Big credit goes to @cloudtoad for putting together this thought provoking-post over at http://www.packetpushers.net.
He makes some very interesting comments and observations, none of which I can actually disagree with.
SDN is a business dream; Where they can buy commodity hardware, do away with high priced router-jockeys, stop paying a premium for mid-range value products, and just focus on whatever their core business actually is.
Unfortunately, I think there’s a lot more to this discussion that I haven’t seen a lot of people address yet. I’m not saying I have any answers here, but I do have a few questions. Hopefully, some of you will post in the comments as to how you see this playing out, because my crystal ball is getting pretty foggy these days. 🙂
1) Skills Gap: It’s been 20 +/- years since the fallacies of distributed computing were laid out. And yet, we’re seeing the past repeat itself again and again. I was asking about the DevOps trend with a customer a few weeks ago and he laughed at the question. They had already tried it he said.
And it failed miserably.
These guys are a startup doing some pretty impressive stuff with BigData ( Hadoop ) and they have a lot of talented coders on staff. This really got me curious. So I asked him “Why?”
His response, which I think applies equally to SDN as it does to DevOps, was the following
” It took you 10-15 years to become a really good network guy. It took them”
pointing over shoulder
” 10-15 years to become really good programmers.”
” I’ll guess you aren’t a good coder, but I can TELL you that they aren’t good network people.”
This has been racing around in my brain for weeks now. Other than the odd exception like @lynxbat
( check out his awesome VMware cloud demo here
I think it’s safe to say that 99% of the network engineers I know are capable of nothing more than rudimentary scripting, and most of that is regurgitated code from examples downloaded off the Internet.
I have a hard time calling someone who downloads a perl script and hacks in a couple of locally significant values a programmer, And yet this is very much the world we are all talking about moving to.
So where are these new breed of GUI-jockeys going to come from? With the hybrid of both coder and network knowledge that they will deliver us from our current state of one-protocol-per-problem. Because sadly, I see a shortage of good network folk in general, let alone good network folk with coding skills.
We’ve been slowly automating out all the jobs that green networks kids used to cut their teeth on. So where are these new wizards of SDN going to get their network experience to learn the valuable “just because you can…” lessons that all of us have over the years?
More than likely, they are going to make some snide comment, as the young are prone to do, on how our ipv4 knowledge, just doesn’t apply here anymore. Offer us a piece of tin can with a string, and offer to write us a new protocol.
all I can tell you for sure?
“Not on my network.” : )
Is it just me? does anyone else see this as a problem? And if so, what are you doing to prepare yourself for the coming divide?
6 thoughts on “SDN. Who’s going to run it?”
I think what’s going to happen is a new set of SDN vendors will emerge that have the skills to build new network applications that take advantage of SDN enabled devices. While this may result in an explosive burst of creativity in configurable and programmable network apps, I’m still not convinced it will have the economic disruption some predict.
Don’t want to conflate too many things, but remember that OpenFlow is sometimes described as the x86 instruction set for networks, not perl. When it matures to the point of supporting higher level semantics/languages, how different is it really from what exists today?
Thanks very much for the comments. I think we’re already seeing some of these new SDN vendors like Nicara, Big Switch, and NEC. I know NEC’s not new, but who knew they made switches?!?!?!?
It’s definitely going to be an interesting time in the networking industry, but I do agree with you that I’m not sure that the impact hasn’t been greatly exaggerated. Only time will tell on that one I guess. But it’s definitely going to fun to watch.
I’ll have to think about the x86 instruction set angle. There’s something about that analogy that doesn’t sit well with me, but I’m not quite sure what it is. I started thinking along the lines of the single asic ( Broadcom=Intel ), but then AMD makes x86 too.. and well. Not sure. I’ll mull it over. More pieces to the puzzle. 🙂
Where are these people ? Well, these people are at the companies that are having problems dealing with the complexities of their network. The ones that need to simplify or die, so to say.
Like the Amazons and Yahoos of this world, which have large installations and have new demands all the time.
I think like with anything new, you’ll have to start out small. Openflow allows starting small, because you can run the existing network and run some parts even on the same switch with Openflow.
From one presentation I understood that Yahoo has problems dealing with their L2, obviously their whole network layout and where the servers sit on the network is fully mapped out in a database. So why would they not use that database as the basis of their L2 ?
After these companies get operational experience, the switch vendors will have some operational experience.
Maybe by that time people can get an idea if it would fit their network as well and only then implement some of it.
Do people really think a new technology or technique should be apply to all ? Or they all need to adopt it just after it becomes available ? Of course not.
Thanks for reading. It’s still amazing to me to see people actually paying attention to what I’m writing. 🙂
I agree with your “where are these people” points 100%. Of course they are at the places like Yahoo and Google that are really dealing with issues of scale that 99.999% of us will never see. And of course that operational experience will float back out to the vendors whom they are working with.
What I don’t see though is how that operational experience gets translated into critical mass and socialized throughout the networking practice. How do people who were not working for those companies, or people who were not working for the vendor and assigned to those accounts get that operation experience? I’m just not seeing that tie yet.
I’ve worked for a couple of manufacturers now, and there’s still a very strong tribal knowledge component that relays wisdom from the old to the young. ( This is known as buying the old guys beers for stories ). Without some method of institutionalizing that information into a widely consumable format, like Cisco has done so successfully with the Cisco Press line of books, I just don’t see how we’re going to get to where the promise of SDN is trying to drive us.
Again, no answers here, just more questions. I sincerely hope that you’re right and more of this knowledge leaks back out, but I’m just getting skeptical in my old age. 🙂
My 2 cents is that, if they are going to come from anywhere, they will come from the same place they did when the Internet started – from universities. It’s my understanding that 25+ years ago, the very first IP networks were on university campuses and they were run by computer science departments. The network was, in some sense, a “playground” for CS professors and grad students. About the time I graduated from college (mid-1990’s), this changed and the university IP networks became professionally managed by an IT department. CS faculty and students became disconnected from what was happening with the production university networks. Over the past 12 years since I started working for a university IT department, I’ve seen a number of unsuccessful attempts to bring IT and CS departments back together. In the past year, I’ve seen IT and CS departments both get excited about the promise of SDN and have seen them start collaborating. I’m hopeful that this strong common interest will lead to a new wave of computer science students that are knowledgeable about networking and are proficient software developers.
I think your probably right that this is where the first new batch of people will come from. But I guess my concern still remains is that they although they will be trained, they will not be experienced. It will still take them a few years of stumbling around before they get enough war wounds to figure out ” just because you can…” situations.
I’m really glad you shared this though. Anything in our industry where a technology can actually bring people together rather than driving them further into turf wars and technology battles is a VERY good thing.
If the only thing that SDN accomplishes is to get apps and networks folk to actually start talking. It’s got my vote.