Inconvenient Truths, SDN, Innovation and Other Ramblings
I started this post on March 27, 2012 and it resided in the draft folder in various forms till today. During the intervening months a lot has happened. Google publicly detailed their OpenFlow based WAN at ONS 2012. Cisco announced ONE proving even large companies progress through the Kubler-Ross five stages. SDN went beyond mainstream. Over espresso on Sunday morning, I read this essay in the WSJ by Nichols Carr. You are welcome to read the essay, but I will quote the concluding paragraph if you do not want to click out:
“Knowing that the cause of our innovators’ faltering ambitions lies in our own nature does not make it any less of a concern. But it does suggest that, if we want to see a resurgence in big thinking and grand invention, if we want to promote breakthroughs that will improve not only our own lives but those of our grandchildren, we need to enlarge our aspirations. We need to look outward again. If our own dreams are small and self-centered, we can hardly blame inventors for producing trifles.”
I found the essay by Carr to be well timed. On Friday of this past week, I had two casual interactions that subtly support Carr’s thesis. The first was a casual discussion with the CEO of Plexxi, who is a MIT grad. I asked him where are the software grads or interns from MIT? We are located next to the MIT campus. We are building products that are complex, interesting, difficult to engineer, challenging, but they hold potential to have a profound impact on our industry. If this was 1992, we would have a line of people out the door trying to get in, but it is 2012 and it seems too many innovative people take the network for granted and assume there is nothing to be done.
Our CEO’s response to me was that the young software engineers were all focused on trying to start the next Instagram, the next social mobile site that leverages the Cloud and big data to deliver a break through app. In the words of Nicholas Carr “One consequence is that inventions have become less visible and transformative. We’re no longer changing the shape of the physical world or even of society. We are altering internal states, transforming the invisible self or its bodily container.”
The second interaction I had was with a person who architects a network. He is consumer of technology, a practitioner of networking products and technology. Our interaction was on Twitter, which in the context of Carr’s essay is ironic. Here is the exchange which started with tweet from Plexxi. I removed the Twitter handles as it was the tone of the interaction that is important, not the participants:
XXXXX “Direct network capacity in real time.” Smells like Unicorns. How does one do physical setup (like CLOS, etc) with Plexxi?
WR_Koss Just two cables.
XXXXX @WR_Koss I think you know that’s not what I meant ;-).
WR_Koss Oh yea…we use a mouse.
XXXXX @WR_Koss Maybe you didn’t know what I meant. I know you’re not suggesting that a large DC build out will consist of just two cables.
WR_Koss True, but DC network innovation has been dormant for a decade+ so lots of room for improvement.
XXXXX @WR_Koss Indeed.
My point in reproducing this exchange is to highlight that even the smart, experienced people with a CV full of certifications are skeptical of innovative change when it comes to networking. Perhaps they have been burned too many times in the past and that is why they label big ideas Unicorns. Maybe a foundation of a thesis can be found in this micro exchange. Why should an entrepreneur do something hard? Why take on a hard challenge when your market is full of skeptics? Maybe it is easier to build an online mold removal app and get some subscribers and sell your site? Why use hard science and engineering? We should all just stop trying to do the hard stuff and just buy the ebook version of Networking for Dummies. I can tell you the reason why there has been little if any innovation in networking in the past decade. It is hard and many people see no reason to wake up in the morning and embrace hard.
Bullshit is my answer. There is value in building something tangible. There is value is conceiving something that has never been done before. That is called innovation. Step two is to take your innovation to market and convince people to exchange something of value for your innovation. When customers use your innovation and see value in it, that is the standard by which to measure one self. Customers are the true mark of success. That is how you build a company, create jobs, create wealth and give back to the community. That is what America was built on; the Man in the Arena. America was not built by the Man on the Couch.
Over the past few weeks I have been collecting some inconvenient truths that few people want to acknowledge or discuss. I think the reasons are many: fearful of change, ignorance, denial, anger, loss, etc. Yea, there are lot of reasons, but the good news is I meet new people every week who want to acknowledge the truth. As a microcosm of failing to realize a broad trend I read this on Friday after the early release of earnings from one networking company “...weakness is a combination of a) macro issues and b) share loss, as competition is increasing. [He] does not believe XXXX’s miss is indicative of continued carrier spending weakness and argues that other telco-exposed names actually begin to show signs of recovery this earnings season.” Today, I read “XXXX’s: After yesterday’s miss and guide down…[He] is recommending investors stay out of telco-exposed names through the rest of earnings season and would prefer to revisit XXXX, XXXX, and XXXX later in the August/September timeframe — actually, the only name he is pointing people to is XXXX, as he prefers the ‘wireless’ exposure.” A big change in three days.
Here is the problem that many analysts, pundits and participants are unwilling to acknowledge or maybe grasp. The service provider business is becoming a Bits R US business. That is the home of lower prices and prices are always falling. The value has moved to the compute element inside the data center. The people who own and operate data centers (i.e. servers), just want a managed wave or dark fiber and they will direct flows to and from the compute element. The stuff in between is only important to the service provider. The delivery network has decoupled from the compute element. It does not matter if you are wireless or wired, IP is IP.
The layered protocol stack is all done. Back in the day it was great. It was developed to solve the internet problem within the constraints of 128kb of memory and 50 MHz processors. Fortunately we have moved on from those constraints and today we do not need to guess the network state from the wires up – we can compute state and many states from the top down. Indexing massive databases with MapReduce and utilizing processor power has been proven, we can certainly apply the same techniques to the network. My thoughts on what Google was doing after Urs Holzle’s presentation at ONS 2012 are posted here.
