A Revolution in networking is on the horizon…

Date: Mon, 03/05/2012 - 18:57

Network technology had its roots in the Cold War era, when full decentralization was a smart strategy for defending the Advanced Research Projects Agency Network (ARPANET) and maintaining reliable performance across unreliable or war-damaged network links. In today's data centers, however, a pair of redundant, centralized controllers should be all that’s needed to ensure reliability and survivability. Time for a revolution?

A Revolution in networking is on the horizon… Dan Pitt, Executive Director, Open Networking Foundation, at NetEvents EMEA Press and Analyst Summit, Garmisch, Germany

Yes, indeed! And we, at NetEvents EMEA Press and Analyst Summit, Garmisch, Germany (February 22, 2012), are privileged to welcome one of the leaders of this revolution – Dan Pitt, Executive Director of the Open Networking Foundation (ONF). With 20 years at the coal face developing networking architecture, technology, standards, and products at IBM Networking Systems in North Carolina, IBM Research Zurich in Switzerland, Hewlett Packard Labs in Palo Alto, Bay Networks in Santa Clara, and Nortel's Enterprise Solutions Technology Center – not to mention his contributions to academia – he was among the first to recognize the potential for Software Defined Networking (SDN) and to sign up with the ONF.
Judge a revolution by the quality of its leaders: Deutsche Telekom, Facebook, Google, Microsoft, Verizon, and Yahoo! got together last year to found the ONF, to rethink networking and accelerate SDN standards and solutions. Its first conference was massively over-subscribed – a sign of a world hungry for this revolution.
What is SDN? What are the benefits and who reaps them? What has already been achieved and is the ONF living up to its dazzling promise? Where is this revolution taking us in the longer term?
We call on Dan’s wisdom and experience to answer these and other questions...

Dan Pitt, Executive Director, Open Networking Foundation
Good morning, everyone. I appreciate meeting all of you and being here today. I want to thank Mark Fox for inviting me to the session. I’m very happy to be here with you today, and I’m looking forward to a very interesting dialogue today and tomorrow about these new concepts in networking.
I have come back to networking after ten years away. I did my career in networking with IBM, Bay Networks, Hewlett Packard, Nortel Networks and even some start-up companies. I drifted off about at the time when telecom bubble burst because I was really tired of it, and I thought it was old and it had run out of steam and energy.
I was lured back in to take this on about a year and a half ago by the early founders professors, Nick McEwen and Scott Shanker for the Open Networking Foundation. I have been increasingly excited during the past year and half that we have spent. What’s happening is a movement in the networking industry. I do think it’s a revolution for reasons you probably don’t expect, and we will get on to that.
I am the Executive Director. I am the sole employee of the Open Networking Foundation. This is my full-time job.
The Foundation was formed kind of accidentally, but based on a number of frustrations, particularly by network operators and networks users. We can debate this all afternoon if you would like. I’ve listed a few of them up here, some of these trends that are happening. It’s not true of everyone, but there is increasing urgency amongst more and more people who operate and deploy [news] networks about how inflexible they are compared to the needs of business – and that’s always the valid comparison.
So many of them are homogeneous, single-vendor or closed. This frankly was the motivation for the research that led to the invention of the OpenFlow protocol and the general concept of software-defined networking. I will do my best to let you know what these two concepts and terms mean, because they are NOT identical.
The networking community has been a bit envious, I think, of the worlds of computing and storage, in that they have virtualised their operations, but networking hasn’t. It’s interesting, from a research standpoint, networking had its heyday in the 1980s or 1970s, and since then it really has not added anything new, in terms of fundamental concepts or abstractions. We have added numerous protocols; there are 6,000 RFCs. But we still have a lot of custom configuration and programming of each network element, and what’s worse, the applications that reside above it. So it’s become very labour-intensive. I talk to many operators, both public and private network operators, about how labour-intensive it is, and how many of the errors are introduced through the human configuration steps.
There are a lot of what I call point ‘Band Aids’, that others call point solutions. I see Band Aids upon Band Aids upon Band Aids in networking. Each one fixes a particular problem, but only that problem.
I was intrigued in the 1980s to discover this notion of ‘fractals’. Fractals are curves of fractional geometric dimension, which came out of some research by IBM in France by Benoit Mondelbrot. He had some terrific illustrations of these curves. If you look at it with a microscope, it looked the same as the original curve. And if you looked at with another microscope, it looked like the original curve. They were what’s called self-similar.
What I saw happening when I was managing Nortel’s Advanced Technology operations in the late 1990s, and the protocols going into the IATF, they looked like new little curves, the previous curves on the previous curves on the previous curves.
And it was not leading to any fundamental new approaches to solving the problems that kept arising in networking.
