We are streaming more and higher quality video. Overall, time spent in viewing streaming video has increased rapidly to between 60 and 100 minutes per person per day in 2019. Also, video quality is going up as we are increasingly streaming HD and 4K content. Hence, bandwidth is an ever-increasing factor to consider for online video services with the ambition to scale to mass audiences. With the COVID-19 outbreak, video providers, including Youtube, Netflix, Amazon Prime, and Disney+, have even reduced streaming quality to avoid internet congestion.
HESP drastically changes video delivery and reduces bandwidth consumption by up to 20%
Existing HTTP Adaptive Streaming (HAS) protocols use a segment-based approach. This means that a video is cut up into segments of a few seconds each. This approach leads to high zapping and switching times, causes bandwidth overhead and introduces increased latency. Instead of using a segment-based approach, HESP leverages a frame-based streaming approach, which allows for a much better viewer experience and a reduction in bandwidth consumption up to 20%. The figure below compares HESP to CMAF-CTE, LL-HLS, HLS, DASH, WebRTC and RTMP.
Bandwidth test results: HESP outperforms CMAF-CTE
In order to measure gains achieved by trading latency for bandwidth, we designed a test setup where the encoding parameters were set to achieve a 2.3s end to end latency for HESP, aligned with the lowest latency measured for DASH CMAF-CTE. More details on the technical setup can be found in the HESP whitepaper.
The test assets which were looped within the live channels were three videos from the Xiph collection:
- Big Buck Bunny: The slow moving movie we all love.
- Elephants Dream: Relatively fast moving movie asset.
- Meridian: The Netflix test asset designed to test encoders.
When comparing bandwidth used after 10 minutes, we found the following results. The percentage in bandwidth saved, can of course map one-to-one with cost reduction for example the CDN egress. Depending on the content being broadcasted (ideal GOP-length), the bandwidth saving can become quite significant, averaging between 10-15%.