What's the Best Way to Virtualize the Network Edge?

Posted by Mae Kowalke on Wednesday, June 22, 2016 with No comments

Virtual CPE is one of the hottest early markets for network functions virtualization, but the approaches to implementing it vary from putting everything in the cloud to a blend of on-premises and centralized resources. What are the costs and ROI of each strategy and how are they being used today? At Big Communications Event last month, Accedian's VP of Strategic Marketing, Scott Sumner, was on a panel exploring those and related questions. Some of his points are summarized below; the full panel video is also now available for a deep look at virtualization at the network edge.

What is the "network edge"?

Sumner pointed out an important distinction to keep in mind: 'network edge' vs. 'service edge'. For example, a cloud service might be isolated to one or two locations at the physical edge, yet have a multiplicity of service endpoints. Even in a single wiring closet, there may be network layer physical demarcation that uses a different strategy than the virtualized functions themselves. 

Given this distinction, virtualizing functions is the higher-ROI action, because it means replacing many physical boxes with a (potentially) simpler and less costly software stack. 

How and where is the network being virtualized?

In a lot of networks, virtualization is being done in a piecemeal fashion. During 2015, commercial deployments began to take off in the sense of the network edge being combined with virtual network functions, using technology like smart SFPs.   

When it comes to business services, such deployments are being rolled out based on the size of the customer: SMBs are an easy target, whereas enterprises are a harder sell (for now). 

"In the SMB market, it's very easy for operators to make the move to virtualized services, because it follows the residential model," Sumner explained. "For example, it's easy to put a firewall in the cloud and offer that to an SMB as a managed service. The customer will accept that because they have very simple needs for managed services at the edge."

Delivering virtualized services to enterprises is more complicated, though that is coming too. For the time being, the providers embracing this service model are those with the vision and market positioning to go out and do it (not necessarily dependent on operator size). 

How is the business case for virtualized edge playing out?

A key dilemma that's emerged from virtualization deployments during 2015 is that there's a mismatch between Moore's Law in terms of compute and the actual demands of real-time traffic processing. Operators learned (some the hard way) that x86 platforms deployed at the customer premises quickly resulted in bottlenecks, because CPU couldn't keep up with new applications. 

"According to Moore’s Law, compute capacity is increasing about 60% per year," Sumner noted. "And business services consumption of broadband is growing at 50% a year. On the surface, it appears x86 is keeping ahead of packet processing requirements. But in practice, that's not true. Over the past three years, business users have shifted 60+ percent of their workload from private to public or hybrid cloud. That means the amount of data running through that edge box or x86 is experiencing a much higher exponential growth rate than Moore's Law can actually handle. x86 is static, captive capacity; it is finite and will run out. It can't grow fast enough to keep up unless scaled into data centers." 

What are the operational challenges with a virtualized network edge?

One thing operators need to avoid is reinventing the wheel when it comes to operational procedures. Things like turn-up testing, monitoring, and remote troubleshooting have been around for a decade; operators have worked out automated, rapid procedures to address these aspects of operations. There's no need to start from scratch converting all aspects of operations to a virtualized world. 

This necessarily means that, "rather than just making everything into an on-board VNF, you also virtualize solutions at the network level, allowing them to easily talk to legacy systems," Sumner stressed. "Operational support systems should be able to operate as they did before, and technicians shouldn't have to change their practices."

To make all this efficient, it helps to innovate with solutions like a controller that sees all endpointsaka virtual NIDsas ports of one large device, and don't have to 'talk to' each individual endpoint separately.

"Using cloud scale to manage services, do testing, and collect monitoring metrics, means operators can focus on VNFs that are going to being new business and new revenue," Sumner concluded.