The industry's first L3.5 autonomous driving network is solving intractable problems
The most common vision for the self-driving car is one where you can climb into a car, say "Take me to work," and it does. Along the way, the vehicle navigates a number of obstacles, from choosing a better route because there's a big jam on the interstate, to breaking suddenly and changing lanes because another driver swerved into your path, to handling bumper-to-bumper traffic getting through the bridge tolls, to finally finding a good parking space in the downtown garage.
Engineers making safe-driving cars have developed a nomenclature that helps define just how autonomous the car's technology really is. At Level 0, the driver is fully responsible. Level 1 adds one somewhat automated system (think cruise control). At Level 3, the car can steer and control speed, but with the driver ready to step in at any time. It's at Level 4 to 5 where the driver can take a nap and the car does it all.
Autonomous driving involves a mixture of sensors, artificial intelligence, machine learning, and high-speed communications. Your car is, essentially, one node on the nation's road network, and the AI in your car is mostly responsible for the safe functioning of that one node.
But what if you wanted the network to manage all the nodes? That's the idea behind network automation, or automatic driving networks (ADN). Here, we're no longer talking about cars and roads, but actual computer and data center networks. In fact, many of the problems are similar.
Computer networks can have thousands of paths, hundreds or even thousands of devices, many thousands of running workloads, possible rogue or destructive players, and constantly changing conditions. As an IT professional, you're not responsible for one single node on the network, but the whole network overall.
Despite the complexity of network management, ADN helps network engineers arrive at their destinations safely and hassle-free, just like a self-driving car.
And that network is incredibly complex. It's no longer just what you can fit into a data center building. Modern networks are mixtures of public and private clouds, physical data centers, and many different devices and solutions from a never-ending parade of vendors.
Managing all that has been a huge challenge, because even the management tools differ from vendor to vendor. But what if AI and machine learning could help? What if AI could learn about all the devices and how they communicate? What if you could provision and operate networks based on specific goals and intents, rather than by constantly fighting device drivers? What if data center networks (DCNs) could vastly reduce tedious manual work and "self-drive" the networks like self-driving cars will one day take you to work?
From automating cars to automating data center networks
That's what ADNs, or automatic driving networks, are helping with. Huawei is doing considerable work in this area and, in concert with the IEEE UAE Section, released a white paper about their L3.5 Autonomous Driving Network at the recent Huawei Connect Dubai 2022.
The white paper provides deep insight into the architecture and key capabilities of L3.5 autonomous driving networks (ADNs) in data center scenarios, as well as deployment practices across key sectors from finance to public services
That's right. Data center ADNs have autonomy levels, too. Level 0 (or L0) is all manual operations. L1 is called "tool-assisted operation." If you've ever written a batch file or a script, you've been working at this level. L2 is a step up, where fabric and virtual private clouds are being deployed. And then there's L3, "conditional autonomous network".
L3 is when you start to see the network perform some of the heavy lifting. From an engineering perspective, you're no longer defining specific network connections, but you're driving network operations through service intents. An L3 ADN starts to make decisions, starts to deploy workloads, and starts to relieve a lot of the manual or hard-coded algorithmic work. Essentially, the idea of data center intents at L3 is that the network engineer describes what they want the network to do (the intent), and various orchestration tools figure out how to do it.
Before L4 "highly autonomous network", Huawei has been working at L3.5. This is a big jump, because instead of focusing on one network, like inside one data center, L3.5 deploys intent-driven automation and AI across data centers, clouds, and vendors.
It's here that the operations and maintenance (O&M) load can be substantially reduced, and the quality and responsiveness of service can be substantially increased.
Key technologies
There are a number of key technologies that Huawei's L3.5 ADNs for data centers are able to deploy. Let's look at some of them now.
Open programmability platform: With Huawei's low-code open programmability platform, engineers can use text-based instructions or drag-and-drop operations to describe workflow orchestration and service knowledge to specify operational intents.
A key strength of this system is that it does not require network engineers to be professional programmers. It provides tools that fit within the experience of a network manager. The open programmability platform provides both a development and runtime environment that allows network engineers to commission and release processes, manage component lifecycles, and maintain and manage the development resources themselves.
Intent orchestration platform: Building out network operations can be enormously complex, tedious, and error prone. Intent orchestration allows engineers to focus on service goals rather than specific device interfaces and vendor-proprietary configurations. Huawei's intent orchestration platform gives engineers more than 100 network element models they can use as building blocks to create service flows on a network-wide basis.
This approach provides for powerfully automated scalability across networks, clouds, and vendors. With a drag-and-drop interface, network professionals can provision complex multi-cloud heterogeneous networks in seconds.
APIs published from orchestrated service flows: Once intents and sophisticated automated workflows are created, it's often necessary to connect them to other IT processes, legacy systems, vertical applications, cloud-based SaaS services, and more. Workflows created by L3.5 ADNs automatically publish API libraries that allow external systems to query, update, manage, and invoke service flows.
Huawei has data showing that this sort of seamless integration can substantially free up the time of network professionals and cut down repetitive manual network engineering tasks by more than 70%.
Huawei's implementation: CloudFabric 3.0 Data Center Network
Today's data centers are more of a concept than a physical building. The modern data center consists of a mix of physical data center facilities, public clouds, private clouds, and the networks that bring them together.
Keeping all this running smoothly requires management, control, and analysis functionality, but it's so complex that artificial intelligence (AI) assistance is an absolute necessity.
That's where Huawei's iMaster NCE-Fabric platform comes in, enabling highly automated and intelligent network O&M across the entire lifecycle of data center networks. It is "the brain" of the Huawei CloudFabric 3.0 Data Center Autonomous Driving Network Solution, the industry's first L3.5 autonomous driving network solution, drawing on feature-rich CloudEngine series data center switches and iMaster NCE technologies including an intent engine, automation engine, analytics engine, intelligence engine, and network digital twin base.
Huawei CloudFabric 3.0 integrates cloud systems as well as physical devices, and functions as a network control hub for management as well as collaborative provisioning. Via L3.5 autonomous driving, it manages tasks that previously had to be done manually, greatly increasing service efficiency. It helps simplify deployment via a GUI that allows you to add or delete network resources and network intents. And, of course, it's intelligent, utilizing the technologies previously discussed to keep your networks running optimally.
Some powerful case studies
There's a lot to like here, but the best way for you to see how valuable these new systems are is to look at some real-world examples. Let's look at a financial institution, a public sector health and services provider, and an automobile maker.
Finance: A top commercial bank in China wanted to grow by increasing its retail offerings, knowing that digital services provided through digital transformation could generate that growth. It soon became evident that meeting the rollout goal without help would not be possible.
After incorporating Huawei CloudFabric 3.0 into its architecture, and using Huawei's Agile Open Container (AOC) to integrate devices from multiple vendors, service rollout efficiency jumped 100-fold, allowing the bank's previously unattainable goals to be easily met by its current level of IT staffing with AI-driven solutions.
Public Service: In 2013, the largest public healthcare and social security institution in Latin America began the process of transforming its business and processes to digital, with a goal of increasing the quality and reducing the cost of social services by building a medical cloud and telemedicine platform that would let patients book appointments, file for insurance reimbursements, and access pension services.
Implementation of Huawei's ADN features via CloudFabric 3.0 enabled the institute to reduce service interruptions by monitoring the network in real-time, identifying problem areas, intervening automatically when possible, and proactively eliminating the conditions for faults before they could happen.
Efficiency in locating faults improved 10-fold. Data center reliability was increased, negative comments went down, and positive comments increased along with the public's trust of the popular new system.
Manufacturing: A large automobile manufacturer in China needed to overcome its legacy of siloed management, poor service provisioning efficiency, and laborious manual O&M.
The company deployed Huawei's multi-data center controller (MDC), which provides end-to-end software defined network management across hybrid cloud networks, talks to public cloud APIs, and provides the ability to manage policies across the entire hybrid cloud environment.
Huawei's CloudFabric 3.0 introduced unified collaboration and automated management capabilities across the various business units, and its automated AI-based solutions replaced the meetings and inter-jurisdictional logjams of the past.
We talked about self-driving cars as an inspiration for self-driving networks. This auto manufacturer needed to build and operate actual autonomous driving vehicles, and it benefitted greatly from the help of an autonomous driving network.
Deep insight into real-world service scenarios and pain points
"Huawei's data center autonomous driving network is the industry's first to evolve from L3.0 to L3.5. This is the result of our deep insight into real-world service scenarios and pain points of customers across sectors such as finance, public service, and manufacturing, as well as our joint innovations with industry forerunners," said Arthur Wang, Vice President of Huawei's Data Center Network Domain.
To learn more about these innovations, download the L3.5 Data Center Autonomous Driving Network white paper here. This white paper shares best practices across these sectors, to enhance enterprises' competitiveness by optimizing your network architectures and operating modes and helping you build agile and reliable services while reducing your OPEX and CAPEX.
More information about Huawei's Data Center Autonomous Driving Network is available here.