For many people, poor Quality of Experience (QoE) is a fact of life when it comes to video streaming. Videos can take forever to start, suffer from low quality, and repeatedly rebuffer. This also applies to video conference calls and to cloud gaming, which both often suffer from latency issues and playback freezing during crucial moments.
That’s why I was excited to talk about the bold steps we’ve been taking at Compira Labs to remedy the QoE problem for such services. Recently, I spoke with Dror Gill and Mark Donnigan, the hosts of The Video Insiders podcast, which tackles the hottest topics in video streaming, including the latest in video compression, encoding, and beyond.
The podcast featured how Compira Labs is leading the way in QoE and online streaming optimization.
The problem with transmitting video content
The discussion started out by touching on some of the problems with transmitting data, including congestion issues that many networks face, which are often caused by competition from other services and users, as well as failure to properly utilize available bandwidth.
Transmitting data too quickly might overwhelm the network, causing network congestion and, in turn, lost or delayed data, which might manifest in the dreaded rebuffering wheel. On the other hand, not sending data fast enough doesn't utilize the full network capabilities and can’t support HD, and certainly not 4K or 8K.
The main problem for determining how fast data should be injected into the network at each point in time lies in the network being dynamic and unpredictable, and the trend of consuming more and more video content using volatile mobile/5G networks further aggravates the situation. This is where Compira Labs’ software solution comes into play.
The key observation underlying Compira Labs’ solution is that today’s content delivery solutions are one-size-fits-all and cannot be customized to different network environments (e.g., 5G vs. LTE vs. wired) or to different QoE requirements (e.g., VoD vs. live streaming). Not surprisingly, this often results in bad QoE.
Cloud gaming’s QoE issues
Even more challenging than video streaming in terms of QoE is cloud gaming. In many cases, extremely high responsiveness is crucial during gameplay. This responsiveness is hugely affected by network latency issues. Moreover, to support high resolutions, large volumes of data must be delivered, further aggravating the situation.
As I mentioned in the podcast, we now have working versions of our solution in both WebRTC and QUIC, which are implemented on top of UDP and are more appropriate for real-time content delivery.
Using machine learning to improve QoE
There are two parts to Compira Lab’s solution:
Compira Edge: A real-time component installed at the edge (video cache, gaming server) that makes decisions at the granularity of milliseconds regarding how fast data traffic should be injected into the network.
Compira Cloud: A big data analytics engine that customizes, over time, the configuration of the first component to the prevailing network conditions at different locations and to the service-specific needs. By aggregating useful performance-related statistics, Compira Cloud also provides content distributors and service providers with visibility into the quality of their content delivery.
Both components of our solution utilize machine learning for decision making. Compira Edge builds on the rich body of literature on online learning in ML and game theory for real-time decision making, whereas Compira Cloud leverages statistical methods and ML for longer-term big data analytics.
Deploying our solution
One thing that we were able to figure out was that the QoE problems plaguing services like video streaming and cloud gaming can be addressed without having to change anything within the network, i.e., routers, switches, and so on, and without having to touch the receiving end (for instance, video client) at all.
All users have to do is install one (Linux kernel/webRTC/QUIC) software module. Incorporated into the server-side network stack, Compira Labs’ solution allows service providers, operators, and Content Distribution Networks (CDNs) to offer optimized QoE to their subscribers, resulting in higher overall user engagement and satisfaction and in increased revenues.
Compira Labs’ solution is ideal for the last mile
During the podcast we discussed BBR, a congestion control algorithm developed at Google. BBR is much more sophisticated than traditional TCP, but, unlike our solution, makes strong assumptions about the underlying network that, when violated, result in erratic protocol behavior. In addition, BBR, similarly to TCP, is incapable of customizing its rate-selection method to specific networks and service requirements.
We believe that our framework, which combines online-learning real-time rate selection at the edge with longer-term, data-driven customization, is much better suited for the last mile, which is often chaotic and unpredictable, making it hard to explicitly model and rendering one-size-fits-all solutions ineffective. Our experience from large-scale live deployments of our solution at tier-one video streaming service providers corroborate this.
Compira Labs advanced QoE optimizing solutions
Compira Labs specializes in the personalization of the delivery of content through groundbreaking research on applying machine learning to data delivery across the internet.
Our goal is to dramatically improve the user experience across services, networks, and devices.