W3C Web5G Workshop Report
Executive Summary
On May 10 and 11 2018, W3C organized a workshop hosted by GSMA. Network Operators, vendors, application developers, content provider and standard makers participated in this event which was designed to explore how the Open Web Platform could help drive the adoption of 5G innovations from the applications layer to the network level.
During the two days, participants reviewed opportunities that new emerging innovations and capabilities at the application layers can bring to the 5G network.
The workshop concluded with the proposed creation of a task force of participants to explore how the 5G and Web communities might work in a productive and cohesive manner.
In particular, there was wide agreement on the benefit of developing compelling business and technical reasons and objectives to incentivize and drive a close collaboration among the W3C, 5G standard organizations (e.g. 3GPP), browser vendors, developers, equipment vendors and network operators. The goal is to create an environment conducive to the development and deployment of technologies that are supported by all the stakeholders in the ecosystem.
Introduction
Fifth generation networks (5G) are intended to provide higher bandwidth, lower latency, denser coverage and more network-exposed services and control. The investments needed to make this vision a reality in many ways, rely on the of capabilities of web technologies and software. In particular, they need to be exposed to and adopted by application and service developers before end users (consumers and business alike) can appreciate their benefits.
The Web5G Workshop, which took place on May 10 and 11 2018 in London, hosted by GSMA, aimed at starting the conversation among the various stakeholders involved in the deployment of new network technologies and the advancement of new application fields.
5G as an Application Enabler
As Dan Warren (Samsung) presented, the requirements for 5G networks were set by the Next Generation Mobile Networks Alliance based on their analysis of new needs emerging from B2B use cases.
These requirements are expected to find implementation by a combination of:
- increased spectrum with frequencies providing greater throughput,
- Inclusion of more antennas and correlation of their signals,
- Increased flexibility brought by the lower reliance on custom hardware (softwarization) and the scalability of deploying new software ("cloudification") in the operation of networks.
- The actual implementation and deployment of these improvements are expected to appear progressively over multiple years, and in parallel with the continued operations of older generation networks for the foreseeable future. The actual capabilities of the networks will also vary depending on location - some of the frequencies with the higher throughput are for instance known to behave badly indoors, or through leafage.
The investments needed for the deployment of these capabilities will be driven on the one hand by the seemingly unlimited appetite of consumers for more bandwidth (driven in particular by videos of increasing quality, and other bandwidth-hungry applications such as virtual reality), and on the other hand by B2B use cases (network as a service, IoT, industry automation, professional usage of virtual and augmented reality, e-health) which mobile operators hope will open new markets opportunities.
A key to the latter relies thus on the ability of mobile operators to expose these new network capabilities to make them easily usable and adaptable to a wide variety of service developers.
3GPP, where the new 5G architecture is being standardized, is developing a framework for REST APIs that developers are expected to integrate when interacting with 5G-exposed services (so called Northbound API Framework). ETSI has already started defining APIs to expose the capabilities of Multi-Access Edge Computing (MEC).
A number of participants evoked the various previous attempts made in the telecommunication industry at building and exposing network-based capabilities (OMTP, WAC, OneAPI, Parlay), many of which struggled to gain wide adoption due to the combination of the difficulty for developers to gain uniform access to these features across operators as well as to the different pace and culture of API deployments across the Web and telco worlds.
Workshop participants thus noted that the aforementioned new efforts would benefit from more direct participation and input from the application layer, and identified this as a task that W3C could help with. This should involve not only technical considerations (API design, integration points in the platform), but also and primarily business considerations that would enable all stakeholders to find common interest in exposing these capabilities.
Experimenting with new Web/network integration in private networks
A salient feature of the 5G architecture is the ability to seamlessly create network slices which can be approximated as on-demand virtual private networks (VPN) with well-defined characteristics in terms of e.g. bandwidth or latency, including over the radio network interface.
These network slices are expected to play a key role in how certain applications would gain access to some of the most advanced performance settings of 5G (e.g. ultra-low latency).
There remain a number of unknowns about the deployment and dimensioning of these slices: when and how will they be made available? how many slices will a given network support? How many profile of slices?
Some of the possible ways these slices would be used would be to deploy vertically-integrated applications, where the network layer is considered to be an integral part of a specific service, and would thus be relied on for its performance characteristics and its exposed capabilities (e.g. for configuration or access to edge computing). As a result, they might be fertile ground for experimenting with new approaches to integrating application and network layers, the same way private enterprise networks have enabled specific optimizations (e.g. DSCP markings in WebRTC).
Workshop participants thus encouraged W3C to explore how and when features primarily or initially targeted at private networks (by opposition to the "public internet") might be considered for experimentations in Web browsers.
Transport layer innovation
In addition to the improvements brought by the new radio capabilities of 5G and its associated service-oriented architecture, a number of changes at the transport layer were reviewed at the workshop.
John Grant (Nine Tiles) presented the ongoing work in ETSI to define a next-generation protocol to replace the IP layer, aiming at reducing latency and increasing security in the context of operators networks. Several workshop participants provided feedback on the overall comparison with the existing IP stack and the improvements that the ongoing work on QUIC are expected to provide, and warned that any replacement for IP would struggle to gain adoption, esp. as any performance gain would likely be lost at the foreseeably necessary connection to IP networks.
Colin Perkins (University of Glasgow) offered lessons learnt from the TAPS (Transport Services) effort in IETF to replace the current standard socket API used across most environments to interact with IP networks by a semantically-richer API in an age where the network stack has grown a lot more complex and featureful.
