The US and Polish examples in the previous report are two of many which epitomize the complex choices operators will need to make about network architecture, and how that will influence the physical assets they will need to access. The balance between centralized and edge-based resources, and the distances between the two, will be important, and will vary depending on the key use cases which will drive the 5G model. While virtualization and software-defined networking (SDN) enables an MNO to be far more flexible in allocating its digital resources, those towers, data centers and local sites and servers will not be so easily rearranged, so the topologies will need to be considered carefully.
Some of the options are not fully clear yet, as they will depend on how certain technologies evolve. The biggest challenge to centralizing the RAN over long distances is the cost and availability of fiber of sufficiently low latency and high quality for fronthaul. Ethernet-based alternatives to CPRI are helping greatly with fronthaul economics, but other innovations in the fiber itself will need to emerge, and be cost-effective, to improve the case further.
At the edge, there are debates about the best architecture – the telco-centric MEC, the IT-focused OpenFog, other approaches like cloudlets. There are also disputes over whether edge computing will be most effective in a mobile network if the cell site and the processor are converged, or whether the edge server would naturally sit somewhere separate from the antenna. Nokia and Intel had pushed the former idea with early developments like Liquid Apps, which places a processor at the cell site.
The advantage of that is close integration of radio and IT functions so that distributed cloud services, such as context-aware marketing, can be linked to precise awareness of a user’s location, and the best connection available to them. However, recently there has been a swing towards planning the cells and the edge servers separately, with some arguing that there are trade-offs when placing them together. Much depends on where the operator defines the edge – on one side, a MEC server might be placed with a switching center and support a large area and many users; on the other, some visions of massive Internet of Things connectivity might require many microservers at the very edge of the network – which might lend itself to small cell integration, and swing the pendulum back to the Nokia approach.
Projects like the Flat Distributed Cloud (FDC) at the UK’s 5G Innovation Center at the University of Surrey indicate the full 5G potential of this concept. This was demonstrated in late 2016 in partnership with virtual packet core specialist Quortus, plus Cisco and Huawei. The demo showed the FDC running a ‘5G’ core, together with a fully orchestrated virtualized RAN. It aimed to show that a 5G radio will greatly enhance the effects of SDN/NFV, by supporting a far flatter, more distributed IP mobile network than has been feasible in the real world before, and so enabling a far more sophisticated interpretation of central concepts like network slicing.
The 5GIC work will be closely watched in all these respects. Many other 5G projects are likely to come out with their own versions of the fully dynamic core network soon, but 5GIC is early in the game. Its approach pushes content to the edge and flattens the architecture to reduce latency – the FDC design will require just one layer in 70% of cases, and two in the rest, says the team behind it, rather than three for current LTE. FDC proposes a “horizontally distributed cloud model that relies on new user and control plane protocols allied to the implementation of NFV and SDN, with all network nodes equipped with service compute power and storage as well as communications processing”.
So far, so similar to many other 5G efforts, such as some of the European Union’s 5G-PPP initiatives. But the 5GIC insists that its approach relates general concepts to “the very specific challenges of mobile architectures”, rather than going for “blind adoption of SDN, NFV and cloud principles”.
Each node in the FDC architecture is enabled for both services and communications, with processing and storage. Clusters of dynamic, virtualized cells are linked to hardware clusters in data centers, which may be nearby or remote depending on available backhaul/fronthaul. Another technique borrowed from LTE is dual connectivity – the operator can connect to both the control plane node (mapped to a macrocell) and the user plane (the most appropriate user plane cell within a cluster of small cells with a virtualized controller). The clusters can be reorganized according to changing conditions such as load versus time, in order to optimize the network – NFV is used to reconfigure the clusters on-demand while SDN allocates resources where there are highly dynamic traffic patterns and dense environments.
The FDC was demonstrated end-to-end over LTE-A, from commercial mobile devices to internet services. The cloud itself was running on off-the-shelf Intel processors and Linux. This means, 5GIC says, that operators can deploy VNFs as network services, without engineers having to perform physical upgrades on the physical cell sites. A VNF can be deployed on the FDC in 10 minutes, it claims, compared to 10 days for traditional systems.
The demonstration was part of the EC Horizon 2020 virtualization project SoftFire, using the open source ‘OpenBaton’ orchestrator plus OpenStack as the VNF controller.
Amon the key focuses of the FDC project are network slicing, which will be the subject of the next FDC demo, and context awareness. The system uses information about the user and network status to provide better quality of experience over its dynamic and distributed cloud architecture, directing resources where they are most needed.
FDC is able to understand a user’s (or ‘thing’s) context and deliver just the right amount of capacity, latency and so on required for that particular situation. That would achieve one of the key aims of 5G – to support a user experience which is perceived to be always sufficient (as opposed to over-delivering, for a given usage, on some occasions and then disappointing on others).
The FDC approach also proposes extending the current RRC (Radio Resource Protocol) into a Common Resource Connection Protocol (CRCP), allowing multiple bearer types from multiple technologies to be combined to support a single dynamic, virtual connection from a device to the FDC network. This would allow multiple radio and fixed access technologies to be treated as a single resource, mixed and matched on a per-user basis, depending on their context and usage at the time. Like all the efforts focused on mobile edge/fog computing, the aim is to turn the MNO’s access network into a far more powerful tool to understand customers’ context in enormous detail and deliver services, QoS levels, promotions and actions accordingly.