We’ve certainly had the scale, first of all of Ethernet, and a lot of bandwidth from 1Gbps to 10Gbps, to 40Gbps to 100Gbps, and the need to process our data, particularly packets, at that speed. And then the unbelievable scale of data centres that we have witnessed today. What’s remarkable about the data centre expansion is – if you look and Zinga, Facebook and Amazon – it’s not done with new technology.
It’s done with basic pieces of very fundamental technology, garden-variety servers and computing elements.
So what I have here is just a picture of a lot of plumbing. And it’s there because I think we’re tired of talking about so much plumbing. One of the notions I’ll introduce today is how we’re going to reduce the importance of plumbing. It’s there all right, but it’s not going to be so prevalent in the applications.
So what are networks good for? I can think of no one I feel more sympathy with than the network administrator, the network manager or the IT manager at many enterprises. You will find that in my remarks this morning I will kind of drift around from enterprises to carrier to data centre operators. When I do that it’s because one of two things in the case. Either what I’m talking about is a very general notion that applies to all, or because there are specific things relating to the one I’m talking about at the moment.
If I look at enterprises, there is really no more thankless job than being the IT manager or the CIO. I think of the problems that they’re facing today. We have this notion in the United States called ‘Take Your Daughter to Work’ Day. One day a year, you’re supposed to take your daughter to work. This was ostensibly to motivate girls to become more ambitious about their career. It’s especially important in the fields of engineering, technology and science. So April 23rd every year I dutifully took my daughter to work and was very gratified. She went off to the finest State Polytechnic University in California, where she majored in hospitality and art. You can never predict what your children will do. But now you find, it’s not just ‘take your daughter to work’, it’s bring your own device to work. Bring your own applications to work.
Bring your own viruses to work. And the IT managers now have to cope with all of these things.
If you look at what it takes to succeed as a CIO, an IT Manager or Network Manager, it’s not that you can do anything great – it’s that you cannot screw up. I had a friend who was running all of the networks and all of the servers for eBay. His goal in that job was not to have CNN on the lawn in front of eBay at 7 o’clock in the morning because something had failed. Could he really ever succeed? Or did he just not screw up?
Think about Mathaud’s hierarchy. This is a psychologist. What motivates people from the time we separated from the apes. It’s a pyramid which starts with breathing, eating and finding warm, private place to go to the bathroom, and then it goes to survival and having housing and safety, and then sex and procreation, then sex and recreation, and then love, then getting a company car and appearing on Britain’s Got Talent or American Idol, and then eventually self-actualisation. But nowhere in Mathaud’s hierarchy do you see ‘didn’t screw up’. That’s what motivates the poor IT managers.
What we have always wanted the network to do, especially for enterprises and for customers of carriers, is to contribute to business value. And by contributing to business value – this is a Gartner chart that many of you have seen. We’ve got various capabilities here of how networks might be operated, other characteristics, whether it’s SLAs and their costs, how it’s organised in the enterprise, how long it takes to respond to opportunities or to failures. But when it leads to business agility, that’s where we want to go.
I don’t know very many network managers, IT managers or CIOs who run anything other than a cost centre. And the CEOs always want to reduce the costs of networking and IT. So the best you can do is to cost less. They all want to be able to contribute to business value. So how does the network actually serve the business goals of the enterprise or the carrier? That’s a motivation behind what we are doing.
So how did OpenFlow happen? Well, it happened accidentally in a research environment, Stanford University, with two people. One was a student, the other a professor. The student was frustrated because students are used to programming everything they want. If they want to write something, they write it. If it doesn’t work, they change some lines of code. They feel they have control over their world because they can programme it.
They get into networking, and the grad student says, I cannot programme this thing.
We’ve got these big, closed boxes, like mainframe computers, but one can’t get inside of them. The professor is thinking something different. He’s thinking, how much research is left in networking? He can’t even pry a protocol beyond IP. Is he stuck in this forever? He cannot build a system to test and experiment and do research with at scale. He can build one, get some net FPGA chips and do some [few] chips. He can simulate it in software, but that doesn’t cut it. You cannot do real world networking research without using the scale of the internet itself.
So these two frustrations led to this notion of what we call OpenFlow and software defined networking. So the transition looks like this. Most of today’s networking boxes are stand-alone boxes that have their own custom ASICs, their own operating system and their own features put in by the networking vendors.
The operating system and the features are embodiments of these IETF protocols and IEEE protocols, and these other protocols you’ve heard about. Each one is an autonomous system that learns about the topology from its neighbours. It’s standalone.
This came out of the mandate of the US Advanced Research Project Agency in the 1960s that we have to have a network that survives a nuclear attack. So if one box gets knocked out, it’s fine. Everything will reconfigure around it.
Nobody has a global view of the network. So that’s led to the development of what are the networking equivalent of mainframe computers. Back in the 1960s, if you wanted some software run, you went to IBM, and a year later they gave you the software that ran that. We don’t do that anymore. In networking, this is kind of the way it is.
