Guavus Blog

All the latest from the world of Guavus

AI: The Missing Piece to the IoT and 5G Scalability Puzzle

by Stephen Collins, Principal Analyst, ACG Research, Network Visibility and Analytics

When future historians look back at the early 21st century, one of the hallmarks of this era will be the pervasive extent of network connectivity. Today there are more than 4 billion Internet users, exceeding 50% of the world’s population. Network access is nearly ubiquitous throughout the developed world via a combination of fixed broadband, cellular data, WiFi and satellite coverage in remote areas.

Despite this impressive achievement, we’re just getting started. IoT and 5G promise to revamp the mobile networking landscape but also present service providers with operational challenges driven by sheer scale.

Explosive Growth in Connected Devices

The Ericsson Mobility Report forecasts the number of smartphone-based mobile broadband subscriptions to reach almost 8.3 billion by 2023. Gartner forecasts that the number of IoT devices will top 11 billion this year and Statistica projects that number to exceed 75 billion in 2025. It’s no stretch to say that we’re looking at an order of magnitude increase in the total number of connected devices over the next ten years.

5G to the Rescue

Of course, mobile operators are aware of the looming IoT tsunami and the race has already begun to build out 5G networks in anticipation of the expected demand. 5G technology is multi-faceted and conceived to address the fundamental problem of scaling wireless networks in several dimensions.

First, there is the need to accommodate the vast number of devices. Network densification efforts will result in a huge increase in the number of cells. The Small Cell Forum forecasts the installed base of small cells to reach 70 million worldwide in 2025, which is roughly ten times the current number of cell sites.

Second, 5G incorporates narrowband technology for connecting IoT sensors distributed across broad geographic areas using battery-efficient, low power radio signals. This is a critical requirement for instrumenting the physical world to monitor weather, the environment, agriculture production, and transportation networks, etc.

Third, mobile operators envision that 5G networks utilizing millimeter wave technology will enable a new generation of high bandwidth, low latency applications at multi-gigabit speeds. Fixed wireless access at 28 GHz promises to alleviate the cost burden associated with fiber or cable last mile connections. Speculation about other 5G use cases ranges widely from smart cities to self-driving cars to virtual/augmented reality.

Pushing the Cloud to the Edge

Extreme connectivity and capacity at the edge of networks will drive the adoption of edge computing to distribute processing, memory and storage resources closer to devices and the sources of data. Downscaled cloud-native infrastructure deployed near the edge will offload centralized hyperscale data centers and conserve bandwidth by distributing workloads, which will improve application latency and performance.

Sounds amazing. Too good to be true. There must be a catch, right? Of course!

Scaling Up Involves Heavy Lifting

Network service providers are facing a lot of heavy lifting in order to scale their networks up and out by an order of magnitude in terms of capacity, the number of connected devices and vast numbers of networks. The operational challenges are daunting.

Fortunately, network operators are already moving to software-driven infrastructure based on SDN and NFV, which will reduce CAPEX and OPEX while increasing service agility. By adopting a software-centric approach based on DevOps practices and a continuous integration / continuous delivery model, operators will be better positioned to scale and to respond rapidly to changing network conditions and market needs.

However, SDN/NFV and DevOps are just “jacks for openers.” Service providers are already excessively burdened with the operational costs of their existing networks. Scaling up and out will only increase that burden. Throwing more people at the problem is a non-starter because it is simply too costly.

Scaling Depends on Automation

The obvious solution is automation, enabling operators to streamline workflows and automate the majority of routine operations. Operators are starting to deploy a new generation of software tools for automated service provisioning and network configuration. Open source communities are hard at work developing the various components that comprise sophisticated lifecycle service orchestrations solutions. A leading example is the Linux Foundation’s Open Network Automation Platform (ONAP).

Yet key pieces of the puzzle are still missing. How to derive the real-time, actionable intelligence needed to drive automated service orchestration?

Automation Depends on Machine Learning and AI

A key benefit of software-driven infrastructure is that it can be easily instrumented to generate streaming telemetry data that can be fed into Big Data analytics engines. However, given the constant flood of a diverse array of telemetry data collected from many different sources across the network, operators need to move beyond dashboards and visualizations as the primary output of monitoring and analytics tools. Alarm fatigue, which is all too common in existing networks, is only going to get worse.

Machine-based intelligence that offloads human operators is critical to achieve the degree of automation needed to significantly reduce operating expenses. Machines will have the ability to detect gray failures, subtle anomalies and keep pace with constantly changing network conditions.

Machine intelligence encompasses advanced machine learning algorithms that can rapidly correlate information from multiple data sets in order to extract real-time insights for driving network and service orchestration. Cognitive AI techniques that emulate the decision-making processes of the human mind need to be applied for automating operator workflows.

Filling in the Missing Puzzles Pieces

We’re not there yet, but a lot of smart people are hard at work filling in the last few pieces of the 5G scalability puzzle. Both service providers and vendors are making progress applying machine learning and AI to many aspects of network and service operations. Yet much work remains be done until operators will be able to support the delivery of a new generation of IoT and 5G applications at vast scale.

With the race to 5G well underway, it is imperative that service providers ramp up fast on machine learning and AI to establish partnerships with the leading vendors and integrators who have proven expertise in applying these new technologies in large-scale operational environments.

Image attribution: