The metaverse is going to challenge webscale companies across every dimension of their operations. Metaverse technologies like AR/VR, 8K video and cloud gaming in the consumer space and Industry 4.0 technologies like digital twins, blockchain, IoT and autonomous operations will demand a re-engineering of today’s webscale networks.
Webscale companies and the cloud are at the heart of communication’s future. Thus, they will need to evolve their data center fabrics and their IP and optical WANs. As 5G and 6G networks connect everything from pacemakers to low earth orbit satellites, the stakes will become increasingly high. Automated assurance and security will need to be baked into the DNA of their networks. To ready themselves for these future consumer and enterprise services, a multi-dimensional strategy becomes indispensable.
Scaling at every dimension
Until the early 2000s, scaling computer processors relied on a one-dimensional strategy: increase clock speed. As we hit the physical limits of clock scaling, multi-core processors emerged. Today, we’re hitting new limits, causing the proliferation of specialized, task-oriented processors and other ways to offload parallel processes. Future scaling proposals are envisioned, but most increase the complexity of the hardware, which must be handled by software.
We can see a similar trajectory occurring in networks. Different strategies have been followed in optical transport than routed wide area networks. Both differ from data center fabrics, and for a long time, increasing bandwidth and speed was the principal scaling strategy. However, we are reaching limits that simply over-provisioning won’t be affordable or successful. Delivering the metaverse will require more than brute speed and oodles of capacity.
Previously treated as separate domains, optical transport, IP WAN and data center fabrics will have to work together in smarter and more complex ways to scale, assure and secure the connection from end to end. This will require orchestration across all dimensions in a responsive and scalable way. Many different but interrelated technologies will need to operate in unison. No longer can companies over-engineer their network layers and expect cost, efficiency and scale to meet their customers’ needs.
Converging data center, IP routing and optics
The rising importance of data center interconnect (DCI) puts the WAN in the spotlight. With the rise of edge clouds and the need to deliver ultra-low latency, especially to autonomous and industrial applications, the network has to become more hierarchical and distributed. Where workloads in the past were mostly moving within the data center, they are increasingly moving between data centers, and not just in the metro area network.
Simultaneously, the requirements of workloads are becoming stricter. AI workloads and their training requirements often involve workload-to-workload communications across multiple data centers. These communications are sensitive to packet loss as well as to latency. Not only must the data center’s network fabric be designed to accommodate these requirements, but the IP-routed WAN must also support these performance metrics.
Routing decisions made at the IP layer in turn need to be supported at the optical transport level. Over-provisioning at the optical layer has been the historical norm, but Ethernet ports on routers have increased from 100 Gigabit Ethernet (GE) to 400GE and even 800GE, which means that the underlying optical capacity cannot simply be assumed. It is also important to know the quality and reliability of the optical link. Does the optical signal carry sufficient distance to reach a new edge cloud and still support the Ethernet port speed? Webscalers planning their DCI networks need to think about both the IP routing layer and the optical transport layer.
Multi-dimensional networks
Just as with the example of parallel processing in computers, complex network architectures have to be handled by increasingly complex software. Such network software is responsible for dynamically coordinating the different elements, turning network elements on and off, and shifting tasks to the least cost, most power-efficient or best-performing resources depending on the workload or application being supported. And this all has to happen automatically and securely.
Unlike single computer designs, networks are built incrementally with layer upon layer of legacy network elements and software that can’t simply be overhauled. Despite these limitations, the overall network is slowly becoming automated to provide the operational scale, efficiency and cost-effectiveness needed. So far, automation is not as available across all network technologies, but this level of control is being built, more commonly in the DCI space than public networks with legacy services like voice that don’t pose new requirements.
For webscalers, the coming metaverse applications and the growing dependence on the cloud mean that automating both the IP-routed and optical networks are being driven by economics and opportunity. Especially concerning the industrial metaverse, webscalers have strong incentives to invest in their networks. Establishing a converged multi-layer network infrastructure with unified control and open, software-defined networking (SDN) interfaces across both IP and optical layers will be essential to meeting the performance requirements of these new use cases.
As well as offering new services, this approach to the network can also make operations more efficient and cost-effective. In addition to automating many tasks, it simplifies and abstracts many network processes making it possible to program the underlying infrastructure — often referred to as infrastructure as code. This not only empowers network engineers, but also saves on network equipment with cross-domain coordination and orchestration to optimize resources.
Bolstered by AI/ML analytics, network telemetry can be rapidly analyzed to troubleshoot network and service issues. Dynamic awareness of traffic flows across multiple layers allows for coordinated operations to minimize the effects of, for instance, optical topology changes on IP traffic. Diversity analysis can automatically assess true comprehensive path diversity to ensure ultra-reliable service delivery. The result is more efficient, reliable and scalable data center interconnectivity that can meet the workload requirements of AI and metaverse applications.
Securing the metaverse
Even a cursory review of the many proposed metaverse use cases - from autonomous transportation and remote healthcare to financial services and smart city infrastructure - highlights the growing mission-critical aspect of tomorrow’s services. This calls for a multi-dimensional approach to security as well.
Looking at Distributed Denial-of-Service (DDoS) alone, it is now the fastest-growing traffic category, exceeding even gaming or video. To provide the scalability and functionality required to protect highly distributed, mission-critical networks, network security must become like packet forwarding: a high-performance, highly scalable function of the routed network. Bolt-on security appliances cannot scale. Just like assurance, security must be built into the DNA of the network. The routers themselves will be required to be highly precise attack sensors and mitigation elements without compromising other services running on the same router.
The opportunity for webscalers is unprecedented. They are building the metaverse platform of the future. As the cloud becomes generalized and distributed, as well as mission-critical, every dimension of the network has to be renovated, re-architected and reimagined for it to meet the needs of the emerging metaverse and the essential services it will deliver.
