The cable industry, especially in North America, has increasingly been threatening to mount a challenge to MNOs that could be emulated round the world. First cable operators invested heavily in public and homespot WiFi to add wireless connectivity to their broadband/video bundles without the cost of licensed spectrum. Now they are looking to add cellular connectivity, either buying some of those licences and/or using shared spectrum schemes for 4G/5G, like the USA’s CBRS and future 5G-Unlicensed.
Initiatives like that of the two largest US cablecos, Comcast and Charter, which have joined forces to develop mobile services, are worrying for MNOs, but the real threat from cable is not in connectivity. Even Comcast and Charter are unlikely to build a national mobile network to vye with those of the MNOs (unless they partner with Dish Network). They are more likely to use a combination of cable (access and small cell backhaul), WiFi and cellular to deploy local, neutral host or private networks in locations where there is real revenue potential. And their differentiator will not just be the multi-layered access and xhaul networks, but more importantly, their control of the local core and the edge computing node.
In the past, the packet core was specific to the MNO, managing traffic, connections, policy and billing as users moved around. But it is evolving, in the 5G era, into a cloud-based, multi-access platform that goes well beyond subscriber management.
- As a virtualized system, it is more easily deployable by a wide variety of organizations on COTS hardware
- As a converged system, it is of interest to wireline as well as wireless operators
- And as it can potentially be deployed on an edge compute node, its capabilities can be integrated with the data processing and applications on that distributed cloud, greatly increasing the range of services that can be supported.
Mobile operators have never had an inbuilt advantage in deploying and managing edge cloud, except that they have many locations in which edge data centers could be sited, plus connectivity. But cablecos also have highly distributed locations and infrastructure, and they are working hard to develop their capabilities in converged core and edge – which in future, could stand them in good stead to move up the stack and take on functions such as orchestration and slicing across many networks and clouds, private or public.
One of the latest initiatives is called Project Adrenaline, and is led by CableLabs – the R&D arm of the US cable industry, which is contributing to so many programs related to 5G and edge, and greatly enhancing the chances for the global cable sector to play a major role in the 5G era. In this project, it is working with Altran, an engineering service provider, on an open source platform that would enable operators to build and manage edge networks, and deploy containerized applications on them, more easily.
Adrenaline is initially focused on cable, but has the broader objective of applying its developments to multiple industries.
“It’s a cable-first initiative, but this is also a general purpose platform,” said Shamik Mishra, VP of research and innovation at Altran; while Randy Levensalor, principal architect with CableLabs’s future infrastructure group, added: “We’re not trying to constrain what use cases there are.”
If localized, virtualized and converged cores are the heart and brains of the private networks we are describing, Adrenaline has its focus in the right place. It aims to accelerate work in the emerging virtualized converged cable access platform (vCCAP) core. But it is also making efforts to virtualize mobile cores too, especially as about half of CableLabs’ members are also MNOs.
The aim of Adrenaline is to simplify the way that new applications are deployed. Mishra feels that there are industry initiatives focused on how compute resources are placed close to end users, and on supporting developer tools and ecosystems, but not how to save those developers the time and money it takes to onboard new apps across many locations, and to inform them about which edges have the performance or latency capabilities needed by the app.
“There’s definitely a need in the market to simplify the deployment, which, I think, has not been a focus too much in the edge computing world,” Mishra told LightReading. “The latency benefit through network placement actually gets lost because the compute itself takes a lot of time.”
CableLabs aims to open source the platform in the first half of this year and then explore the first use cases that can sit on top of it, probably starting with virtual CCAP and a virtualized mobile core. Later in the year, the group will produce reference architectures for access and core networks, and some early trials. Throughout, there will be alignment with other open source projects, especially the Kubernetes container management community.
Project Adrenaline evolved from of CableLabs’s recent work focused on hardware accelerators, including field programmable gate arrays (FPGAs) and graphics processing units (GPUs), to enhance cloud infrastructure to support highly demanding applications. This is another gap in the edge market, argues the R&D organization, saying that frameworks like OpenCL allow apps to be written once to run across multiple hardware types including CPU, GPU and FPGA, but this comes with a performance trade-off that could negate the benefits of running workloads at the edge.
Another CableLabs development which may prove to be increasingly important in the edge market is low latency DOCSIS, final specs for which were recently released last June as an annex to the DOCSIS 3.1 standards. These support round trip response rates of one millisecond. Although low-latency DOCSIS is initially targeted at the cablecos’ HFC (hybrid fiber-coax) networks and particularly at video streaming, CableLabs stressed that it will also be applicable to many access technologies, including FTTP (fiber-to-the-premise) and wireless, and to many applications including those in the Internet of Things (IoT), which are often a trigger for edge and private network projects.
In fact, this goes for most of the technologies included in the cable industry’s ‘10G’ programme (see inset), led by CableLabs, Cable Europe and Internet & Television Association, which sets out the roadmap for next generation networks, starting from 2020. It seeks to support speeds at least 10 times faster than current top end cable networks, as well as many other new capabilities – ambitions which CableLabs is clear will require multiple technologies used in unison.