Here is a link to an article on Infosys who reported disappointing forward guidance with the CEO stating that “…a large percent of the above $1b deals have disappeared.” This is not a surprise to me as I wrote about it recently. There is a significant sea change event building in the land of IT. New techniques, technologies and regulations are enabling enterprise to own their IT again, because IT really does matter. SDN will be the component that changes the network portion of the IT model.
SDN this and SDN that and what the killer apps for SDN! Who cares? The SDN use case is universal. Riddle me this Batman…what happened to the service provider voice networks with the introduction of VoIP? As far as I can recall, voice became pretty cheap as the carriers separated the data plane from the control plane. For those not steeped in the difference between a Class 4 and Class 5 switch, a Class 4 switch is the control plane and the Class 5 switch is the data plane. Did I just write about separating the data plane from the control plane? I think I have heard that before. Where would I have heard that before…hmmm….oh yea, SDN! SDN is separating the control plane from the data plane; and now you know the rest of the story.
I think these times of innovation and then disappointment (or lacking urge to change) go up and down depending on the markets over the years. I remember being in the supercomputer business where we were going to change everything. Then in 2001 no one cared about technology or innovation, as it was ‘too hard’. Now we do again. But, having said all that, the closing arguments from my business school were: “The mind is like a parachute. It only works when it’s open.”
I have no problem owning that interaction. The issue I have is not with innovation. Plexxi sounds like they are doing something very interesting. We’ve had this interaction before. The network is supposed to be delivering applications, etc.
My problem is with with language and presentation. Particularly in the realm of SDN. Let’s examine the statement “Direct Network Capactiy in Real Time.”
Define “Network Capacity.”
Define “Real Time.”
I can guess what you mean. I think you’re referring to the ability of applications to request network capacity? Or for a controller to predict and allocate network capacity on behalf of an application? Many network engineers however do not think of capacity relative to an application. This is where the unicorns remark came into play (and maybe there I sounded like a dismissive a-hole), but my point is that network engineers tend to think of capacity in an aggregate way relative to an uplink or connection. From this point of view, this statement sounds like unicorns indeed. You can’t allocate more resources in real time since there is a physical dependency on adding resources to the network: you have to run additional cables and steer traffic over the new paths. I was hoping maybe my reference to CLOS would have been a clue here.
However, the “two cables and a mouse” remark smacks of the kind of marketing that Brocade engaged in with the commercial of children running cables all over the place to form a magical network that passed packets. As if you couldn’t do that with an IP network. Lots of crappy physical designs exist in the world. No amount of software will compensate adequately for it in the long run.
This kind of marketing, however, is intentionally misleading and it is obviously directed towards manager-types, not engineers. It seems to me SDN startups are taking advantage of certain ambiguities in terminology to deliver false impressions of network utopia. But who knows, maybe you’ve engineered something that defies physics and you can build out a 10k host environment with “two cables.”
Going back to my cookie crisps. Also, to be clear, I think highly of you Bill, so don’t take this as an attack. I’m not entirely caffeinated yet this morning.
I never thought it was an attack. I also accept that I am public in my blog and that people will comment, disagree, confront, etc. IMO…do not blog if you do not want people to read it. So I am cool. I also understand the marketing versus reality statement.
In terms of Plexxi, I think the solution will make more sense at the end of August when we provide more details on the product. You are not the first person to ask for a design of 10k host environment and yes it is more than two cables, but there are better ways to compute and direct capacity without running cables.
I have to agree with Derick. I think this post contains a few wishes for unicorns.
Even more disappointing, it attempts to re-label technical debate as anti-innovation rhetoric. Sure, some people will attempt to undermine innovation to protect their own business interests, ego, etc. I work with quite a few people like that, in my day job as well as across the industry. But some of us, especially those of us that are attempting to innovate ourselves and are familiar with the fundamental “hardness” of the problems, are justly skeptical of generalized and unsubstantiated claims. If it makes you feel better to think of us as anti-innovation, rather than actually engage in difficult discussion, then you’ll have to forgive my lack of professional respect.
As for wishful thinking, when I see statements like “the layered protocol stack is all done” I (usually safely) assume the author simply doesn’t know much about networking. There are a bunch of people that *use* the network and impatiently think it should just work. Magically. All these bits of complexity are irritating, so we must not need them… right? As much as I agree with the argument that networks need to be more responsive to applications, that we need SDN interfaces and APIs, I’m sorry to say that application developers are usually just network *users* that don’t necessarily understand what they’re talking about. As network innovators, it’s our responsibility to provide abstracted interfaces so they don’t have to understand. But that’s a hard problem, and the “big idea” doesn’t make it any easier.
Which leads to what I believe is the underlying technical argument in your post… I think you’re suggesting that by removing the control plane from the network we end up with some kind of magically better network environment. That somehow we can make the data-plane just work better by decoupling it from control, as if the ability to compute offline topologies somehow makes up for the physical constraints on capacity. I’m dismayed to see you refer to (whether you realize it or not) an old network architecture like SS7 and ISDN as a vision for the future. Circuit switching with out-of-band control is hardly a network paradigm that will defeat packet switching. History has already played out that debate, and I see no point in repeating it.
On the other hand, if we accept the value of packet switching and the layered protocol stack, but acknowledge that our control-plane was developed to serve the network and not the applications, then maybe we’ll find ourselves on a constructive path. There is plenty of innovation to be done, and I hope that people keep thinking up “big ideas” to drive progress. But the hard problems are still hard, and the technical debate is necessarily technically detailed. Otherwise it’s just empty marketing.