Periodic Reporting for period 1 - FAST-STREAM (Solving the ‘last-mile delivery challenge’ for quality Over-The-Top (OTT) streaming content)
Période du rapport: 2022-05-01 au 2023-01-31
In recent years, Over-The-Top (OTT) streaming has become increasingly popular for a wide range of applications. During the coronavirus pandemic, the need for social distancing led to increasing numbers of people who work and study remotely. Despite massive investments in network infrastructure, which includes newly deployed 5G and fiber networks, severe problems in service quality persist. These include frequent re-buffering and low resolution for video streaming, to long "lags behind live" for real-time services such as video conferencing.
The root cause of these problems is the "the last mile network". To reach its destination, content must often traverse volatile network segments (such as mobile and Wi-Fi networks) and must compete with traffic from other services over scarce network bandwidth. Bad user experience results in customer churn, decreased user engagement, and loss of income for service providers. It also harms the ability to effectively work and study from home and has a negative impact on the economy.
Compira Labs' solution is utilizing Performance-oriented Congestion Control (PCC) algorithmic framework, data analysis and machine learning methodologies. While it requires no changes to the network or applications, it offers better utilization of the existing networks thus reducing the digital divide by providing better access to Internet services to wider communities.
The goals of this project are focused on scaling up our technical solution to support successful large-scale field trials for a diverse set of customer use cases, followed by successful qualification and validation of our technology under operational conditions.
The root cause of these problems is the "the last mile network". To reach its destination, content must often traverse volatile network segments (such as mobile and Wi-Fi networks) and must compete with traffic from other services over scarce network bandwidth. Bad user experience results in customer churn, decreased user engagement, and loss of income for service providers. It also harms the ability to effectively work and study from home and has a negative impact on the economy.
Compira Labs' solution is utilizing Performance-oriented Congestion Control (PCC) algorithmic framework, data analysis and machine learning methodologies. While it requires no changes to the network or applications, it offers better utilization of the existing networks thus reducing the digital divide by providing better access to Internet services to wider communities.
The goals of this project are focused on scaling up our technical solution to support successful large-scale field trials for a diverse set of customer use cases, followed by successful qualification and validation of our technology under operational conditions.
To date two major technological milestones have been completed:
We adapted our technology to 3 different commonly used protocol stacks and successfully benchmarked the performance of the integrated solution against the state-of-the-art alternatives.
For HTTP/TCP, the prominent service delivery method today, we packaged our software as a loadable Linux kernel module. For HTTP3/QUIC and for WebRTC we have integrated our PCC logic with open-source software stacks.
We devised methodologies for inferring video-related Quality of Experience scores, namely, average video bitrate, and video start time, from network-level statistics (e.g. throughput, packet loss rate, packet delays, etc.).
We modeled this challenge as a supervised learning task and showed how statistical decision models can be utilized to accurately predict the quality score.
We adapted our technology to 3 different commonly used protocol stacks and successfully benchmarked the performance of the integrated solution against the state-of-the-art alternatives.
For HTTP/TCP, the prominent service delivery method today, we packaged our software as a loadable Linux kernel module. For HTTP3/QUIC and for WebRTC we have integrated our PCC logic with open-source software stacks.
We devised methodologies for inferring video-related Quality of Experience scores, namely, average video bitrate, and video start time, from network-level statistics (e.g. throughput, packet loss rate, packet delays, etc.).
We modeled this challenge as a supervised learning task and showed how statistical decision models can be utilized to accurately predict the quality score.
Our solution for HTTP/TCP was tested in the field with many thousands of live customers and was able to reduce rebuffering metrics by 30% as well as reduce the number of low-quality video sessions by more than 30%.
In the lab, our solution achieved better user experience metrics than the state-of-the-art alternatives for both HTTP3/QUIC and webRTC implementations.
In the lab, our solution achieved better user experience metrics than the state-of-the-art alternatives for both HTTP3/QUIC and webRTC implementations.