Architecting Your Network for 5G
5G has the unique challenge of enabling not only well-known services, such as mobile phone calls and ultra-broadband, but new services with very different characteristics. These services include low latency machine communications, such as autonomous vehicles, and Internet of Things (IoT), which may see as many as a million devices per square kilometer. This will call for significant changes to the wide area network to meet the reliability, scalability, low latency and connectivity requirements that these diverse applications demand.
5G will also make possible decentralized data center architectures, such as multi-edge computing, or MEC. This article will focus on how the IP/MPLS (multi-protocol label switching) backbone needs to accommodate MEC and the changes required to enable WAN services to interwork seamlessly with the data center, including which protocols are required to make this happen.
But first we have to understand the basic principles of the 5G network architecture.
As mentioned, 5G will introduce more elements and services than ever before. Designed to provide 10 Gb/s peak data rates and massive amounts of device connectivity, the first requirement is to cost-effectively avoid scaling bottlenecks. This will require employing cloud-native principles, such as the separation of user and control planes (CUPS). Cloud-native services will require end-to-end (E2E) traffic engineering and the network fabric will be designed to allow granular slicing on a per-service and per-tenant basis simultaneously. Full virtualization will also require dynamic placement and automatic connection of network functions (NFs) into service chains. To ensure that the intent of policies equals actual QoE outcome for end users, closed feedback loop assurance will enable ccontinuous monitoring of SLAs and dynamical resource adjustment.
Looking at how this will affect the data center/WAN interface, rather than looking at the WAN and data center networks in isolation, we will need an open, scalable and unified network architecture that seamlessly unifies these worlds in the telco cloud. Thus physical network functions (PNFs) must be virtualized, which means that the user-network demarcation point shifts from a physical interface on a network appliance or gateway to a virtualized network interface on an application server. As the UNI moves to a server or a leaf switch, the service data path is extended across the WAN into the data center infrastructure in order to meet deterministic performance requirements.
How to achieve this? Seamless MPLS is a sound basis and good starting point for providing some of these capabilities. Now 10 years old, it is an established protocol that provides E2E network architecture, seamless interworking between network domains at large scale and layer 2/3 service convergence. However, it does not provide E2E traffic engineering, seamless VNF interconnection into the data center, nor does it support transport network slicing or service function chaining.
For these functions we need to evolve from seamless MPLS to meet the more comprehensive capabilities of 5G. An evolved seamless MPLS needs to enable agile and seamless connectivity for distributed NFs in the telco cloud edge and core datacenters. A unified IP/MPLS wide area network is still critical but it needs to support E2E topology awareness and service-aware traffic engineering for deterministic SLA guarantees.
There are a number of key protocols that should be supported:
Ethernet VPNs (RFC 7432) are the de-facto standard for network virtualization in data centers and provide Layer 2 (L2) or Layer 3 (L3) virtual private network connectivity between VNFs that are part of a virtualized network service. NF reachability information is exchanged by means of multi-protocol extensions of the BGP control plane, which is far more scalable than conventional address flooding and learning strategies. EVPNs can also include PNFs to support services that include both network-native and/or cloud-native functions.
Segment Routing (SR) is a highly scalable approach to source-based routing that leverages regular IGP routing protocols such as OSPF, IS-IS and BGP to distribute topology information and only requires SR head-end routers to keep forwarding state information. As a result, SR enables dynamic traffic engineering (TE) services and granular per flow/application steering with various loose or strict routing constraints including bandwidth, latency, path diversity, and explicit objects to include or exclude in the route.
SR-TE policy defines a constrained shortest path between a source and a destination node through which an NF can be reached (e.g., aggregation router, a Top-of-Rack switch or a leaf router). The SR-TE policy concept decouples the forwarding control and data planes, which enables the implementation of abstract SR-TE policies on various underlying data path technologies including MPLS, UDP or IPv6. SR-TE also implicitly supports multi-path routing, which allows for load-balancing of traffic over all available links that meet the SR-TE route policy. SR-TE policy also enables service chaining using the service function chaining constraint.
In summary, the evolution to a cloud-native 5G architecture will create a more open, scalable and unified network architecture that will make it possible to unify the telco cloud and the data center. That means, in effect, no artificial barrier between the data center and the WAN and no involvement of data center gateways in the service plane. Eliminating the gateways will allow E2E traffic engineering, an E2E BGP control plane and E2E segment routing in the data plane based on SR and SR-over-UDP.
This evolution of seamless MPLS will also allow for one global H-SDN controller with E2E topology, service awareness and service chaining for policy-driven virtualized networking across the WAN and data center using SDN. This will enable us to dynamically connect physical and virtualized network functions requiring deterministic QoS. It will also allow for automatic mapping of NF connectivity SLAs to underlay network traffic engineering policies using SR-TE.
We still have a way to go before the full 5G architecture is rolled out, but these changes represent some of the best practices for interoperability, compatibility and scalability, all of which will be in high demand to support the exciting new cloud applications that are on their way.
Wim Henderickx is director network consulting engineering & PLM Technology, ION, Nokia.