Yes, I’m crazy enough to get up early on a Sunday to go to an SDN conference. I won’t say that there was a lot of “new news” in this session yesterday, but it certainly validated a number of hypotheses. I wound up taking many pages of notes even so, and I’ll summarize the main themes and highlights here.
Open Source and Standards Bodies
While it shouldn’t be a surprise at the Open Networking Summit to hear attendees and speakers beat the open source drum, it was interesting how completely dismissive of traditional vendor implementations most of the speakers were. Prodip Sen of Verizon, also Chair of the ETSI NFV group, commented matter-of-factly,
“Open source is the new standards body. Equipment manufacturers need to build openness, interoperability and modularity into their products. We’re not interested in vertically integrated stacks—that day is gone.”
I could caveat that by noting that even the “enterprise” speakers (from Ebay/Paypal) were rather Web 2.0-flavored and so not completely representative of the broad mass of non-tech enterprises and their needs, but the general direction is pretty unmistakable.
That said, Dan Pitt (ONF Chair) did offer a bit of pushback at one point on the dangers of overfocusing on modularity at this point. He pointed out that modularity comes with the cost a lot of effort expended on building abstractions for interoperability between modules, vs further developing core functionality. At this stage in the game, Pitt feels driving adoption is the most important thing to get SDN off the ground, with a solid user-dev feedback loop in place to refine SDN development as it goes along.
Pitt took a similarly measured view on the role and timing of standards development. He described the OSI stack as an hourglass, pointing out that standardization at the “narrow waist” had actually facilitated more options above and below it because all can leverage the common TCP/IP platform:
In Pitt’s view, most would-be standardization impulses in the SDN space are premature. For example, although the ONF has a Northbound Interfaces Working Group, he emphasized that it’s far too soon to drive a single standard implementation, pointing to the multiple NB API projects OpenDaylight has and stating that the ONF would be observing what ODL users learned in trying out the different APIs.
This led to a discussion of what area of SDN would be the “narrow waist”—OpenFlow? The Service Abstraction Layer? Multiple things? Neela Jacques of the OpenDaylight Project commented that there might be multiple “narrow waists” in different parts of the ecosystem, to which Pitt reiterated that this is something that will be determined by users and network operators. Sen had a more subtle take:
“Not any one of these is The Answer. It’s about the reference model, which will dictate all of the necessary abstractions.”
Use Cases and Technology Trends
Roy Chua (@WireRoy) of SDNCentral gave a really good presentation on this topic, and I highly recommend downloading his slides once they’re posted. In the meantime, since he was kind enough to send them to me last night, here are two of the most interesting ones:
Not too many surprises in that one, though I don’t think I’ve seen a more concise summary. On the white box front, Roy was somewhat cautious: on the one hand, there are active PoCs in the largest data centers. On the other hand, those same entities are ones who are most likely to do their stack development largely in-house, so the market opportunity is fairly murky.
In this one, I thought the concept of an “ODL-Lite” was interesting…and wondered if it could be offered by the OpenDaylight Project at some point in the future once the core architecture is worked out. Roy also made the point that while controllers are certainly a strategic element of SDN, they have thus far proven hard to monetize. Which is another solid argument for ODL as a rallying point—the narrow waist above and below which money-making innovation can happen.
The “Applications” portion of this slide was mostly centered around SDN applications. Roy commented that we haven’t yet seen a lot of development in this area—there was a longer conversation around this at the end of the day which I describe below.
The session featured speakers from carriers, a Web 2.0ish enterprise, and a cloud provider (Azure) who have implemented OpenFlow-based SDN. The drivers were quite consistent across all groups:
- Service agility. Especially with the ability to mass-customize and mass-optimize offerings to users.
- Efficiency. Network simplification for elastic scale, better utilization, and op-ex relief. All were very explicit that Cap-Ex reduction via cheaper hardware was a one-time thing that wouldn’t do that much for them. And revenue via service agility and differentiation trumped Op-Ex and Cap-Ex for service providers.
- Embedded tools to monitor usage for chargeback.
- NFV and soft switches. With some variation, the general architecture was L2 in a vSwitch (Ebay mentioned several times that their V:P switch ratio is 50:1) and L3 at top of rack. Azure doesn’t do much service chaining; they just use routing instead. In all cases, there was a mix of physical and virtualized L4-7 services depending upon place in the network; some of this is driven by regulatory/security zoning requirements. Azure also made the comment that each service type has its own controller, optimized for that service, eg one for load balancers, one for firewalls, etc.
In the subsequent analyst panel, Rohit Mehra of IDC noted that one thing slowing Enterprise adoption is that the complexity of the SDN space has increased exponentially over the last year. There are now multiple decision points, all in a high state of flux: multiple commercial and open-source controllers, all relatively immature; northbound and southbound interface models; overlay vs underlay vs hybrid; bare-metal and open networking OS’s; and so on.
The Role of Hardware in SDN
This wound up being the focus of much of the discussion of the last session of the day, which featured a panel of financial and industry analysts.
Amitabh Passi of UBS pressed vendors to begin thinking about new business models for SDN. 80-90% of network R & D is in software, but the sales model is hardware-driven. Rationalization is in order. Vendors also need to understand that users are moving to an Op-Ex-driven operating model and package their offerings accordingly.
Eric Hanselman of 451 argued that specialized hardware with specific optimizations would continue to be needed at transition points (DC and WAN edges, for example), where traffic volume is highest and scalability is needed to maintain performance.
Hanselman also pointed to the quiet movements Oracle is making in SDN as a possible harbinger of giving enterprise applications more influence over network activity.
This morning the conference started with hands-on sessions covering OpenDaylight and Floodlight controllers. I attended several sessions covering many of the same topics at OpenDaylight Summit last month, so skipped this morning’s sessions. I’ll be attending the Technology track sessions this afternoon.