Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.
Designing, testing and provisioning updates to data digital networks depends on numerous manual and error-prone processes. Digital twins are starting to play a crucial role in automating more of this process to help bring digital transformation to network infrastructure. These efforts are already driving automation for campus networks, wide area networks (WANs) and commercial wireless networks.
The digital transformation of the network infrastructure will take place over an extended period of time. In this two-part series, we’ll be exploring how digital twins are driving network transformation. Today, we’ll look at the current state of networking and how digital twins are helping to automate the process, as well as the shortcomings that are currently being seen with the technology.
In part 2, we’ll look at the future state of digital twins and how the technology can be used when fully developed and implemented.
About digital twins
At its heart, a digital twin is a model of any entity kept current by constant telemetry updates. In practice, multiple overlapping digital twins are often used across various aspects of the design, construction and operation of networks, their components, and the business services that run on them.
Peyman Kazemian, cofounder of Forward Networks, argues that the original Traceroute program written by Van Jacobson in 1987 is the oldest and most used tool to understand the network. Although it neither models nor simulates the networks, it does help to understand the behavior of the network by sending a representative packet through the network and observing the path it takes.
Later, other network simulation tools were developed, such as OPNET (1986), NetSim (2005), and GNS3 (2008), that can simulate a network by running the same code as the actual network devices.
“These kinds of solutions are useful in operating networks because they give you a lab environment to try out new ideas and changes to your network,” Kazemian said.
Teresa Tung, cloud first chief technologist at Accenture, said that the open systems interconnection (OSI) conceptual model provides the foundation for describing networking capabilities along with separation of concerns.
This approach can help to focus on different layers of simulation and modeling. For example, a use case may focus on RF models at the physical layer, through to the packet and event-level within the network layer, the quality of service (QoS) and mean opinion score (MoS) measures in the presentation and application layers.
Modeling: The interoperability issue
Today, network digital twins typically only help model and automate pockets of a network isolated by function, vendors or types of users.
The most common use case for digital twins is testing and optimizing network equipment configurations. However, because there are differences in how equipment vendors implement networking standards, this can lead to subtle variances in routing behavior, said Ernest Lefner, chief product officer at Gluware.
Lefner said the challenge for everyone attempting to build a digital twin is that they must have detailed knowledge of every vendor, feature, and configuration and customization in their network. This can vary by device, hardware type, or software release version.
Some network equipment providers, like Extreme Networks, let network engineers build a network that automatically synchronizes the configuration and state of that provider’s specific equipment.
Today, Extreme’s product supports only the capability to streamline staging, validation and deployment of Extreme switches and access points. The digital twin feature doesn’t currently support the SD-WAN customer on-premises equipment or routers. In the future, Extreme plans to add support for testing configurations, OS upgrades and troubleshooting problems.
Other network vendor offerings like Cisco DNA, Juniper Networks Mist and HPE Aruba Netconductor make it easier to capture network configurations and evaluate the impact of changes, but only for their own equipment.
“They are allowing you to stand up or test your configuration, but without specifically replicating the entire environment,” said Mike Toussaint, senior director analyst at Gartner.
You can test a specific configuration, and artificial intelligence (AI) and machine learning (ML) will allow you to understand if a configuration is optimal, suboptimal or broken. But they have not automated the creation and calibration of a digital twin environment to the same degree as Extreme.
Virtual labs and digital twins vs. physical testing
Until digital twins are widely adopted, most network engineers use virtual labs like GNS3 to model physical equipment and assess the functionality of configuration settings. This tool is widely used to train network engineers and to model network configurations.
Many larger enterprises physically test new equipment at the World Wide Technology Advanced Test Center. The firm has a partnership with most major equipment vendors to provide virtual access for assessing the performance of actual physical hardware at their facility in St. Louis, Missouri.
Network equipment vendors are adding digital twin-like capabilities to their equipment. Juniper Networks’ recent Mist acquisition automatically captures and models different properties of the network that informs AI and machine optimizations. Similarly, Cisco’s network controller serves as an intermediary between business and network infrastructure.
Balaji Venkatraman, VP of product management, DNA, Cisco, said what distinguishes a digital twin from early modeling and simulation tools is that it provides a digital replica of the network and is updated by live telemetry data from the network.
“With the introduction of network controllers, we have a centralized view of at least the telemetry data to make digital twins a reality,” Venkatraman said.
However, network engineering practices will need to evolve their practices and cultures to take advantage of digital twins as part of their workflows. Gartner’s Toussaint told VentureBeat that most network engineering teams still create static network architecture diagrams in Visio.
And when it comes to rolling out new equipment, they either test it in a live environment with physical equipment or “do the cowboy thing and test it in production and hope it does not fail,” he said.
Even though network digital twins are starting to virtualize some of this testing workload, Toussaint said physically testing the performance of cutting-edge networking hardware that includes specialized ASICs, FPGAs, and TPUs chips will remain critical for some time.
Culture shift required
Eventually, Toussaint expects networking teams to adopt the same devops practices that helped accelerate software development, testing and deployment processes. Digital twins will let teams create and manage development and test network sandboxes as code that mimics the behavior of the live deployment environment.
But the cultural shift won’t be easy for most organizations.
“Network teams tend to want to go in and make changes, and they have never really adopted the devops methodologies,” Toussaint said.
They tend to keep track of configuration settings on text files or maps drawn in Visio, which only provide a static representation of the live network.
“There have not really been the tools to do this in real time,” he said.
Getting a network map has been a very time-intensive manual process that network engineers hate, so they want to avoid doing it more than once. As a result, these maps seldom get updated.
Toussaint sees digital twins as an intermediate step as the industry uses more AI and ML to automate more aspects of network provisioning and management. Business managers are likely to be more enthused by more flexible and adaptable networks that keep pace with new business ideas than a dynamically updated map.
But in the interim, network digital twins will help teams visualize and build trust in their recommendations as these technologies improve.
“In another five or 10 years, when networks become fully automated, then digital twins become another tool, but not necessarily something that is a must-have,” Toussaint said.
Toussaint said these early network digital twins are suitable for vetting configurations, but have been limited in their ability to grapple with more complex issues. He said he likes to consider it to be analogous to how we might use Google Maps as a kind of digital twin of our trip to work, which is good at predicting different routes under current traffic conditions. But it will not tell you about the effect of a trip on your tires or the impact of wind on the aerodynamics of your car.
This is the first of a two-part series. In part 2, we’ll outline the future of digital twins and how organizations are finding solutions to the issues outlined here.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.