The low latency DOCSIS (LLD) standard will be supported by existing DOCSIS 3.1 equipment with a software upgrade and is particularly targeting virtual reality gaming, also an application beloved of 5G/edge supporters. Indeed, one European operator recently confided that it had separate departments developing business cases for 5G and for FTTP, which found themselves competing over the same target revenue stream, VR gaming.
One US cableco, Cox Communications is already testing a low latency gaming service for PCs in Arizona that costs an extra $14.99 per month, and 5G operators are considering similar propositions to boost wireless revenue (though in mobile environments, AR will be more relevant). Many of the early demonstrations and applications developed for 5G edge relate to AR – not just gaming, but for industrial uses such as training or remote factory support, which would often be running on a private network with low latency requirements.
LLD grew out of discussions about the next technologies and user experiences after ultra-high definition video. Examples such as holographic telepresence, the successor to videoconferencing, would demand ultra-low latency as well as colossal bandwidth. The drive for low latency streaming is coming from two categories of application – one-way live video and two-way interactive applications. While the former requires lower latency to bring streaming into line with broadcasting, it is not as demanding as interactive applications and is being met with protocols such as SRT (Secure Reliable Transport) and RIST (Reliable Internet Stream Transport), which strike a balance between delay and error correction for video quality.
However, videoconferencing, gaming, and a number of professional applications such as telesurgery, require sub-second latency as far as possible, comparable to the demands of traditional voice telephony. Some applications are even less latency-tolerant, notably in-sports betting, where every millisecond lost can be expensive for punters as odds on results fluctuate rapidly in line with unfolding events on the field or court of play. For some gaming applications, delays over 20ms can have impacts on the experience, including motion sickness when the eye and ear experience slight offsets between signals that are supposed to be totally synchronized.
CableLabs decided it needed a purpose-built protocol suitable for both live video and gaming, that would slot naturally into the DOCSIS infrastructure without any significant reconfiguration. The foundation had already been laid in the latest DOCSIS 3.1 with its support for Active Queue Management (AQM), which typically achieved latencies around 10ms over the access network link, although with occasional peaks of up to 100ms or a little more under heavy load.
AQM avoids the previously normal practice of waiting for a buffer queue to be full before dropping IP packets by necessity, which can impair video quality and also impose latency while that process is taking place. Instead, AQM drops packets before the queue is full on a statistical basis, minimizing the impact on video. It also informs the source end point when congestion is occurring before a queue is full, so that intelligent action can be taken, such as holding packets up temporarily, or again deciding which ones to drop.
Since the access link is just one component of the end-to-end latency path, CableLabs deemed AQM’s performance was not good enough and was anxious to shave that contribution down to 1ms. It also wanted to make this latency contribution almost invariant under heavy loading, eliminating those spikes.
Hence LLD, whose underlying principle is that data can be split into two categories, that which is tolerant of delay and that which is not, allocating each type to separate queues. This was decided after considering that the HFC (Hybrid Fiber Coax) access network contributes to latency in five ways:
- the (slight) delay from switching and forwarding IP packets
- propagation delay – the time taken for a signal to travel across the HFC plant
- delay associated with serialization and encoding
- media acquisition, which is peculiar to DOCSIS, resulting from the shared-medium scheduling still supported until eventually fading away as fiber reaches the premise
- queuing delay associated with TCP itself and its low latency variants, which is connected again to the bottleneck imposed by the DOCSIS access infrastructure
The main contribution of the new standard is on the queuing front. This is achieved through this dual-queue mechanism, which cuts around 20ms from latency as well as slashing jitter, that is latency variation. It does that first by taking account of the fact that most applications are somewhat tolerant of delay and yet tend to build queues by sending data faster than the slowest link can tolerate through their legacy and rather selfish algorithms designed to avoid congestion for themselves. The bulk of traffic, including downloads, comes under this queue-building heading.
The other traffic category comprises interactive apps where low latency is critical but which usually generate far less traffic, including conferencing apps and gaming. These applications tend to send at a relatively low data rate with packets corresponding well-spaced if the bit rate is high, which means that queues do not accumulate under their own steam.
But such less selfish apps suffer from delays imposed by the queue-building ones, which LLD neatly sidesteps by building separate queues for each.
A key point which CableLabs likes to emphasize is that this is not a prioritization mechanism and would not fall foul of net neutrality principles, because neither queue is held up for the other. They operate in parallel, each optimized for the characteristics of the applications they support, one data intensive but somewhat tolerant of delay, the other at a lower data rate but requiring ultra-low latency. The total amount of capacity required is unchanged.
The cable industry’s 10G roadmap will be influential at the multi-access edge:
The 10G technologies include:
- 25G-PON and 50G-PON (passive optical network fiber standards)
- FDX (full duplex)
- Point-to-point coherent optics
- DOCSIS 3.1, the latest release of the cable standard, along with FDX DOCSIS and low latency DOCSIS
- Micronets
- Various WiFi technologies, namely WiFi Easy Mesh, Passpoint, Vantage and WiFi PNM (proactive network maintenance)
- Low latency mobile backhaul.