disney's Mark Arana, and Yahoo's Omer Luzzatti discuss the ongoing need for ABR streaming in the fast-changing streaming landscape.
Companies and Suppliers Mentioned