Unleashing the true potential of 5G with cloud networks

For many, the intersection of 5G wireless and the Internet offers very little by way of new information, just from a quick, surface-level glance. However, things look very different when considered in light of the 5G New Radio (NR) standard.

For example, consider a new technology in the 5G NR called URLLC—which is short for ultra-reliable low-latency communications. Designed for mission-critical and latency-sensitive Internet services, URLLC is a configuration and technology feature that enables a new class of applications and services. These services include things such as remote control of robots; thin, tether-less augmented reality headsets, smart car collision avoidance systems, traffic control on roads, and entertainment services such as fast-action multi-player multi-region gaming.

The delay specification delivered by URLLC can be anywhere from one to four msec on-air latency, well within the required hard limit of the latency required by these applications. However, in many ways, this delay specification is simply not sufficient in light of the fact that many of these latency-sensitive applications only make sense in the context of the Internet. Interacting with users, devices, and cloud services over long distances forces one to think about the end-to-end latency characteristics that include the Internet.

Stretching the reach of low-latency applications

No one network can encompass all end-users and cloud services, so every network operator ultimately depends on the Internet at some point. Based on the locations of the source and destination, Internet latencies can vary anywhere from a single order to several orders of magnitude more than on-air 5G wireless latencies. This high latency can effectively remove all the benefits of the low-latency properties offered via the 5G network. As a result, the scope of the services that people enthusiastically look forward to gets reduced considerably.

Beyond URLLC, 5G traffic in general (voice and high-resolution video calls) will demand performance and reliability from both the wireless and Internet paths, even more so than traditional web and enterprise traffic.

Operators spend a lot of money to manage and maintain their networks and peering relationships, but so does Microsoft. The question then is, why are two massive industries doing the same thing? Because both parties move packets around, doesn’t it make more sense for them to collaborate?

Here, the well-managed, reliable, and performant Azure network should be thought of as the backbone that operators trust. With this shift in thinking will come all the advantages of innovation that IT companies like Microsoft are rapidly bringing in.

For example, large operators that run extensive national backbones can get tremendous benefit from extending their wide-area network (WAN) with Azure’s WAN. In this example, a 5G network that spans both will allow 5G devices to more effectively reach cloud services deployed on Azure datacenters. This includes first-party services such as, Xbox cloud gaming, as well as third-party services that Azure customers run. Smaller operators (or new operators) that do not have their own national backbones can save resources including time, human capital, and money, and instead leverage Azure’s extensive investment to build a unique 5G network on top of something that has already been proven reliable.

Azure’s planet-scale WAN

Azure maintains a massive WAN with significant capacity and one that is continuously growing. We have over 175,000 miles of lit fiber optic and undersea cable systems. This connectivity covers close to 200 network points of presence (PoPs) over 60 regions, across 140 countries.

Azure’s network is connected to many thousands of ISPs and other networks with significant peering capacity. Our global network is well-provisioned, with redundant fiber paths that can handle multiple simultaneous failures, it also has massive reserve capacity in unlit dark fiber. These optical fibers are fully owned or leased by Microsoft, and all traffic between and among Azure datacenters within a region or across regions is automatically encrypted at the physical layer.

This combination of redundant capacity to handle failures, dark capacity for significant growth, and research advancements being made in increasing transmission speeds means that we have a massive amount of spare capacity to serve 5G traffic to a broad array of new operators.

Challenges related to carrying 5G traffic over a cloud WAN

Figure 1 illustrates how packets can move from one client (in 5G terminology, a “User Equipment” (UE)) to another over the Azure cloud network, with extremely low latency.

Figure 1: Unified platform.

With 5G, there is an intense amount of pressure on traditional metrics that include latency, jitter, throughput, reachability, and loss, by which transport quality is typically judged. With on-air 5G latencies reaching close to the sub-ms range, wired transport latency will likely dominate end-to-end performance. However, with on-air 5G throughputs in the tens of Gbps range in Enhanced Mobile Broadband (eMBB) mode, just a small number of UEs could overwhelm a single peering link that today would typically have a capacity of tens to hundred Gbps.

Beefing up capacity at the peering surface area and on backbone links that support them remains an expensive endeavor. Finding efficient paths through the maze of WAN links is also crucial to achieving business success. Outages are a fact of life for any cloud network, but 5G experiences will be more sensitive to packet loss and reachability than traditional web and enterprise traffic.

Reliability will need to be increased. This will happen by guarding against peering outages and overload, and by verifying configuration changes. The performance will also need to be increased. This includes supporting streaming audio and video services by actively reducing jitter and access queuing delays. Additional factors such as cost, safety, reachability, and regulatory and business policies also must be addressed.

Further, there will be a need to orchestrate wired transport for 5G deployments. 5G networks will deploy many network functions in a distributed fashion. Rigorous needs around scale-out and fault tolerance will make these deployments more complex than typical enterprise application deployments. Orchestrating cloud network capabilities, such as Virtual Networks, Virtual Network Peering, Virtual WAN, and private endpoints, all while meeting performance and policy constraints (business and regulatory), represents a new challenge.

Making Azure WAN great for 5G traffic

For many years, Microsoft researchers and engineers have been working on a hybrid-global traffic orchestrator for routing network packets across Azure’s WAN. Our orchestrator takes control away from classic Internet protocols and instead moves that control into software that we build and control for 5G traffic. We place the 5G flows that demand high performance on low-latency, high bandwidth paths to and from the Internet. Network flows that are cost-sensitive are instead routed through cheaper paths.

In effect, we have developed a fast-forwarding mechanism to build a 5G overlay on our existing WAN, thereby supporting a variety of 5G network slices with different wired transport properties, while avoiding interference with the operation of the underlying enterprise cloud network.

We have also extended our state-of-the-art network verification capability to cover complex network topologies by modeling Virtual WAN, Virtual Networks, and other network function virtualizations (NFVs), as well as modeling reachability using formal methods. Using fast solvers, we can verify reachability constraints on customer topologies, at deployment time or when undergoing a config change.

We have applied machine learning to predict the impact of peering link outages and congestion mitigation strategies and use the data to improve the availability of the WAN peering surface area.

Our expertise in optimization algorithms has been shown to ultimately reduce cloud networking spend. Techniques like these will be invaluable in carving out 5G paths on the overlay that are cost-efficient, but still meet the performance needs of every network slice.

Figure 2: 5G WAN technical architecture. This is how we envision the different technical components will need to come together to serve 5G operators.

The significant upside for operators

To reiterate, Microsoft is heavily invested in running a well-managed, always-available global network. We have been incorporating multiple groundbreaking technologies, including scalable optimization, formal verification of routing policies, machine learning, and AI. We envision operators to not only be able to use our WAN to transfer 5G packets, with low latency, but also to benefit from multiple network services such as DDoS protection, firewalls, traffic accelerators, connection analytics, load balancers, and rate limiters, many of which we use in running existing Azure network workloads.

At Microsoft, we bring the full power of research and engineering leadership into our networks, rapidly incorporating innovation and new features to provide reliable, low-latency, low-cost service. In turn, this effort will open up the significant potential of next-generation services and applications as envisioned by the community at large. It is no understatement to say that collaboration between operators and Azure is key to unleashing the true power of 5G.

Source: microsoft.com