So the transition away from this is in a couple of simple steps. First of all, we take this operating system and we move it to a place that has a globalized view of the network, which is true absolutely in data centres, and increasingly in enterprise networks and also in carrier networks. So instead of replicating all the routing protocols in every device, you have it in one place. It’s called ‘one place’ because it’s logically in one place, but it’s part of a distributed computing world. You can reproduce it, you can back it up. We’ve had those tools for 25 years.
The next step is to take the features out of each box and we put it up there, if we have the networking operating system now implemented on open platforms of servers, languages and operating systems, then the introduction of features becomes a simple matter of small programs that run in normal operating environments, written by anybody who wants to write them. It can be a networking vendor. It can be the telecom operator. It can be the enterprise. It can be a third party. And it’s just a small matter of programming, and you can do it as fast as it takes to write software.
So what’s left? Custom ASICs. Already there’s a movement away from that toward merchant silicon. Right now it’s prevalently where there is an Ethernet Layer 2 environment. But if you complete that step, then these become simple packet forwarding devices. This is a general concept; it can apply in any of the use cases that I mentioned (data centres, telecom operators, private networks, …) and other networks we have not even talked about. Instantiations will be very different in all of these environments, but the concept is the same. Packet forwarding is simply hardware or hardware-assisted high-speed, low-cost packet processing.
Today’s merchant silicon IPO is important for mass market at low cost. Today’s merchant silicon handles only certain kinds of packets with certain kinds of headers.
It does not do deep packet inspection. If you want that you can employ FPGAs, custom ASICs, network processors – or you can do the whole thing in software. And some people are doing that whole thing in software at network edges where the speeds are not 10Gbps.
We have this network operating system here, as I showed before. That’s where the routing protocols run. Then we have the features and the controls all up there. So what have we done? We have separated the control plane from the forwarding plane.
This is just a forwarding plane. Packets come in, they get processed very quickly.
This is the control plane where you determine how that processing is done.
Now, to those of us in the data networking world, this is a revelation. We’ve separated the control plane from the forwarding plane. Aren’t we brilliant? Well, we’re only half as brilliant as the people in the telecoms world who have been doing this for 100 years with circuit switching. We’ve had SS7 and all of its predecessors.
So it’s not a new notion, but it’s new to data networking.
Early in my career, I was working for IBM and we were doing the first generation of IEEE 802 local network standards. The job I was given was to standardize something called source routing bridging for the IBM token ring. We had no control plane, but we had to get control information to and from these token ring nodes. It was so ugly, it was embarrassing. I was in the network architecture group at IBM, what we called systems networking architecture (SNA). I knew it was ugly, but we had no notion of a control plane. We had management planes, but you couldn’t do it with a management plane.
We didn’t have a control plane. So it was the biggest kludge I was personally involved in. I’m sorry to admit. We had to do it. It worked, but it’s not a problem that we want to have again. It’s not a mistake we want to make again.
So we’ve separated the control plane from the forwarding plane. And then the features, as I mentioned, are just ordinary software. Now consider the routing algorithm from [Dykstra], shortest path routing. Instead of having it in every device everywhere in the network, you have it in one place. If you look at the RFC that contains that, it’s about 300 pages. I think four pages are [Dykstra] and 295 are all of this other stuff about how do I find my neighbours, and change state and try to stabilize this network, and all this complication. It’s cost and it adds time.
So it’s a simple concept that actually brings a lot of value of enabling you to govern your network from a single place with ordinary software. Now, once you’ve separated the control plane from the forwarding plane, you have to find a way for the forwarding plane to know what to do. That’s where the OpenFlow protocol comes in.
It’s a small part of the whole SDN framework. It’s the open interface to packet forwarding. This is the interface that grad students could not find in the networking boxes. They could not program the boxes. Now they program them in here. We’ve separated them.
What does OpenFlow convey? Basically, it fills a forwarding table. If you see a packet, it has this in the header, you take this action. What is the action? It’s pretty simple. Either you forward it out of port X, or you change it and forward it out of port Y, or you throw it away, or you send it to me because you don’t know what to do with it and I’ll figure out what to do with it. It’s very simple.
So this protocol I consider to be really essential, but really boring. We’re designing a Ferrari here. This is just the drive shaft; nobody cares about the drive shaft, but without a drive shaft the engine and the wheels are not connected and you can’t do anything. You can have a gorgeous body, but the car won’t go. So it’s really needed.
And what’s important about it is that it is now an industry convention. It is a standard.
There is a lot of momentum behind it and it does what we need to do.
