One of the key promises of 5G is to provide near-zero latency with high reliability for mission-critical applications. The industry calls this ultra-reliable low latency communication (URLLC). It has use cases in factory automation, autonomous driving, smart grids, augmented and virtual reality (AR/VR) and haptic systems (e.g. giving a sense of touch to a remote machine operator). Communication network equipment vendors are making tremendous advances to reduce latency to meet 5G goals and IP routing is part of this effort.
What is latency?
Latency is the delay that a data packet experiences as it makes a round trip through the network.
LTE networks cut latency by about half compared to 3G networks. But there is work to do to achieve the target 1 - 10 milliseconds required for 5G URLLC applications. The following table shows average LTE and 3G latency in milliseconds for selected countries, according to recent OpenSignal reports.
Average latency in selected countries
Addressing the factors that contribute to latency
As shown in the figure below, there are many things that contribute to a data packet’s latency. These include processing at the user equipment (UE), application processing performed in the cloud, and transport in both directions between the UE and the cloud.
Network elements that contribute to latency
Industry experts are finding solutions for reducing latency at each step along this pathway through the network, notably in the radio access network (RAN). Mobile transport, or anyhaul, accounts for a portion of the end-to-end latency. Anyhaul latency is determined largely by distance, geography and network architecture. The Evolved Packet Core was introduced for LTE and end-user devices have widely adopted IP interfaces. IP packet data networking (a combination of IP, MPLS, and Segment Routing) is now typically used to provide a service layer for mobile transport. As part of the anyhaul network, IP routing plays an important role in latency.
Latency across a router is measured from the time the first bit of a data packet arrives at the router until the last bit of the same data packet leaves it. For routers in today’s networks, latency is in the range of tens of microseconds. This amount seems negligible, but every microsecond matters when the goal is to take end-to-end latency below 10 milliseconds. Fortunately, 2018 will see the arrival of new routers with latest-generation silicon that can reach latency targets under 10 microseconds.
Network congestion makes latency targets more difficult to achieve. Applications that require low latency need the right quality of service level so they can be given priority and transported through the router with minimal or no buffering. Delay variation, or jitter, must also be minimized. For network slicing, the appropriate class of service – including buffering, priority, jitter, policing and queuing – will need to be provisioned to deliver the latency required for each slice.
The amount of processing required by networking functions such as synchronization and encryption can also affect latency. Routers that use multi-core processors with efficient multi-processing can help minimize the impact on data stream processing.
Ultra-reliability goes hand in hand with low latency for 5G mission-critical applications. To guarantee service levels, routers need redundancy, fast convergence times based on the best possible control-plane performance, and comprehensive traffic management including in-depth, per-service packet flow monitoring.
Radio access network architectures and anyhaul latency
Anyhaul latency also comes into play in RAN architecture implementation. RAN functionality can be split over different locations. In LTE networks, operators moved to centralized RAN architectures to reduce costs and support more efficient communication between centralized units. This efficiency is essential for advanced LTE features such as coordinated multipoint (COMP) and carrier aggregation.
In a centralized RAN, most of the radio signal processing occurs at the hub site. The common public radio interface (CPRI) or Open Base Station Architecture Initiative (OBSAI) protocols are used over optical fiber to move high volumes of physical-layer data between the cell and hub sites with low enough latency. The distance between these sites is limited to 20 km based on the speed of light in fiber. This approach is inefficient, inflexible and insufficiently scalable to support the volume of traffic and different services expected in 5G.
For 5G, standards bodies are looking at different ways to split RAN functionality between distributed and centralized locations. Moving the split to a higher layer would mean that more physical layer processing is done at the cell site. It would also relax the transport latency and bandwidth requirements between the nodes. The tradeoff is that fewer processing functions can be centralized for performance and cost improvements.
Moving the split up the RAN functional stack – for example, by splitting the PHY layer as shown here – may permit the use of packet networking. A packet network has the advantages of supporting statistical multiplexing for greater link efficiency, supporting multipoint-to-multipoint connectivity, having the ability to scale to higher bandwidths without hardware upgrades, and allowing for multipurpose use of the network links for fixed and mobile transport. IP routing allows for different service levels for user data, control and management signaling, and synchronization.
The use of IP routing for fronthaul is attractive because it supports lower cost, standardized network control and management functions, software-defined networking, and the extension of network monitoring and tools.
The question is whether packet transport will be workable in cases where a lower-level RAN functional split is used. A lowering of the split increases the amount of processing done in the centralized hub site. It also increases the importance of high bandwidth, low latency and low jitter. The latest generation of cell-site routers have latency and jitter that, in many cases, will be low enough to make IP routing a viable choice for architectures with a low-level functional split. New interface protocols such as eCPRI, which is designed specifically for this packet-based implementation, are being standardized.
Gaming, AR/VR, remote operator control, autonomous vehicles and drones will all depend on low-latency communications. Improved latency and reliability will open the door to new competitive service opportunities.
Operators seeking to upgrade their fixed networks for 5G must determine whether and how the latest generation of transport equipment can help them achieve their 5G latency goals. Those thinking about IP networking should look beyond throughput and port counts and consider router latency, quality of service, management and reliability features. These features present a tremendous opportunity to create new services and generate new revenues while meeting the needs of 5G-era applications.