The theme of today’s symposium is Applied OpenFlow, but so much of the discussion about OpenFlow is about its promise. I’m interested in bridging the gap between those two pictures.
Very quickly, here’s what I’ll be coveringSome things seem to be settled about OpenFlow, we might call it conventional wisdom. But in OpenFlow, even that sometimes feels unconventional. In addition, some of the so-called benefits are arguments in favor seem to cut both ways. We’ll visit a few with Good News / Bad News.Then I’ll talk about expectations, which I believe is useful. Maybe partly “CYA”, but also keeps you (and me) from going crazy. And it’s not just about our own expectations, but also about others in your organization.And, seems like we should talk about what is actually possible to DOLastly, I’ll talk about what Brocade’s been doing. We don’t share the gory details (except under NDA), but I can say a few things.
[CLICK?]One of the founders of the OpenFlow movement, UC Berkeley’s Dr Scott Shenker, made a statement about OpenFlow that has often been quoted. “OF doesn’t let you do anything with your network that you couldn’t do before.” That kind of makes everyone perk up. Did I hear that right? If it’s more of the same, what’s the point, right? In just about the next sentence he acknowledges that there is something new, which is a new API.What that means is that if you’re a independent software developer, you actually can do something new with a network that you couldn’t do before. And once the coder has done something, then maybe a network operator can do something new. But if you’re a switch developer, no, OF does not enable the switch to do anything new.[CLICK]On the other end of the spectrum we hear that OF enables everything. Sliced bread, dishes, cleans windows, etc. And yet, OF 1.0 and 1.1, the officially released version of OF protocol, are extremely basic. So when folks say “OF enables nirvana” they’re really talking about the concept of SDN which OF hopes, over time, to deliver.But even today, OpenFlow changes the dynamic for developing networking algorithms. In these early days, that’s more true for the soft switches, but it will also shift gears for hardware switches as well. Decoupling the software development funding will allow for a burst of innovation.
Okay, so how about some of those “benefits” about OpenFlow seem oddly like landmines?[CLICK]For example, I often here statements that sound like “You can do whatever you want!” Yeah, that’s sounds a wee bit scary to me. Of course, savvy people with tricky problems prefer the freedom and will do their own testing and add their own safeguards.[CLICK]Another supposedly cool thing is the vast array of interesting applications. In a way, that is cool, because it does help justify the work. But this comes at cost, which is trying to agree on features and interoperability, etc. More on this in bit.[CLICK]One more item of good news is that we can try modeling networking similar to the way we have modeled computing for years. Applications on top of an OS on top of a common hardware framework with a common CPU. This notion kind of works with OpenFlow, and even the idea of a hypervisor can be inserted. And yet this also highlights the differences between networking and computing. The networking stack components are not (at least not yet) as de facto standardized as they were for much of the development of the computing paradigm, and early apps were not simultaneously running on many separate (and different) pieces of hardware. In a way, the fact that we can see this mismatch ought to be good news, as long as we stay conscious of where the models diverge.
Now I’d like to set some expectations about OpenFlow. There are so many variables, it’s hard to make firm predictions, but I think we can make some, or at least “expect” certain things. Maybe we’ll be surprised. Let’s touch quickly on some of what we know.First, we know that the released versions of the spec are very basic, and lots of important bits are still missing. You may or may not already be aware of that. If you’ve seem the demo videos, you may wonder how they could work. Many are leveraging pre-standard implementations, or they use extensions. They are still really valuable proofs-of-concept, but they’re a bit ahead of the OpenFlow curve.We also know that several key features have been prioritized and the ONF workgroups have formed and are eagerly wrestling with new features. We can “expect” that they will arrive in the coming 12 months; many members will have advance runway in building toward these standards, so we might expect some real products to include these sooner than a full development cycle after the standards are released.Of course, there will still be plenty missing, and we can expect a steady stream of stuff to find its way into the standards
Let’s take a step back and look at some implications, essentially 2nd degree expectations.[CLICK]In the near term, we can expect that interoperability will be iffy for a while until we get a lot more stuff disambiguated in the standard, and we have some independent certification. Until that happens, providers will need to partner in order to deliver robust solutions that are ready for prime time. So don’t expect mix-and-match real soon. You may see you killer app in a demo video, but be unable to buy it. It may be that in the near term, OpenFlow turns out to influence your vendor choice, even if it doesn’t guide your product choice. Also, despite some claims that OF will drive down hardware costs, that’s not likely for a while. It’s possible that you’ll be able to solve your problem in a cheaper way with OF, but don’t look for openflow to lower your capex for quote some time.[CLICK]In the Medium term we’ll get certification. Basic apps will be interoperable, but a lot of interesting apps will depend on advanced functionality , and that means extensions, and that means partnering will still be common. Also, there will be a good variety of applications, and vendors will tend to specialize (true in the near term as well)[CLICK]Longer term we’ll see more interop and some consolidation and broad coverage. At this point, we may see more competition which will eventually result in price pressure.
Finally we arrive at the slide about actually doing stuff. First, there’s no one-size-fits-all option for engaging with OpenFlow. What makes sense depends on your context. There are a million ways to categorize contexts, like type or size of firm, but I decided to view across two dimensions. The vertical represents the nature of your problems. Aggressive technology oriented firms are at the top, conservative late adopters at the bottom. On the horizontal axis indicates whether you have resources like budget, time, people, lab space, etc. Each network lives somewhere on this. Probably everyone feels like they’re underwater, but if you’ve got time to check out a new tech, you’re not totally underwater.First let’s notice that some parts of this grid are not very populated. [CLICK] Upper left, the building is on fire. It’s unsustainable, but no doubt some pour soul has that job. [CLICK] Lower right is Easy St, with no urgent need to learn new tech, but this also seems unsustainable. Years past we might jokingly call that a government job, but no more.[CLICK] Most firms are near the diagonal, I’ll show suggestions there. [CLICK] Lower left, skimming the web is the least you can do, hard to even call it engagement. [CLICK] If you’ve got a tad more time, watch some of the demos to see what apps are most intriguing, then ask vendors about them (this helps!). [CLICK] Next, actually check out the published specs. They’re not huge, and it’ll help clarify some things and raise other questions.[CLICK] If you’ve got more time, I strongly recommend the free/open Tutorial posted on the openflow.org wiki. You can run the whole thing in a VM on your laptop. Warning, some coding involved.[CLICK] That tutorial can be extended to multiple boxes in your lab if you want to do some lightweight experimenting.[CLICK] Beyond that, get some early vendor units into your lab and beat them up. [CLICK] Tons of resources and need? join ONF and help advance the spec. [CLICK] More need? Start to deploy! You heard earlier today that some are actually doing this. That just about covers the spectrum, but [CLICK] if you have even more need and resources there’s still more you can do: develop your own apps/controller. That easy for me to say, since I’m from a switch vendor.So there’s a handful of things you can do, whether you’re on the diagonal or not. If OpenFlow is truly interesting to you, you can do a service to the movement by studying a bit and pressing your vendor for details, since that helps firms know where to invest.
I may say one or two forward-looking statements about Brocade’s products, so our legal folks ask that I show this disclaimer.