The Journey from NFV to uCPE
Virtualization, a mainstay of enterprise IT, has also been adopted by telco IT departments, hosting mainly back-office enterprise applications. However, it wasn't until 2011 that telcos embarked on the journey to virtualize their operational network as part of the Network Functions Virtualization (NFV) initiative, with ETSI as the organizing body.
As we will discuss, NFV ran into issues when it began, evolving in fits and starts but never becoming a breakout movement. Nevertheless, ten years later, we see the fruits of that struggle. This post will examine the NFV journey and how today's universal customer premises equipment (uCPE) sidesteps past challenges as it brings about the virtualization of the enterprise edge.
Early NFV Challenges
The telcos working on NFV aligned around a high-level architecture consisting of virtualized network functions (VNF) hosted on an infrastructure platform (NFVI), managed by VNF managers (VNFM) and orchestrated by the NFV Orchestrator (NFV-O) function.
NFV promised many advantages: agility, resiliency, scalability, and an automation-friendliness not possible with physical hardware systems. Telcos looked to create restorable snapshots of running VNFs, moving them around when needed, and spinning up VNFs as the load increased. They dreamed of achieving this without racking and stacking physical appliances or running new Ethernet cables.
However, early telco initiatives to build a unified common NFVI—often powered by virtualized infrastructure manager (VIM) OpenStack—proved challenging. Plagued by the inconsistency between vendor implementations, and more importantly, a lack of incentives for many network equipment providers (NEP) to undercut the rich margins they were making with proprietary platforms, NFV stalled.
In addition, carriers had taken on too much at once and were surprised by the level of complexity; simply wrapping a physical appliance into a VM wasn't enough. Performance challenges, confusion around I/O optimization, memory management, lack of consistent onboarding processes, incompatible interfaces, and different schemas impeded progress.
The community trudged on and stood up NFV clusters as distinct virtualized silos—one for easy-to-virtualize mobile core functions like IP multimedia subsystem (IMS) and eventually the 4G evolved packet core (EPC), another for virtual customer premises equipment (vCPE) and the virtualized WAN, yet another at the central office with the virtual broadband network gateway (vBNG). The issue was the lack of a unified platform managed from a single console.
SD-WAN and the uCPE are like PB&J (peanut butter and jelly)
Fast forward a few years, and while vCPEs struggled to gain widespread traction, enterprises pushed forward on WAN modernization initiatives. Driven by the need for increased bandwidth at lower costs, increased resiliency, and an explosion of many different cloud-based applications, CIOs sought a new approach different from legacy branch routers and firewalls. Software-defined wide area network (SD-WAN) emerged, promising new capabilities while aggregating multiple enterprise edge functions, including routing, multi-link management, edge security and firewalls, WAN optimization, and sometimes wireless LAN (WLAN).
SD-WAN exploded, multiplying and resulting in multiple billions of mergers and acquisition activities to date. Across analyst projections from IDC, Gartner, and Dell'Oro, SD-WAN (and related categories) is expected to see 20-30% compounded annual growth rates (CAGR) in the next few years.
In parallel, carriers jumping on the SD-WAN bandwagon and seeing strong managed service provider (MSP) revenue from enterprises are looking at reinvigorating CPE virtualization. The uCPE was a major NFV initiative, a single platform at customer premises that could host all manner of VNFs (security, unified communications, routing, WAN optimization).
Today, our conversation and work with carriers demonstrates a renewed interest in the uCPE as a platform for SD-WAN. Carriers are also showing increasing interest in Secure Access Service Edge (SASE), an evolution of SD-WAN that pulls in cloud-hosted enterprise edge security services. The uCPE is viewed as an on-premises component that could enhance SASE, too.
NFV is Dead, Long Live NFV or What's Different this Time?
While we don't talk about NFV as prominently in the context of uCPE, SD-WAN, and SASE, the principles from NFV carry over to the uCPE. Today, the community understanding behind NFV has matured. We understand how to build platforms that marry virtual machines (VMs) with Linux containers. Those who continue to use OpenStack know how to keep it lean and fold in containers on VMs. Others who bet on Kubernetes as the platform know how to leverage KVM and KubeVirt technology to handle VMs.
From AvidThink's viewpoint, the most important thing is not the technology. SD-WAN (and SASE) avoids previous NFV challenges because SD-WAN solves an obvious and immediate problem, with significant revenue dollars available. Plus, SD-WAN and SASE is a constrained domain around the WAN and on-premises footprint, making it tenable and feasible to focus resources.
Under the umbrella of a positive business case with straightforward returns, internal teams at carriers can justify infrastructure building and pre-investment with uCPE. The uCPE can lay the foundation by using SD-WAN/SASE as the delivery vehicle (the “foot-in-the-door”), opening up the option for other VNFs in the future.
We see the uCPE as a flexible edge platform at customer premises and becoming an enterprise edge site equipped with secure and reliable connectivity. We already see carriers launching virtual enterprise application environments on these uCPEs that can run different workloads and expect this trend to continue. In the next few years, uCPEs could host key components of a variety of workloads, including unified communications, IoT, 4G LTE and 5G private networks, manufacturing systems, and video surveillance. By running latency-sensitive or high data consumption workloads locally, uCPEs can improve application quality of experience while lowering costs and maintaining data compliance. At that point, we'll finally achieve our NFV dreams, but by then, we'll probably call it edge computing.
