Original Post Read More
It’s been 12-plus years since we embarked on the paradigm-shifting edge computing story, which brings the cloud closer to the source of data generation and consumption. Nowadays, the cloud provides resource-rich compute and storage capabilities, remote management, and new applications and services as latency continues to be reduced. Edge computing has gone mainstream, as evidenced by numerous conferences and workshops; thousands of research papers, mainstream media articles, Ph.D. theses; and many products, including those from Microsoft.
Years ago, an article we wrote stated that the killer application for edge computing was video analytics. The article, as published by IEEE, envisioned cameras and video located everywhere, increased ability to understand these video streams, and improved ability to react appropriately, stemming from real-time video analysis at the edge. Microsoft continues to believe that edge video analytics will be the dominant service for edge computing, just as we noted many years ago. Since then, we have evolved to an edge fabric, enabling ubiquitous computing. Here, the computing fabric is all around us in many different settings—working for us, improving efficiency, protecting us from problems, and entertaining us.
In this article, we focus on what’s next, including the topic of edge computing for telecommunications, which has been evolving into the next wave of innovation, and one we must embrace. Microsoft believes the telecom edge is the catalyst creating a new world where the telecom and cloud industries join forces to eliminate duplication while creating a new era of latency-sensitive applications and services.
Enabling private 5G Networks with Azure private multi-access edge compute (MEC)
A private 5G network is a local-area mobile network; technically, it is the same as a public wide-area 5G network. This next-generation network enables advanced use cases not supported by current mainstream Wi-Fi technologies. For example, private 5G networks can unify connectivity and support a variety of enterprise-specific secure IoT services and applications.
In June 2021, Microsoft unveiled a new product category for the telecom industry when we announced our Azure private multi-access edge compute (Azure PMEC) managed solution. Azure private MEC is a solution for modernizing enterprise networks, comprised of Azure Stack Edge, Azure Network Function Manager, first and third-party network functions, and manageability via Azure Arc. With it, carriers and ecosystem partners can easily and rapidly deploy and manage network functions like 5G mobile cores, radio access networking (RAN) solutions, and Software-Defined Wide Area Network (SD-WAN) products directly from Azure Marketplace. Our open platform solution empowers operators and system integrators (SIs) to unlock the private 5G opportunity by delivering managed, curated solutions to enterprises with the flexibility of first and third-party offerings, including their choice of RAN and latency-sensitive applications.
Many of us in the IT and telecom industries accept edge computing as a game-changing architectural innovation, reducing the time needed to process the packet after it is generated at the source. All edge computing products that exist today provide this, but Azure private MEC enables even more. With the emergence of novel software-only 5G implementations, edge computing is evolving to become an exciting part of the packet creation infrastructure.
Conflation of Virtual Radio Access Networks and edge computing
The figure below illustrates the shift away from specialized, monolithic network infrastructure to programmable, virtualized Radio Access Network (vRAN) elements. Virtualized RAN offers a cost-efficient solution for running the 5G RAN as a virtualized network function (VNF) on commodity hardware. To implement vRAN, telcos need a low-latency connection between their signal acquisition and computing hardware, necessitating edge computing to make vRAN possible.
It is possible to implement vRAN over a hierarchy of edge installations. In 3GPP RAN parlance, the distributed units (DU) that implement the near-real-time functionality of the RAN, which include physical layer processing (often referred to as L1) and medium access control (often referred to as L2/L3), are implemented at the “Far Edge.” The rest of the RAN stack, along with the network core, is implemented at the “Near Edge.” We have been working on providing this edge infrastructure to operators as part of Microsoft’s core offering.
Figure 1: RAN architectural evolution and innovations in 5G networks.
Despite this evolution in 5G networks, there is still more to do. When implementing RAN functionality at the Far and Near Edges, one has to decide how many server cores are needed to support a given number of cell sites. This type of problem is easy to solve. Microsoft computer scientists are able to determine the number of cores needed to serve the client device, and have further invented and developed algorithms and techniques to allow scaling, energy management, fault tolerance, and feature deployment in running systems. Note that server cores can be provisioned to both assist with packet generation and running applications and services.
In ACM SIGCOMM 2021, we published a paper entitled, Concordia: Teaching the 5G vRAN to Share Compute. As noted in this publication, one reason why vRAN is more efficient than traditional RANs is because it multiplexes several base station workloads on the same computer hardware. Although this multiplexing provides efficiency gains, more than 50 percent of the CPU cycles in typical vRAN settings still remain unused.
Here, co-locating the vRAN functionality with general-purpose workloads not only improves CPU utilization, but it also allows us to service low-latency applications on the same hardware. This is important since vRAN tasks have sub-millisecond latency requirements that have to be met 99.999 percent of the time—difficult to accomplish with existing systems.
Microsoft has also built a user space deadline scheduling framework for the vRAN. Our system includes prediction models using quantile decision trees to outline worst-case execution times of vRAN signal processing tasks. Running every 20 microseconds, the ultra-fast scheduler delivers accurate prediction models, enabling the system to reserve a minimum number of cores required for vRAN tasks while leaving the rest for general-purpose workloads. Evaluated on a commercial-grade reference vRAN platform, our design meets the 99.999 percent reliability requirements and reclaims more than 70 percent of idle CPU cycles without affecting RAN performance.
Edge computing was created jointly by Microsoft and our academic colleagues. Edge computing products have evolved, as we fine-tune solutions to new sets of problems we are solving. Beyond implementing 5G infrastructure on commodity hardware, our software takes advantage of the latest discoveries we’ve made in applying machine learning techniques to improve the performance of our edge nodes. We continue to work closely with our academic colleagues, and serve on the advisory board of two National Science Foundation (NSF)-funded Edge AI research centers (The Institute for Future Edge Networks and Distributed Intelligence and The Institute for Edge Computing Leveraging Next Generation Networks). Both research institutes focus on developing AI technologies as part of edge computing that leverages next-generation communications networks to provide previously impossible services.
The future is bright because we are on the right track with Azure private MEC. The architecture we are developing and the products we are delivering will make edge computing indispensable, as every packet in the mobile network will be processed by an edge node, leading to a large ubiquitous processing fabric, the likes of which we have never enjoyed before.
To learn more about our Azure for Operators strategy, refer to the Azure for Operators e-book.