The folk at the Netflix Technology Blog are growing gradually more susceptible to drip-feeding streaming infrastructure insights to thirsty M&E industry observers. The latest blogpost on VBR (Variable Bitrate) has caused quite a stir, even though the technologies and techniques behind VBR—multi-pass and single-pass encoding—are far from novel.
Netflix has revealed that at the beginning of this year, all live events on the streaming platform were encoded using VBR instead of the standard CBR (Constant Bitrate). This technique allows encoders to increase bitrate for complex scenes and reduce bitrate for simple scenes—trimming off bits deemed surplus to perceived visual quality.
Efficiency gains via Open Connect:
Through the shift to VBR for live streaming at scale, the most important bottom line for Netflix is the reduction in traffic needed to fill its Open Connect CDN servers, thanks to the reduction in average number of bytes needed to deliver a full event.
This achieves greater network efficiency, saving Netflix bandwidth costs, while also improving quality of experience (QoE). Netflix says its VBR-based livestreams since January have achieved 5% fewer rebuffer events per hour, while transferring around 15% fewer bytes on average, with a 10% reduction in traffic at the peak minute.
However, when Netflix first applied VBR using the existing bitrate ladder (producing a set of streams at different resolutions and nominal bitrates), A/B tests produced a lower perceptual quality score for VBR compared to CBR, using Netflix’s proprietary VMAF metric (Video Multi-Method Assessment Fusion).
To fix the VMAF gap, Netflix increased the nominal bitrate for VBR just enough to close the difference, whenever VBR fell more than one VMAF point below CBR.
This is much harder to achieve with public CDNs like Akamai and Cloudflare, because VBR spikes can inflate costs disproportionately, and increase risk of congestion on shared infrastructure. Public CDNs are effectively incentivized to smooth or constrain traffic variability due to their fundamental billing models, which is why some streaming services favor capped VBR approaches.
Netflix’s per-title encoding has been using VBR since 2015, following initial discussions around Content Aware Encoding (CAE) back in 2010. The application of VBR in live, over a decade later, represents the final frontier, with live workflows historically struggling with VBR due to unpredictability and the risk of bitrate spikes breaking adaptive bitrate (ABR) assumptions and overloading CDNs.
AWS Elemental MediaLive is used in parts of Netflix’s live pipeline, where a capped version of VBR imposes a hard ceiling so bitrate never exceeds a defined maximum.
For example, VBR uses a low bitrate for a basic “waiting room” scene before an event, which is visually uncomplex and easy to compress. Once the live event starts, a more complex shot filled with fans and confetti means VBR uses more bits to maintain quality (as pictured).
Origins:
The origins of VBR can be traced back to research conducted in the late 1940s when work by Claude Shannon established that data could be encoded more efficiently depending on its entropy.
It would be decades later, in the early-to-mid 1980s, before the actual term VBR began appearing in technical literature, followed by the first commercial deployments in mobile and satellite communications in the late 1980s and early 1990s for bandwidth optimization.
MPEG-1, established in 1993, defined both VBR and CBR encoding modes, which spurred the development of early encoders using VBR to improve audio quality at lower bitrates.
The advent of H.264/AVC in the early 2000s changed the game, followed by the emergence of HTTP Live Streaming (HLS) in 2009 and MPEG-DASH in 2012. The maturation of streaming effectively pushed VBR deeper into encoding techniques used across the industry.
When Netflix aggressively expanded VBR for on-demand content in 2015, there were concerns that content would drop to a lower ABR rung (e.g. from 1080p to 720p) if connection speeds were struggling. Part of the confusion with the introduction of 4K content was that more efficient encoding made 1080p content look closer to 4K quality.
With live streaming, the behavior is different. Poor bandwidth does trigger ABR downshifts, but VBR makes streams more efficient at any given rung. When the network, not the encoder, becomes the limiting factor, the VBR workflow is heavily capped—sometimes so tightly that it behaves almost like CBR.
Why now?
With Netflix recently laying groundwork for streaming live 4K HDR encoded in AV1 at scale, it makes sense that the next stage of savings comes from how bits are allocated. Netflix has already squeezed significant gains from codec transitions (H.264 to HEVC to AV1).
It also reflects how Open Connect has matured to the point where it can tolerate higher average bitrate swings, allowing Netflix to take advantage of VBR for live streaming while public CDNs lag behind.