The CPE market is often touted as the first real business case for network functions virtualization (NFV), but that doesn't mean the issues around distributed NFV are settled -- far from it, in fact. For one thing, the debate over what gets virtualized on premises and what is hosted in the network continues to rage among CPE vendors. And, as one leading analyst points out, distributed CPE requires a well-defined network orchestration strategy, and that's not something most CPE vendors have, on their own or through partnerships.
The attraction to virtualized CPE devices is an obvious one: Carriers today spend billions deploying and supporting a wide range of customer premises-based devices and see immediate value in being able to replace that cacophony of boxes with x86 hardware running virtualized functions such as load-balancing, firewalls, deep packet inspection and more as part of an NFV regime. Being able to sell new features and spin them up via distributed software -- versus rolling a truck so a technician can install a new box -- is the telecom version of a no-brainer. (See Virtual CPE: An Early Mover for Virtual Networks.)
But what functions need to be distributed to that premises box and which ones can be virtualized as part of a network cloud is still a matter of debate. RAD Data Communications Ltd. was the first vendor to publicly push a distributed NFV strategy and has continued to be its major advocate, while both MRV Communications Inc. (Nasdaq: MRVC) and Accedian are advancing more measured approaches to distributed NFV. Overture Networks, which today announced a virtual CPE product for Carrier Ethernet, as well as network orchestration software, is actually offering a range of options for what is virtualized and what isn't. (See RAD Pushes Distributed NFV Forward and Overture Builds on NFV Foundation.)
The new Overture Ensemble Carrier Ethernet device brings Carrier Ethernet demarcation into the NFV realm, as a virtualized network function on a server that can support other Layer 3 VNFs as well.
"The hottest short-term initial deployment and where we are most engaged in trials is seeing things virtualized at the edge of the network," says Prayson Pate, Overture CTO. Being able to program an on-premises device with new features means service providers can not only sell things more easily and cost effectively but even do promotions such as 30-day free trials that would be cost prohibitive if they involved deployment of a device at each customer's site, he notes.
But Overture will support virtualization of the CPE regardless of whether it is distributed, hosted at the edge of the network or in a more central location, Pate says. "The power of NFV is breaking apart the software from the underlying hardware which lets you build functions and put them where it makes the most sense, based on the service, underlying resources, latency and bandwidth," he says. "So we want to offer flexibility in doing that."
MRV is also supportive of some functions being virtualized on premises, but those tend to be "higher level functions" such as DPI, firewalls and load-balancing, says Zeev Draer, VP of strategic marketing. Performance issues continue to limit how networking functions can be handled as software-controlled functions. He continued:
Obviously some functions can be done but some functions are supposed to still be done in ASICs -- when you have wire speed, it can be tricky. When there will be technological progress in terms of cost structure, power consumption and the performance, that will be the time it will be realistic to do this as distributed NFV. Right now, not only MRV but everybody is trying to add this small-sized CPU in their legacy devices, this is the typical trend of NFV from the vendor perspective.
Veteran industry analyst Tom Nolle, president of CIMI Corp. , also sees vendors rushing to add distributed NFV to their checklists, to join in the marketing hype around the trend.
"There is a polar question that no one addresses -- is it NFV that gets distributed or something that is mostly distributed and gets to be NFV?" he notes. "Putting virtual functions on the customer premises does address some issues that could arise, particularly for the class of services that are involved in service chaining."
But distributed NFV needs to be part of a broader NFV strategy that addresses issues such as orchestration of all these distributed functions, Nolle says. "You need to have an NFV strategy" and not just add a board or feature to existing CPE that appears like an OpenStack server for hosting VNFs, he says.
He credits Overture with developing its orchestration play, announced today along with the new ECE and dubbed Ensemble Service Intelligence.
Nolle also cautions that excitement around using NFV-based CPE as a means of driving down costs and making services more flexible may also be over-hyped given the reality of how widely deployed Carrier Ethernet and other CPE devices are today. Replacing those devices has a cost and needs cost justification, he says, and that's not easily done.
The edge, not the prem
Metro Ethernet gear-maker Accedian Networks is backing a version of virtual CPE that also recognizes many of the features now delivered on prem can be done more reliably and cost effectively at the edge of the network with smaller devices with limited intelligence -- auto-discovery, for example -- on the customer's premises.
"You can extract everything you can from the edge including the demark but you need something [on premises] there for visualization and OAM. That could be hardware or software, a small module or software agent that can be controlled by virtual network functions," says Scott Sumner, VP of solutions development and marketing. Even functions such as firewalls can be effectively virtualized from within the network in a way that is more cost effective than putting it on prem, in many cases but not all, he concedes. "There will be a blend of things at different sites because of their requirements."
Determining that blend may be a carrier-by-carrier or situation-by-situation exercise that each service provider must tackle. In its efforts, RAD has been viewing distributed NFV as a starting point -- a place where network operators can begin to see some advantages of virtualizing network functions without dramatically changing how things are done on premises, other than being able to replace a large number of disparate devices with a single box that includes virtualized functions.
All of the vendors involved to date have continued to stress some basics of data networking: the need to do end-to-end service assurance, for example, and visibility for both the service provider and the customer.
The upcoming year should provide a clear indication of how far network operators are willing to go with a virtual CPE strategy and what functions they will choose to distribute to those endpoints, as well as how they will orchestrate the whole lot.
— Carol Wilson, Editor-at-Large, Light Reading