This Industry Viewpoint was contributed by Dr. Yuri Gittik, Head of Strategic Developments and Innovation, RAD
Network functions virtualization (NFV) is all about agility and efficiency, with the objective of automating and speeding service delivery with a streamlined cost structure for service providers. But where in the network can NFV best deliver on its promise?
Is centralization (the cloud) the answer, or should functions be distributed based on where they make the most functional and economic sense for a given enterprise customer?
These questions are beginning to be answered with Virtual CPE, or vCPE, a prime use case conceived as a virtualized networking appliance at the customer edge that delivers communication services to enterprises.
What once was a collection of single-purpose, hardware-based devices at each customer location (such as a router, WAN optimization, or firewall) has been transformed into virtualized appliances that can be dynamically added or dropped as needed.
When deploying vCPEs, there are two equally valid evolutionary options that allow a provider to tailor the deployment to the customer’s needs.
One option is to start with an emphasis on physical equipment at the customer site, along with cloud-based VNFs (virtualized network functions), and then add virtualization. The other is to start by emphasizing the customer-located VNFs and then add hardware-based functions as appropriate. Of course, it’s possible to implement both simultaneously as well.
With vCPE, at least some of the networking functionality associated with conventional CPE is virtualized and relocated. While virtualization facilitates relocating some functionalities from the customer premises to data centers, computational power resident in the vCPE allows maintaining some VNFs at the site or even relocating other functionalities from deep in the network to the customer premises.
Physical and virtualized vCPE functionalities are divided between the customer site and the network, (either the provider edge or the local/central data center), to ensure maximum flexibility and performance. Network-located functionality can also be shared among multiple users, following a multi-tenant unit model.
There is now a consensus that vCPE is deployable over a range of implementation options, with and without virtualization at the customer site. Major vCPE use cases are now better understood; these include IP virtual private networks, Ethernet services with virtualized capabilities and software-defined wide area networks (SD-WAN) as a carrier service.
There are three types of architecture that can be employed to introduce virtualization:
- With only a physical CPE (pCPE, without virtualization capabilities) at customer sites while all virtualization is located in the network/cloud.
- With a universal CPE (uCPE, enabling hosted virtualized functions) at customer sites while virtualization is distributed between the network/cloud and customer sites.
- With uCPE at customer sites with all virtualization at those sites.
These different architectures make it clear that vCPE deployment is anything but a cookie-cutter process. These three options cater to different enterprise needs, and all three can co-exist within a single customer network if needed.
Where to Begin Deployment
Regardless of the architecture, there are two scenarios for launching vCPE deployment. First, to start virtualization at the customer site as the initial stage, using a pCPE that provides tunneling and security. This is the best approach for the architecture in which there is only physical CPE at customer sites while all virtualization is located in the network/cloud. Then virtualization can be added as desired at the customer site using a standalone “white box,” or integrated with the pCPE.
Second, to start with colocation of a separate white box server and existing CPEs at the customer site, and then, as a next stage, collapse them into a single device to maximize performance and reduce costs.
This scenario begins with a white box (basically, a commercial, off-the-shelf system), which runs the vendor’s operating system or the carrier’s own software to host virtualized functions (VNFs). Later on, the white box could be enhanced with hardware-based functionality, such as performance acceleration, switching or routing, Layer 2 or 3 demarcation, or other physical network functions, in addition to VNFs.
This option is suited to the architectures in which there is universal CPE at the customer sites with virtualization either entirely at the customer sites or distributed between the network/cloud and customer sites.
Each of these scenarios is dependent upon technical and business drivers such as cost structure, targeted services, network architecture, use cases and other parameters. These considerations will determine where VNFs will be situated. Some applications are best suited to the customer premises or may have to be located there for the sake of operational and transport issues or governmental regulations or policies related to data protection.
Five Service Quality Considerations
Where virtualized functions are located may have particular effects on service quality. Here are five key factors to consider:
- Bandwidth efficiency. What is the bandwidth “cost” of moving functionality deeper into the network? Excess bandwidth expenditures could have a critical effect on service delivery in areas still served by relatively low-speed connections such as DSL.
- Security. Does moving the virtualized function to the network expose sensitive end-user data? For example, an encryption application located anywhere but at the customer site lacks adequate protection, as traffic interception can occur enroute in an unsecure access segment.
- Survivability. Critical functions must remain operative, even when the access link is down. Hosting IP PBX or router functions at the provider edge or data center would mean an inability to locally place calls or deliver traffic in case of network failure.
- Application performance. Is the delay added by the network for data center-based functions acceptable? This is a critical factor when engineering delay-sensitive workflows such as financial trading applications or considering a firewall that could be affected by frequent session timeouts due to packet loss and reordering.
- Diagnostics and QoE. Testing and troubleshooting applications need to accurately measure link and end-to-end service quality, as well as localize faults, starting from service handoff. When this function resides at the data center, it cannot reliably distinguish between performance issues arising from faults at the access link, traffic handling lapses, or user traffic impairment. That complicates solving the problem and adversely affects QoE.
What Industry Watchers Anticipate
According to Doyle Research, “vCPE is one of the largest drivers of NFV deployment and will account for more than 20 percent of NFV-related services by 2018.” The reason is that providers “are leveraging vCPE to rapidly deliver new services —e.g., SD-WAN and security — to allow customers to easily adjust their services mix.”
Analysys Mason observes that “enterprise vCPE-enabled services represent a significant opportunity … in the business network services market … through agile, on-demand service delivery and customer self- provisioning capabilities.” The organization also noted that vCPE gives providers a scalable, automated NFV and SDN-based solution for rapid deployment of both existing and new revenue-generating services. Yet they can keep their own operational costs down by replacing expensive, stranded physical network appliances with reusable, virtualized off-the-shelf servers and reducing inefficient manual operations.
IHS Markit tracks spending on NFV hardware and software for delivery of software-based services to customers based on vCPE and enterprise vCPE use cases. Its prediction: By 2020, the service provider NFV market will grow at a robust compound annual growth rate of 42 percent to $15.5 billion.
For service providers, vCPE’s sweet spot lies in hardware abstraction and the ability to carry out shorter and more flexible deployment cycles for new services. As the programmable network becomes a reality, the spotlight turns to automation and control. Over time, vCPEs are expected to transform from loosely coupled to integrated entities, with management functionalities increasingly becoming part of a dynamic control plane.