Reimagining the Edge with 5G: A Closer Look at Traffic Patterns

Kaniz Mahdi and Paul Littlewood, Ciena 

IEEE 5G Tech Focus: Volume 2, Issue 1, March 2018 

1. Introduction

Fundamentals of wireless networks have remained essentially static through the evolution of radio from 2G to 3G to 4G, designed mainly to provide connectivity for (geographically dispersed) human communications, with some provision for machine type communication. However, 5G era networks are expected to serve a user environment that is significantly more complex than that of today, as 5G, with its broad range of new radio characteristics and capabilities, benefits from new market dynamics being shaped by SDN/NFV, and technology advancements in M2M communications, Robotics, Artificial Intelligence and Machine Learning. This paper looks at the changes expected in optical networks to mirror the new capabilities being designed into 5G wireless networks. A richer connectivity structure will be necessary to ensure economic support of network applications distributed across the metro cloud, and to support the latency sensitive mobile applications enabled by low latency 5G wireless systems.

2. Service and Traffic Variance

5G services can be broadly characterized into three distinct groupings, each with its unique set of extreme characteristics [1]:

1. Extreme Mobile Broadband (xMBB): high data rates, low latency communications, with extreme coverage.
2. Massive Machine Type Communications (mMTC): scalable connectivity for a large number of devices (tens of billions), efficient transmission of small payloads, with wide area coverage.
3. Ultra-reliable Low Latency Communications (urLLC): ultra-reliable, low latency connectivity for services with stringent requirements on up-time and latency.

This broad landscape of radio characteristics, when compounded with new advances in Machine Intelligence, is expected to drive ‘order-of-magnitude’ scale changes to ‘autonomy’ in industrial processes and consumer applications. Associated radical implications on demand side equations which dictate network topologies coupled with autonomous operations, necessitate that control and intelligence are close, in physical terms, to the point of data collection and consumption to ensure appropriate control function performance.

3. Intelligence Distribution Drives Traffic Flows and Density

Today’s networks are optimized for today’s dominant services, where the majority of traffic flows between users and content centers (U2C) owned by service providers or Internet Content Providers (ICP) located in, or near, metro cores; and cloud data centers owned by ICP’s in rural areas and accessed via Exchange Data Centers, as shown in the left diagram in Figure 3. User to User (U2U) traffic, composed primarily of voice, video and messaging, also follows this pattern since service switching takes place in centralized metro or backbone locations.

Through initiatives such as the Central Office Reimagined as a Data Center (CORD) [2], which aim to transform select service provider Central Offices to Data Centers, current generation services are intended to be supported more efficiently through functional virtualization. Additionally, closer proximity to the end user is expected to improve application response through lowered latency. CORD implementations envisage two tiers of data centers - one or two ‘Central Data Centers’, and up to ten or so ‘Local Data Centers’ (in large cities and associated regions). These locations will equate to ‘gravity wells’ by drawing user traffic towards them thereby increasing surrounding flow density, as shown in Figure 1. Expanded traffic capacity to interconnect the CORD tiers will arise due to the increasing importance of these data centers in agile service provisioning.

 

Figure 1: Distributed functions in Local CORD data centers will draw traffic

 

For the two CORD data center tiers, xMBB and mMTC categories of 5G services are anticipated to source (into the network) a wide range of traffic volumes. Devices sourcing this traffic will range from simple sensors producing a few bits per hour, through communications and entertainment devices producing large volumes of data, e.g. high definition entertainment video, to complex machinery producing telemetry data of many PB daily. The broad range of traffic profiles exhibited by these services is expected to balance the traffic flows in the network which currently are dominated by ‘core to user’ traffic.  Eventually it may even result in reversing the balance to ‘users to core.’

Figure 2: Widely distributed Access data centers for delay intolerant traffic and telemetry pre-processing

User content, which is sourced today from Content Centers such as Video Hub Offices, will remain fairly centralized in Central and Local Data Centers as latency demands for content distribution are relatively lax and caching costs per user are lower when distributed across a larger user community.

Stringent low latency constraints of the uRLLC category of 5G services is anticipated to drive addition of yet another data center tier, the ‘Access Data Centers’, which provide for an even greater proximity of applications to the user. Distribution of functionality, either network or application, between the tiers is governed by response requirements from the network balanced by the cost to perform computation, and data or content storage. For uRLLC services, transport to and from Access Data Centers must consume no more than 1ms from a control loop latency budget, whereas applications located in the Local and Central Data Center must tolerate around 10ms and 100ms latency respectively. As uRLLC services proliferate the effect on traffic will be to localize significant capacity around Access DC’s.

 

 

Figure 3: Traffic pattern evolution

Load balancing between data centers will be important to ensure most efficient use of computing resources. Effective transfers of workloads and data will need additional network capacity to reduce hair-pinning or tromboning driving the need for more direct data center interconnection.  These additional links, illustrated in the right hand diagram of Figure 3, are especially important for mobile low latency services requiring transfer of the control function between local data centers. Direct interconnection will result in the fastest and most predictable transfers.

4. Summary

The complex user environment supported by 5G network technologies, the flexibility required for efficient service support by CORD, the requirement to deliver new services rapidly, and especially, the rigorous demands of uRLLC services are expected to drive significantly large changes in network topology.

Future traffic patterns are expected to evolve with an increasingly rich connectivity at each step, from the centralized environment of today, to the distributed star patterns of CORD, and ultimately toward meshed connectivity deemed essential to support the complex user environment expected with confluence of 5G, Robotics and Machine Intelligence.

References

1. NGNM 5G White Paper: www.ngmn.org/fileadmin/ngmn/content/downloads/Technical/2015/NGMN_5G_White_Paper_V1_0.pdf
2. CORD®: REINVENTING CENTRAL OFFICES FOR EFFICIENCY & AGILITY: https://opencord.org/

 

 Kaniz Mahdi is Vice President of Advanced Architectures at Ciena, and a highly regarded telecom visionary.  In this role, Kaniz is responsible for Ciena’s global technology strategy and target architecture for 5G era Networks underpinned with Open Architectures, Network Function Virtualization (NFV), Software Defined Networking (SDN), Machine Intelligence and Autonomics.

Kaniz spent the last few years as VP/Head of Architecture at Ericsson passionately shaping technology landscape for the multi-faceted transformation of telecommunication industry with Cloud, SDN, and 5G. Prior to joining Ericsson, she headed Communications Services Standards Research at Huawei Technologies, and held various roles in System Architecture and Product Design at Nortel Networks. Kaniz has a stellar record of continuously pushing the envelope on new technologies with more than 60 patents and publications; she is lead inventor of key technologies essential to current Voice over LTE systems and played an instrumental role in adoption of Web RTC. She holds a Bachelors of Engineering degree in Electrical Engineering and a Master of Science in Telecommunications.

 

Paul Littlewood is a Principal Engineer in the office of the Chief Technology Officer at Ciena.  His current areas of study include the impact of next generation wireless and wired network convergence on network traffic and the consequent need for network modernization.

Previously Paul has researched network architecture evolution and has led product management and engineering teams in building optical transport and digital cross connect systems.  He has also led in the definition and development of multiple network technologies including Carrier Ethernet, Resilient Packet Rings and multi-layer control. Paul has seven patents granted and has written a number of papers on optical networking.  He has an Honours Degree in Pure Physics from the University of Newcastle upon Tyne in Great Britain.

 

Editor: Patagiotis Demestichas