Should We Virtualize Functions or Virtualize Networks?

Last week I gave a keynote presentation at MEF and answered two questions that I’m commonly asked:

  1. What’s next after SD-WAN?
  2. What’s the relationship between SD-WAN and NFV?

If you’ve read my previous blogs, you can probably guess my answer to the first question. I believe the software-defined WAN must evolve into the self-driving WAN. By augmenting automation with machine learning and AI, we can build WANs that dynamically translate business intent into action, with central orchestration working in tandem with the WAN edge. For this blog, I will focus on answering the second question.

Network Function Virtualization of NFV has been a key initiative for service providers. The idea is that service providers can drive down costs and improve agility by replacing dedicated hardware appliances including routers, firewalls, load-balancers and wan optimizers with equivalent software (virtual network functions or VNFs) running on industry standard server hardware. This is undoubtedly progress and a move we have supported at Silver Peak. We introduced software versions of our classic WAN optimization appliances back in June of 2010. While VNFs can bring benefits to the service provider, how do they help the typical enterprise? I think it’s imperative for carriers to find a way to use all those compute resources to solve new problems for enterprises, rather than simply supporting conventional functions in a more efficient way.

In my keynote presentation at MEF, I coined a new term: “Customer Network Virtualization” or CNV, that I will explain here. SD-WAN is about virtualizing the wide area network by bonding together multiple underlying transport, and matching this underlying transport capacity with application needs on a packet-by-packet basis. Think of it this way, if the software-defined data center is about running any application on any server, then the software-defined WAN is about running any application over any combination of transport. By virtualizing the customer’s WAN, it’s possible to dynamically create individual WANs for each customer, tailored to each specific customer’s needs. Rather than delineating and virtualizing by function, what if we virtualized each customers’ network? This is what I mean by CNV. It has some important implications. First, if each of the carrier’s enterprise customers effectively has their own virtual network, each enterprise customer can run their own choice of networking software, and whatever version of that software best meets their unique requirements.

In contrast, traditional MPLS networks are built on “carrier grade” multitenancy where giant Provider Edge (PE) routers run software that supports hundreds of customers, using shared hardware to provide each enterprise customer with a virtual IP-VPN. However, because this is a big multi-tenant instance, software bugs can affect every customer.  And when there is an upgrade to address bugs or add features, it has the potential to adversely affect every customer as well as software upgrades. For this reason, the core router vendors and carriers must test this software for months, sometimes years, before making changes to their production networks.  Ultimately, this classic form of multi-tenancy has a crippling effect on agility and innovation.

Imagine instead, a world where each customer’s network is virtualized. Now, every customer can adopt new releases at their discretion. A conservative customer can elect to run a tried and true software release and forgo upgrades.  In contrast, a leading-edge customer can embrace the latest release along with all its new features. If a certain customer encounters a rare bug, the carrier can upgrade that customer’s virtual network without worrying that the fix release might adversely impact other customers. In this new world problems are resolved faster, there is less collateral damage, and every customer can reside where they want on the continuum between conservatism and leading-edge feature adoption. In a sense, CNV is synonymous with SD-WAN. It brings the ability for carriers to mass-customize networks in a way that’s never been possible. It uses compute virtualization to provide multi-tenancy, rather than doing it the conventional way, which while tried and true, is slow and painful, and unresponsive to the diverse and changing needs of the enterprise.

SD-WAN provides enterprises with the opportunity to take things into their own hands, by building and managing their own SD-WAN overlay. However, there are many enterprises that would love for service providers to do this on their behalf. I believe that carriers that adopt a customer-centric rather than function-centric approach to applying virtualization in the network are going to be able to keep pace with the changing needs of these enterprises in a way traditional carrier services have not. Indeed, we’ve entered exciting times as software and virtualization turn traditional networking upside-down!

About the author
David Hughes
David Hughes

David Hughes founded Silver Peak Systems in 2004 after serving a year as an Entrepreneur in Residence at Benchmark Capital. Through 2013 Hughes drove innovation serving as CTO, and then more recently as CEO, leading Silver Peak beyond WAN optimization into the emerging SD-WAN market. Prior to Silver Peak, Hughes served as vice president and general manager at BlueLeaf Networks (2000-2002), where his team developed a unique network switching and transmission system. From 1996 to 2000, Hughes held several positions at Cisco Systems, including director of system architecture for the BPX and MGX product lines, and senior director of product management for the Multi-Service Switching Business Unit. Earlier, Hughes was a key engineering contributor at StrataCom, an early pioneer in frame relay and ATM, which was acquired by Cisco in 1996. Before StrataCom, David worked as an engineer for BNR Japan/Northern Telecom Japan Inc.
Hughes has been awarded more than 50 patents in areas including data acceleration, routing and packet switching, control and scheduling algorithms. Hughes earned his PhD in Electrical and Computer Engineering from Wollongong University, Australia, and holds a BE in Electrical Engineering from Auckland University, New Zealand.