Much of the attention Huawei got at Mobile World Congress was for its smartphones displayed in Hall 3 of Barcelona’s Fira Gran Via. But what we care about is the other stuff Huawei was showing off in a cavernous, invitation-only booth in Hall 1.
“Booth” is actually the wrong word; big companies like Huawei and Ericsson rent out zones of the Gran Via to set up what are essentially miniature trade shows of their own. There, Huawei displayed all its ambitions for network infrastructure: LTE, mobile backhaul — and, of course, software-defined networking (SDN) and network functions virtualization (NFV).
This is the domain of Sanqi Li, CTO of Huawei’s carrier network business group. For about 40 minutes on Wednesday, in one of the small, overheated conference rooms inside Huawei’s MWC neighborhood, I got a chance to talk with him about Huawei’s approach to SDN and its take on how NFV relates to the cloud (something that groups like CloudNFV have been pushing heavily).
How important is OpenFlow to your SDN strategy?
Li: The definition of OpenFlow used to be just decoupling, which made a lot of sense in the data center. Then it moved to a hybrid model, where certain functions are taken out of the hardware devices and certain functions remain, and for the carrier, this is [appropriate]. The carrier is more looking for something that is not forklift, something that is more evolution than destruction.
We do see a lot of opportunities that are OpenFlow-like. The concept is good, but also, it is being extended in good ways. They need to address management; they need to do certain control functions — and abstraction of the networks, that’s crucial. We are looking more to help the carriers, step by step, in that direction.
Let’s talk a bit about your OpenFlow controller.
Li: SDN controller is what you mean, right? Because OpenFlow — it depends on the definitions…
SDN controller, right.
Li: What we believe is that there are data center SDN controllers, there are WAN SDN controllers — and furthermore, in the recent discussion on the architecture in the ONF, there was a service controller on top of the infrastructure controller. So, you can think that the controller will not be just a single-layer controller — the data-center controller — but there will be a controller overlayed on other controller functions. I think this will play out, for the carriers specifically.
On the other hand, there are still issues of the scalability of the SDN controller. Not just the interfaces northbound and southbound, which are the big focus right now, but also the eastbound/westbound.
Scaling in the east/west directions, you mean?
Li: Oh, yes. When you’ve built this distributed system, with a distributed architecture, the question is how you can keep the state in real time. Whenever the network fails — people yesterday on a panel talked about (snaps fingers) nanoseconds! I think that’s too much. (Laughs.) But 50 milliseconds would be in a good range for real time observations.
The challenge is — like in any distributed system, you need a strong consistency among these controllers. And if it’s not well defined, then there will be issues between vendors. These things are challenges, especially when you move to the carrier space, rather than just a single data center. Then it becomes [an issue] across regions, across carriers, across vendors.
I take it you’re working on these scalability issues?
Li: We are engaged, definitely.
Is there a standards effort involved, or is it a matter of just getting it to work?
Li: There are a lot of efforts in open-source technologies that we can leverage. A lot of things are happening in the cloud world and in the ecosystem that would be very good for our industry — not just ourselves.
Any particular examples?
Li: At this stage, I wouldn’t want to mention specifics. A lot of things are still in development that we have engaged, that Huawei are testing and monitoring. I think it’s important to leverage the technology in the cloud world.
You were talking about different controllers. Does this mean Huawei has different controllers for the data center and the WAN?
Li: From the major-requirements perspective, the data center controller and what I call wide-area network controllers (the carrier class of networks) are very different. Our goal is to have fundamentally the same framework which can apply [to all cases]. The data modeling can be different for different applications; the focuses can be different for different applications. A unified platform is our goal.
How close are you to that unified controller platform?
Li: Today, [our controllers] are separate but largely share the same framework — but it’s still evolving.
How long before you’ve actually got this unified framework?
Li: (Smiles.) Ah, that’s a hard question.
Not tomorrow, then.
Li: It’s not like a snapshot, because the requirements and the priorities are different — it’s continually evolving. So, it’s really hard to put a deadline [on it]. New requirements come in. It’s like I said, the controller concept is not just the network controller; it’s a service controller, or it’s a data-center controller — but it’s important to have a unified framework.
On the NFV side, you’ve introduced Cloud Edge. I take it this means your strategy is about the cloud and the edge network.
Li: The network complexity is at the edge — multiple generations of technology, and the move to IP and MPLS. If you look at operational headaches, the opex is also at the edge in network transformation. More importantly, the carrier differentiation, to differentiate from non-infrastructure players, is at the edge.
Therefore, we transform the network edge infrastructure to the cloud. We are in 20 PoCs [proof-of-concept demonstrations] and lab trials with major carriers.
What makes you think carriers are going to move quickly on any of this?
Li: The carrier is no longer as technology-driven as [it was] many years ago. It’s still [about the] infrastructure, but they’re largely business-driven, solving today’s issues. The issues are related to three things: infrastructure transformation; operational simplification and cutting capex; and also more services in general. The early deployments you see, like the virtual EPC [evolved packet core], service chaining, virtual set-top box — those are very much business-driven.
The carriers will identify these high priorities that also have less impact on the existing network infrastructure and operations. The network used to be just packet-in, packet-out. Now the edge has become much more like the data centers. It’s where a packet is cached and processed. Application awareness, DPI, firewall functions — all of these things will be very much in line with the edge, so this is where it’s a high priority.