Peer assisted content delivery has been around well over a decade but is only now coming of age for video distribution in the streaming era, because of its potential to improve core bandwidth efficiency by up to 10 times. This can help negotiate peaks in demand during major live events and also allow streaming services to be deployed over bandwidth-constrained networks, such as 3G or even 4G cellular infrastructures, as well over aging DSL or cable services.
Until recently, online video has mostly been delivered via the traditional client/server architecture on a one-to-one unicast basis where clients download content from dedicated servers managed centrally, often via a CDN and then onto an access network operated by an ISP, either a telco, cable operator or cableco. Over managed IPTV networks, multicast may be implemented but that is still a centralized approach where content radiates out over a tree and branch topology with links being pruned when not needed.
Meanwhile, the alternative peer-to-peer approach had become popular through services such as BitTorrent for P2P file sharing, distributing downloaded media files, but without support for the QoS mechanisms essential for streaming video at acceptable quality over unmanaged and unpredictable networks. But more recently, a host of start-ups have emerged to add the required QoS mechanisms, typically using the WebRTC protocol as that is inherently P2P in its operation and is also optimized for ultra-low latency. At the same time, hybrid, or “peer-assisted”, CDNs have evolved, combining both approaches, which we anticipate will be a major direction of travel not just for startups but also mainstream CDN providers like Akamai for video delivery. That is because the bandwidth savings enabled by such peer assistance in the face of relentless growth in the volume of video traffic, especially over mobile networks and in developing markets, will be crucial.
We have covered P2P networking before and noted in December 2019 how the approach received one of its biggest endorsements so far from Liberty Global. The operator had deployed decentralized P2P software from start-up Teltoo, a graduate from the Liberty Global-funded Techstars program, to optimize its existing CDN infrastructure from Akamai and Velocix. Liberty Global had been swayed by the promise that the technology could save as much as 85% of CDN usage, a figure we were skeptical could be achieved consistently on a large scale in major deployments.
Yet similar or even bolder claims of the technology’s ability to prune bandwidth have been made both by CDN provider CenturyLink, and also the UK’s BBC following tests of activity around its iPlayer portal. Admittedly, CenturyLink’s boast followed its acquisition of P2P steering specialist Streamroot, which it had just acquired and whose technology it wanted to promote. But we have no reason to doubt the claims and our concerns now focus more on quality of experience and logistics, given that P2P streaming, or peer assisted steering, however we call it, relies on resources in client devices such as smartphones that are often running short of memory on account of all the apps and user data on board. The other concern is over ability of WebRTC to sustain quality, as that is widely used in P2P steering implementations, including Streamroot’s DNA and mesh products.
These products are worth mentioning because they were among the earlier deployments and were successfully applied at scale by several major operators before its acquisition by CenturyLink, including Discovery’s Eurosport in a relationship dating back to 2015. The use of Streamroot Mesh for Eurosport’s streaming of the 2017 Australian Open tennis championships was one of the first demonstrations of P2P steering on a large scale. The operator says it delivered about half the streaming traffic via its Eurosport Player app using P2P steering during the final match of that tournament, rising to nearly 60% of total throughput at the peak of viewing. No figures were given then on bandwidth saving but CenturyLink is now claiming levels of 88%, reducing it to 12% of the original value.
At least Streamroot avoids need for a client plug-in, since by utilizing WebRTC it operates directly in the browser. With such products though care has to be taken with the implementation to ensure that it actually does deliver benefits on the required scale. Like all P2P systems it requires a critical mass of peers, below which it defaults back to client/server operation. The size of that critical mass will depend on various factors, primarily geographical dispersion, for at a low level of concentration it would still not work even if there were say thousands of peers in total. On the other hand if peers are clustered locally, as few as about 30 or 40 simultaneous viewers might be enough for the P2P steering to kick in for live streaming, or perhaps 100 for VoD where the benefits are not quite so overwhelming.
What Streamroot is less willing to admit is that the technology is still work in progress and there are considerable challenges to be overcome yet before it will become ubiquitous. One key challenge, still the subject of research papers and PhD theses, concerns storage of content on user devices for P2P forwarding, particular in mobile environments. The first point is that clients have to be selfish and only download data they need themselves. The peer may then cache the associated packets on the basis of various algorithms, which could include some form of prioritization, or just on the basis of how much storage is available. Or packets for caching could be selected randomly in the hope that on a statistical basis, given a sufficient number of peers, at least one nearby client would have the required data available.
Then a major stumbling block is that while P2P content distribution can be advantageous to content providers, CDN operators and end users in cost and performance, it often has adverse impact on ISPs on both these counts. In essence it shifts burden of capacity provision from the content providers and CDNs to the ISPs and this has certainly been shown to be the case with BitTorrent. However, this problem can be alleviated by deployment of intelligent “locality-aware” algorithms, designed to keep traffic shuffling around the network edge and avoid bogging down the ISP access infrastructure as much.
ISPs can then benefit from the self-scalability of P2P networks, resulting from the fact that the more peers there are the more capacity is available. This is precisely what has appealed to CDNs and content providers, which can invoke P2P delivery to cater for flash data flooding during the course of major streaming events like blockbuster concerts, world cup finals or top bill boxing matches.
Given perfect implementation, caching would benefit ISPs just as much as users and other parties, since just one stream would be distributed, or file downloaded, irrespective of how many requests were submitted. In practice, P2P systems will not be optimal but should at least work most efficiently at their time of greatest need at peak load.
The other big question is over video quality, given that the underlying WebRTC protocol was optimized for ultra-low latency above error correction, designed for interactive applications such as telepresence. In WebRTC, each frame is assigned a timestamp and each media stream, including voice and video, is sampled at different clock rates independently. The receiver device must synchronize these streams for playback, using RTP Control Protocol (RTCP) Sender Reports (SR), which map that stream timestamp to a global Network Time Protocol (NTP) timestamp. If this is done incorrectly, lack of synchronization between video and audio results, that being one common issue with WebRTC.
The other problem is simply loss of quality through dropped IP packets, given that WebRTC has less rigorous resending mechanisms than protocols designed specifically for video streaming, such as RIST (Reliable Internet Stream Protocol) and SRT (Secure Reliable Transport). Catering for this relies on robust implementation and it is true that P2P operation provides scope for resiliency here. The likes of Streamroot claim they have this sorted and it will be interesting to see how this plays out as CenturyLink pushes it hard beyond its French roots on the back of early big name deployments there, which also include TF1 and Canal+.