Streaming Learning Center Where Streaming Professionals Learn to Excel
Related Articles
In their paper entitled, Multi-resolution Encoding for HTTP Adaptive Streaming using VVenC, Kamran Qureshi, Hadi Amirpour, and Christian Timmerer from the Christian Doppler Laboratory ATHENA (Alpen-Adria-Universität, Klagenfurt, Austria) propose an accelerated multi-resolution encoding strategy for HTTP Adaptive Streaming (HAS) using VVenC, an optimized open-source Versatile Video Coding (VVC) encoder.
Their approach, called MEVHAS (Multi-resolution Encoding in VVenC for HTTP Adaptive Streaming), aims to reduce encoding time by using lower-resolution representations as references to optimize encoding for higher resolutions.
Contents
The Technique: Multi-Resolution Encoding for HAS
HTTP Adaptive Streaming (HAS) requires encoding video at multiple bitrates and resolutions (ABR ladder) to adapt to varying network conditions. Traditional approaches encode each resolution independently, leading to redundant computations and increased encoding time.
MEVHAS reuses encoding information from a lower-resolution representation to accelerate encoding for higher resolutions, particularly when using VVenC (see the featured image atop the page). By leveraging partitioning patterns from lower resolutions, MEVHAS reduces redundant computations and improves encoding speed while minimizing compression efficiency loss.
Key Findings
MEVHAS achieves a 17 percent encoding time reduction compared to the medium preset of VVenC. This comes at a cost of a 2.11 percent increase in bitrate, roughly equivalent to a two-point drop in VMAF for the top rung of an encoding ladder. The method provides a better rate-distortion tradeoff than VVenC’s fast preset but still introduces a quality loss compared to the medium preset.
The Fundamental Tradeoff: Encoding Speed vs. Streaming Costs
The core tradeoff MEVHAS presents is a 17 percent faster encoding time at the cost of a 2 percent increase in bitrate for the top rung. While this tradeoff may seem reasonable for live streaming, where encoding speed is crucial, it raises concerns for video-on-demand workflows, where encoding is a one-time cost, but you pay to deliver that additional 2% to each viewer.
Breakeven Analysis: When Does Extra Bandwidth Cost Outweigh Encoding Savings?
We calculated the real-world impact, assuming the top rung is streamed. This is the most conservative scenario, but also a reality for many services.
We assumed a $10/hour cost to encode, which is about what AWS MediaConvert is charging for AV1 now. We assumed encoding the full ABR ladder costs twice as much as encoding just the top-quality stream, comprised of the cost of encoding the lower 2K, 1080p, 720p, 540p, and 360p streams in the encoding ladder.
Any assumption that increases the encoding cost also increases this encoding technique’s benefit and the number of streams needed to break even. Ditto for any assumption that decreases the delivery cost, so if we assumed 720p delivery or a mixed rung delivery, the number of streams to break even would be much higher. Overall, the assumptions are likely conservative, but you can download the spreadsheet and run your own numbers.
Other Assumptions
- Encoding cost per hour (full ladder): $20
- Encoding time savings: 17%
- Encoding cost savings: $3.40 per hour
- Assume that the top rung was encoded at 15 Mbps
- Base data usage: 6.75 GB per hour
- New bitrate (with 2% increase): 15.3 Mbps (6.885 GB per hour)
- Extra data per hour: 0.135 GB
- Bandwidth cost per GB: $0.01
- Extra bandwidth cost per hour watched: $0.00135
Breakeven Calculation
Here’s what the spreadsheet shows. Using the assumptions discussed, it makes more sense to encode using the traditional technique and harvest the 2% bitrate savings after delivering 2,519 hours of video. If your video is viewed fewer times than that, the encoding savings exceeds the additional bandwidth cost.
You can download an xls file with this calculation here: MEVHAS.xlsx (19 downloads )
Key Takeaways
For video-on-demand, the extra streaming costs will quickly surpass the encoding savings. Since 2,519 hours of playback is trivial for even a moderately popular video, encoding at a slower speed for better compression remains the better long-term choice.
For live streaming, MEVHAS could be valuable if it enables real-time encoding. However, the paper does not confirm that encoding is real-time, only that it is 17 percent faster than the medium preset.
Final Verdict
MEVHAS is an interesting optimization for HAS encoding, offering notable speed-ups with minimal compression loss. However, the long-term economics do not favor video-on-demand use cases, where the extra bandwidth cost quickly outweighs encoding savings. For live streaming, MEVHAS could be valuable if it enables real-time encoding, but this remains unproven in the paper.
Further testing against slower presets or evaluations under fixed bitrate constraints would provide a clearer picture of its practical applications.
