The shift from purpose-built to white-box hardware with network functions enabled by software is building momentum, based on the presentations and demonstrations at the Carrier Network Virtualization conference. Featuring speakers and attendees from carriers and providers, such as AT&T, Google, NTT and Vodafone, this multi-day conference provided an excellent overview of the motivations and current status of the quickly evolving topic of Network Function Virtualization. In the above interview, Michael Howard, Senior Director of Carriers and Research for IHS, describes some of the highlights from this content-packed conference.

Two initial drivers of virtualization will be virtual CPEs and CORD (Central Office Re-architected as a Data center). While one advantage of virtualization is the potential of lower costs due to commodity hardware, the bigger advantage is the agility and time-to-market that software-based functions bring. That is, the same CPE or CO hardware can perform different functions, based on a given need at a given time without having to roll a truck. Further, virtualization of the network is about automation, speeding the fulfillment and improving the accuracy of a customer order.

Sam Aldrin, architect for Google, explained that software defined networking is about the decomposition of the control, management and data planes, which allows these paths to independently evolve. The result is a network that provides much greater up-time, flexibility and with the ability to roll-out of new functions measured in days, instead of months. Based on what Google has learned from rolling out SDN, Aldrin suggests that Software Defined Operations are necessary for SDN and NFV to fulfill their potential; that is, automation of the configuration and management planes.

Along the lines of simplified operations, Vodafone’s Dr. Walter Haeffner said the motivation for Vodafone’s move towards SDN and NFV was to collapse their fixed and mobile networks into a single network. They chose a proprietary hypervisor for initial implementation and will consider open source hypervisors as carrier-grade options become available, the SDN tools mature and Vodafone develops the necessary internal skill-sets to support open source. Their general rule of thumb is to virtualize the brains and not the muscles (that is the existing carrier-grade network equipment).

Like Google, they have found virtualization reduces implementation time for given functions. For instance, Haeffner indicated that it took only 3 or 4 days to overlay an analytics platform compared to ½ year for dedicated compute approach. The results of their virtualization approach is a projected reduction of capital expenditures of 25 to 45% over 5 years and 30 to 60% operating expense reduction over 3 years, respectively. 

This aligns with the comment made by Douglas Ranalli, founder and chief strategy officer of NetNumber, Inc. who indicated that, “Being able to virtualize allows every developer to work faster, as they don’t have to compete for lab space.” It is surprising to Ranalli that service providers haven’t also adopted NFV for lab development and testing. The lack of standards, which means NetNumber has to integrate with five or six orchestration systems, slows development and is also a barrier to and is slowing commercial roll-out, according to Ranalli.

An original “cloud-native” approach is not adequate for telecom networks, according to Haeffner, and he encouraged that engineers and developers think of “stateful solutions” when designing SDN/NFV. He calls this “extended-cloud-native” and it consists of Linux, plus microservices containers and stateful repositories. This allows for an Agile, DevOps approach, not possible with traditional telecom networks and opens up the possibility of implementing things like machine learning to improve network evolution and utilization.

Oracle Director of Product Marketing at Oracle Communications, Doug Tait believes that NFV still has to cross the chasm and that some of the things that must be overcome for mass adoption are paraphrased as follows:

• split of hardware network elements into software is difficult
• fragmentation of open source and standards
• onboarding of Virtual Network Functions (VNF) is difficult (e.g. as simple as an app store?)
• internal skillset gaps
• NFV architectures are vast and varied – there is no one size fits all
• requires defining the right KPI (Key Performance Indexes) and the applications must be able to provide these measures
• OSS/BSS APIs – are not standard, open or available
• Security is still very nascent in NFV architectures

He suggests three things to help resolve these challenges and assure wider spread deployment, which are:

• embrace DevOps, which will allow operators to “fail often”, while simplifying, accelerating and improving their operations
• move to microservices, which replaces user interfaces with APIs and breaks larger blocks of code into independent modules
• embrace hybrid NFV traditional telecom hardware models

To this last point, a recurring theme from the speakers at the Carrier Network Virtualization conference is that the transition to a completely software defined network where all network functions are virtualized will not occur overnight. Michael Howard reinforced this with his comment that he is projecting only about 10% of larger carrier, central offices will be CORD. Howard indicates that this year the deployed number of virtual CPE measures in the hundreds of units, while next year he is projecting 20,000 units and for it to take off there after.

Stay tuned for more video interviews from the Carrier Network Virtualization conference.