For the second installment in our NFV interview series, we spoke with Margaret Chiosi, distinguished technical architect at AT&T and a member of the network functions virtualization (NFV) group inside ETSI. We talked with her about the ETSI effort and the goals behind NFV.
Be sure to check out the series’ first installment, with Christos Kolias of Orange.
We hear both that NFV is moving very fast and getting traction, and yet also hear that it’ll take a while to get standards created — for example, what hypervisor standards, what orchestration standards aren’t defined yet. Do you see the NFV ISG moving quickly within ETSI? When do you think we’ll see early standards being published?
Chiosi: If you have ever been involved in forums, you always want things to move quicker, but you must remember that these are volunteer organizations. Overall, I think it is moving at a fast rate considering we kicked this off in January 2013 and we had to align vendors (about 100) and service providers (20-plus) on the who (chairs/rappateurs), what (focus), when (meetings/conf calls), how (organization of tracks and number) and where (physical location) of this topic as well as accommodate the growth in attendance (we hit 300 attendees in California). ETSI has done an excellent job at supporting this effort.
The goal is to have published, approved documents by the end of 2014. We are not a standards organization; rather the goal is to define the framework and requirements for a network virtualized platform. Due to this short timeline, we kicked off the parallel working group tracks, which can be somewhat chaotic. But parallelism can be effective in doing things quickly. We’ve been handling major issues as they come up, which in some cases have caused us to add more teams, which had a broader view to help the working groups. We are doing agile development.
Vendors are all piling on the NFV bandwagon, like they did with SDN. The NFV standard isn’t even close to publication and yet we already have NFV-compliant, NFV-compatible vendor claims. What do you think about that?
Chiosi: I think the main concept of putting network software on COTS [commodity, off-the-shelf] environment vs. a vendor-specific environment is understood, since we are building on top of the virtualization platforms that are in the data centers today. The main question is: Can we virtualize and not sacrifice performance or reliability?
So, there are certain functions which can easily move onto the data center/IT platforms and there will be others which may need a different platform which supports acceleration techniques for high throughput. So, for the ones which easily can move onto the IT platforms, claims of NFV compatibility can be made.
The forum is trying to define the requirements in one place. Some areas are well understood, and we are tweaking those. And then there are areas which are not understood well — e.g., orchestration and management from a network function point of view.
You were one of the founding members of the NFV initiative with ETSI. What did you see as so compelling about NFV that drove you to push and champion this initiative?
Chiosi: AT&T sees three value propositions from virtualization. First is cycle time — reduced cycle time for introduction of new services as well as reduced cycle time to remove old or poor services. Next is composition: Since it is software only, the composition or decomposition of functions allows us to be more flexible in responding to the market place as well as allowing us to create APIs much more quickly.
There’s also cost — we want to reduce total cost of ownership.
It is critical that the AT&T has one framework that the suppliers can build their products on and allow focused innovation to occur much more rapidly. The ETSI ISG NFV forum’s goal is to create that framework as well as to get customers, suppliers and academia together in one forum.
You have talked about SDN and NFV together in the past–what are your thoughts about how they play together?
Chiosi: They are complimentary yet solve different problems. One can implement SDN without virtualization, and one can implement virtualization beyond SDN functions. However, the combination of the two is powerful.
My view is that SDN’s goals are:
- To create a modular/layered design to allow innovation at the different modules/layers to be independent from each other,
- To separate the control plane from data plane,
- To provide a programmable network through a network abstraction layer for easier development of operations or business applications without understand the complexity of the different networks,
- And to provide the ability to create a central, real-time network weather map.
The goal of virtualization is to allow the software application to run over a COTS environment and allow the sharing of the COTS environment among multiple software applications (VMs). NFV is focused on virtualizing the switches and routers which support the Metro Ethernet Forum (MEF) E-LAN, E-Line, and E-Tree services.
Orchestration and management is a key part in realizing NFV. The MEF is focused on creating a standard API to perform FCAPS on Ethernet services for the industry.
Many dedicated hardware appliances today have certain performance characteristics like consistency and real-time capabilities that will be hard to replicate in the virtual world — do you see these problems being resolved soon? What other barriers do you see to successful NFV roll-out?
Chiosi: In the near term, not every network function will be virtualized. But as the technology improves to handle higher throughputs and the environment becomes more robust for network functions, more network functions will be feasible over this technology.
Management and orchestration definition, as well as products, are fundamental in accelerating the speed of NFV and therefore a major barrier to NFV roll-out.
You’ve spoken in the past that vendors might like NFV less than carriers. What does NFV mean for the vendors of carrier-grade dedicated appliances, like routing, DPI, firewall and also other L4-7 functions? Do you think we’ll see profit margin reduction for vendors?
Chiosi: I’d refer you back to the opening remarks on the value of virtualization.
Will you expect to see topology changes in infrastructure deployment? For instance, will AT&T be driving server farms or integrated chassis from its datacenter into POPs or more regionalized deployments to provide for NFV platforms closer to the edge?
Chiosi: Combinations are likely, depending on performance and geo-redundancy requirements.