Eric Vincke (Cisco) shared the prospects of better integration between applications and networks provided by the IETF work on "Provisioning Domains" (PvDs). The underlying specification enables network operators to inform connecting devices of network specificities (e.g. whether the network is metered, whether it requires Web-based sign-in before usage) and opens up the way for networks to describe and link to services they might expose.
Application layer needs
In his workshop's keynote address, Jeff Jaffe (W3C CEO) illustrated how all the innovation promised by 5G would only be meaningful if it gained adoption at the application layer. He highlighted some of the advantages the Web has as a platform to push this adoption: cross-platform availability, largest developers pool, broad open-source leverage, and largest end-user market.
He offered a number of domains which were candidates for this higher integration of network and applications: real-time communications, next-generation media delivery, Internet of Things, automotive, virtual and augmented reality. He invited the telecommunications and network community in providing the right hooks to make this integration possible, without assuming that the abstract network layering model would suffice.
Dominique Hazael-Massieux (W3C) provided his perspectives on how the Web as a platform is the most network-integrated development platform, noting its unicity in having an architecture built around identity of network sources. He then explored how some of the ongoing changes in the platform (streams, peer-to-peer topologies, deeper integration with network, hardware, OS and cloud services) could shape into new directions for the Web, many of which had strong dependency on network integration. In particular, the sketched notion of a "cloud native browser" proposed to make browsers programmable orchestrator of network and cloud resources on behalf of the end-user.
Workshop participants heard from two specific application domains of their needs from and impact on network evolutions.
Diego Gonzales (Samsung) introduced the role of the Web as a platform for immersive content, i.e. for virtual and augmented reality experiences and noted the specific requirements in terms of bandwidth and latency that these immersive technologies have to create smooth and enjoyable experiences. In particular, the need to keep the photon-to-motion latency under 20ms was discussed as key to avoid motion sickness.
Lucas Pardue (BBC) described the rising consumer adoption of IP-based live media distribution as well as the rising bandwidth that new media technologies (e.g. ultra HD) require. Multicast IP architectures are viewed as one way to avoid wasteful multiplication of one-to-one streams, when a single stream could serve many consumers. While operators can provide such services via e.g. Multimedia Broadcast Multicast Service (MBMS), the landscape of MBMS-capable clients is still limited. BBC R&D has published an experimental profile of HTTP over QUIC that is suited for IP multicast and uses HTTP Server Push to enable unidirectional delivery. They have developed an end-to-end proof of concept using an embedded client and have recently built a prototype browser client using Media Source Extensions and Service Worker. That work identified an API gap for HTTP Server Push in the Web Platform that was filled with a custom interface. Workshop participants discussed what underlying network capabilities (discovery, congestion control) and new browser features (circuit breaker, HTTP Push events) were needed to make such an approach scalable. Further discussions on this problem space are expected to take place in the W3C Media & Entertainment Interest Group.
Shared evolutions across layers
Workshop participants also reviewed a theme with major impact across layers: the recent progress in machine learning has already enabled a number of new use cases in the application layer (including the emergence of capabilities for machine learning in browser), and is being actively explored in its role to automate network management and computing resource allocations - two key aspects of the softwarization of network enabled by and required for 5G.
The goal of these discussions were to get a shared understanding of the expected role and current deployment of machine learning algorithms across layers, as well as identifying possible needs for coordinated approaches (e.g. around models) when machine learning is used simultaneously on these different layers.
Joe Butler and Sharon Ruane (Intel) presented the results and open questions from their research on the role of machine learning for orchestrating the edge and optimizing the allocation of computing resources (see also their related position paper).
Dan Druta (ATT) introduced the Acumos framework which allows cataloging and sharing of models and predictors, and illustrates both the importance of machine learning in the telecommunication industry and the value of collaboration across industries in that space.
Göran Eriksson (Ericsson) shared some of the challenges posed by the need to build, expose and maintain machine learning models at different points and across different layers when building an AI-enabled Web application. Some of these challenges could be addressed by better cross-layer collaboration, which Göran offered might be easier to develop and experiment with in the context of private networks (as discussed earlier in this report). In particular, the potential opportunities of sharing performance metrics between browser and network to enable better optimization was noted.
Open questions and next steps
Workshop participants identified a number of exciting opportunities for bringing more exposure of network capabilities to Web developers.
Making this possible requires shared commitments from different stakeholders:
- network operators need to be ready to relinquish some of their exclusive control on network parameters,
- relevant standard organizations (e.g. 3GPP) need to ensure the way these capabilities are exposed matches the need of application-layer communities by including them in their technical developments,
- browser and operating system vendors need to recognize the additional value that application developers can bring via these network capabilities,
- and application developers themselves need to develop techniques to enable the progressive adoption of these capabilities.
A number of workshop participants volunteered to come together and build proposals that would pave the way to these commitments and will thus drive the core of the follow-up activities in the next six months. This informal task force is expected to report on its progress during the next W3C Technical Plenary day (October 24 2018 in Lyon).
At the more minute level, the workshop also identified the following technical topics worth further exploration:
- Exposure of HTTP Server Push to userland JavaScript. This can enable novel use cases on the Web, such as media streaming managed by a Service Worker. This will be explored in the Media & Entertainment Interest Group.
- Continued exploration of the role of browsers in the deployment of machine learning applications