Cloud/SaaS/SDN, Forum

Network Virtualization 101

Nachman Shelef, CEO & Co-founder, ConteXtream – a BGV portfolio company shares his perspective on Network Virtualization   Q: What is your definition of network virtualization and does this definition apply for Carriers or Enterprises or both? A: Network Virtualization, like other virtualization transformations is all about transitioning from dedicated resources per function to shared, pooled, common resources for all functions. For compute and storage virtualization this meant transitioning from dedicated compute and storage resources per application to shared, pooled, common compute and storage resources, allocated as needed when needed to support each application. For network virtualization this means transitioning from dedicated (and in some cases with proprietary hardware) networking hardware per function to shared, pooled, common compute resources and basic switching resources, allocated as needed when needed to support each network function. Traditional networking makes this transition very difficult, with three specific barriers: a) High functions are built into the plumbing (hardware) and are applied hop by hop, link by link, port by port and sub-port by sub-port. b) Identities are tightly coupled to locations. c) Control and forwarding are tightly integrated. Key to enabling this transition is to separate basic connectivity between locations from the higher functions that enable carriers to provide services per flow based on the context of the particular flow. This is done by separating functions from junctions and separating locations from identities. And by adding separation of control from forwarding we complete the disaggregation of networking nodes to allow network virtualization. We believe that while the basic concept behind network virtualization is similar for both the carriers and enterprises market segments but the requirements are quite different. For example, the scaling requirements of carrier network virtualization are very different than those needed in an enterprise network. Similarly carrier network virtualization may also require subscriber awareness, which is not needed in the enterprise scenario. Q: Do you position yourself as a – network virtualization company, Software defined networking company, Network function virtualization company? A: We are focused on helping service providers virtualize their network, improve service agility, while decreasing CAPEX and OPEX.  We do this with our unique approach to carrier network SDN. There are other ways of virtualizing networks when going the NFV route, but our approach fully leverages the capabilities of SDN while creating a very flexible, scalable and programmable network. Using our solution, carriers can virtualize their network and it also gives them many network design options on how to scale their network functions. Q: What are the top 2-3 business problems that network virtualization solves? A: Based on our experience with the Tier-1 operators, network virtualization attacks the two biggest problems that carriers are facing:

  1. The first big business problem for any carrier is that traffic is growing more rapidly than the revenues. Operators have to keep adding capacity to cope with demand, but with existing network design paradigms and products the utilization of the resources is poor. Service providers have realized that the legacy methods of network design, deployment and the legacy products have reached their limits. The Internet players and advent of cloud technologies have shown that virtualization can help achieve better utilization of resources which is a major need of the day. This is driving the current interest in Network Function Virtualization. Service providers who adopt our solution will find that they can deploy capacity in both small and large chunks, they can be extremely granular, rapidly move around resources, achieve higher utilization etc.
  2. The carriers also need to compete with Over The Top players. They need to defend/increase their revenue stream with innovative services, and they need to do so with greater agility. Network virtualization with SDN creates a programmable network that significantly reduces the time needed to both experiment and/or roll-out new services.
Q: What are the different approaches for network virtualization and why did you select your approach? A: One approach, attempts to change as little as possible in transitioning to network virtualization. In this approach, big multi-function proprietary networking boxes (sometimes referred to as god-boxes) are replaced by big multi-function Virtual Network Functions in software. The cables that connected the big boxes are replaced by a simple virtual network that provides virtual cables. In this “fat-VNF & simple-Infrastructure” approach each VNF needs to itself provide for scaling through Virtual Machine elasticity, for distribution across locations, for sharing states between VMs, for load balancing, for service chaining of sub-functions and for subscriber awareness as needed. An alternative approach is to rely on the virtual network infrastructure to provide these capabilities (scaling through elasticity, distribution, sharing state, load balancing, service chaining and subscriber awareness) to all VNFs as needed. This approach enables simpler VNFs, faster development, more innovation, reuse of sub-functions for many VNFs, less vendor lock-in, best of breed for every function. Putting these common capabilities in the virtualization infrastructure creates a VNF isolation and abstraction layer that enables fast time to market with new functions that are not connected directly inline. Implementing and thoroughly testing these inline common capabilities of the infrastructure once is more reliable than implementing them separately in each NFV. In short – though we support both the “fat-VNF & simple-Infrastructure” approach and the “thin-VNF & smart-Infrastructure” approach, we highly recommend the 2nd in order to get more of the potential benefits of network virtualization. Q: What are top 5 criteria to use in evaluating network virtualization solutions? A: The top 5 criteria for solutions that support network virtualization that we have found are: 1)    Production deployment experience: Carriers look for solutions that are based on proven technology and are looking for examples of real live deployments.  The needs of the carriers are different from enterprise and most solutions cater to the enterprise. 2)    Overlay solution: To attain highest utilization of network functions, the network should be able to support the function anywhere but be available as one logical function. We believe that overlay is in fact more than support for an encapsulation technique and that for providing scalable network virtualization, a solution needs to provide a function distribution mapping system, which allows this location decoupling. In a true virtual network a function can be anywhere and that should not impact the performance of the network. 3)    Subscriber and application awareness in steering: This is a very unique requirement for network virtualization for carriers because carrier services are typically subscriber services and the best network design is only possible if the network has the best of breed function vendors available, can right size them and then steer subscribers to these functions, based on the end consumer service requirement. We see this increasingly in both fixed broadband and mobile networks. 4)    Application-aware load balancing functionality: The right-sized VNF network virtualization approach increases the number of instances of the functions which greatly increases the complexity of managing the load across these instances therefore the solution needs to support built-in application-aware load balancing. 5)    Standards compliance: This is something operators have always required and will need for interoperability. We are committed to this and in fact take a leadership role in this whenever appropriate. Q: What are the primary benefits that customers seek when they talk to you about network virtualization – COST/CAPEX reduction; Revenue Enhancement; Other? A: From our experience, the primary benefit the carriers are seeking is the optimization of their network. They are trying to reduce both the CAPEX and the OPEX needed to provide the services to the end-users. The other benefit that they are seeking is programmability of the network. Thus they are seeking ways in which they can reduce both the operating expense while increasing the speed with which innovation can be delivered. Consequently operators can introduce new services with greater agility. Q: Which use cases do you see taking traction and which ones are still in the definition phase? A: The use case that we see in deployment at mobile carriers is the domain providing value added services between the Enhanced Packet Core and external networks also called the Gi-LAN. Carriers are motivated to address it to reduce and consolidate the number of VAS middle-boxes and the sprawl of Gi/SGi proxies and purpose-built appliances. There are also monetization opportunities through more customized service-chains to process consumer and/or enterprise traffic. Another use case that is increasingly popular is driven by the adoption of Voice-over-LTE. This is driving virtualization of IP Multimedia Subsystem (IMS) and virtualization of Session Border Controller (SBC). We support all these use cases in a manner where there is no need to rip and replace existing functions with virtualized functions but rather virtualize by using both physical and virtual instances. Other use cases that we see in the medium term are around virtual EPC, vCustomer Premise Equipment, vRadio Access Network, vContent Delivery Network. These use cases are in definition phase because either they are very complex or significant changes can cause a catastrophic service disruption. For example: EPC is a complex monolithic system with multiple interfaces and while given the growth in traffic, advent of Machine 2 Machine etc. there is a definite need to virtualize and create an elastic EPC that is more programmable. But the reason it is in definition phase is because failure can cause a serious service disruption therefore the carrier needs to take an approach that will allow them to experiment without taking too much risk. We expect vEPC to move into field trials and eventually deployment in the next 2 years.