The true value of software-defined networking is in being able to program the network using ordinary software as rapidly and as customized as you wish to do. I will say another little benefit of OpenFlow is that these forwarding instructions can be based on flows, not just IP end points. Between IP end points you can have multiple flows at the same time, even between your browser and some web page, you can have multiple flows. Some are doing the video. Some are just conveying news. Some have images. Some might be speech. They have different characteristics. You could actually have them go through different paths based on that. I’ll say more about that in a moment.
So the Open Networking was formed for these two reasons. One, there was a technology gap here. So Stanford started this research programme, and they had corporate member (they always do, it funds a lot of their research). I’ve been on both sides of the fence. I’ve been in academia; I’ve been in industry. We have what I call the centre for adjective noun. Pick your adjective, pick your noun. And they contribute money. They get to come to the table.
The centre is not cheap. It was called the Clean Slate Program -- $150,000 a year from corporate members. They had a lot of members. They became excited by what was going on here, and they thought they needed to standardize this because they’re all interested in doing it the same way. Let’s not duplicate it.
So the need for speed is in the creation of standards. So we were formed, instead of taking this technology and giving it to an existing standards organization, because of the need to innovate, not only in the technology, but also in the standards-making process. And that’s what we are doing.
I’ll explain why we are user-driven. This is the Board of Directors. We have two professors. And then we have seven companies, and these are their directors. If you notice, there are no vendors on this list. That’s because the Open Networking Foundation allows no vendors on the Board. Moreover, the Board is empowered to do a number of things that only the Board can do. Only the Board can charter the formation of a working group and approve its work plan. Only the Board can approve standards that get policed by [indiscernible]. And only the Board can appoint chairs of the working groups.
We have a bunch of working groups, some informal discussion groups, a technical advisory group. Vendors have prominent roles in these, but they are appointed by the Board after they apply. And the working group charters are limited in time and defined by deliverables. And these are short timelines of six to nine months. And then the working group goes out of business unless it applies for renewal of the charter with some new roles.
This is an organization which the users are empowered to set the direction of the organization. Vendors are very happy to join and participate because when the customer says, if you do this I will buy it, the vendors are more likely to say, all right.
I’d love to sell you that. I’m glad you told me what you want to buy.
I’ve articulated here some of the principles that govern how the Open Networking Foundation operates. As mentioned, we have the users in the driver’s seat. We have a very high cost of entry -- $30,000 a year. We want the seriously committed to be involved in writing these standards because we want everyone who writes the standard to have a commercial stake in those decisions, in the technical specifications.
The success of ONF is measured, not by how many members we have or how many standards we produce, but by the market pull for solutions based on OpenFlow and SDN. That’s how we measure ourselves. So we want what we produce to be valid in the real world.
We have implementations of OpenFlow all over the world already in all kinds of networks, including public networks and companies like Fidelity and eBay.
Another difference between ONF and other standards organizations is that we want to standardize as little as necessary. We’re not trying to make work for ourselves. We don’t announce a lot of meetings well in advance in exotic locations and attract the kind of people who love to go to standards meetings for a living. So we have them more occasionally. We do almost all of our work online, conference calls and things like that. And there are things that we’re working on that we do not necessarily thing will be standardized. We’re trying to iterate rapidly. We want get valid feedback from the market as quickly as possible. We’ve released our standards on a time schedule, not a features schedule.
We’re based on Palo Alto. We’re following the model of a Silicon Valley start-up company. We iterate with our customers. Our customers are our members. We are not process-bound. We create processes only when we have a demonstrated need for them. So we try to make decisions quickly to respond to the needs of our members.
These are the members that are not on the Board. These are the ones that pay $30,000 a year; the Board pays more. You can just take a look at this. What’s interesting is that it really covers an entire spectrum of the IT space, except for enterprises. We’re talking to some enterprises. We have chip vendors here. We have merchant silicon and network processor vendors. We have system vendors and networking companies.
We have computer companies.
I got out of IBM, and IBM got out of networking. IBM is back into networking in a big way. I asked somebody from IBM recently how important networking was to IBM’s computing strategy. He said, Dan, I cannot overestimate for you the importance of networking to IBM’s computing strategy.
You can’t offer an IT computing-based solution without having networking integrated into it now. We have other service providers, Korea Telecom, Comcast, Tencent (a large ISP in China). We have software companies, such as Citrix, VMWare, Infoblox.
Take your pick. We have training companies here.
The only thing that’s notable about this is the companies that are not here. I don’t like to call them out by name, they all start with the letter A. Two will be on our panel, I think, of eight or nine of these. One of them is Apple. I don’t want Apple. I’m fine with the iMac, I’m fine with the iPhone, fine with the iPad, fine with iTunes. I do not want iSDN – I did that once.
So this is what we’re working on. This is where I want to distinguish OpenFlow from SDN. Software-defined networking allows you to program the network using ordinary software to have it do exactly what you want it to do. There are a lot of components there. I’ve made these things up; they’re not different boxes. They’re just sort of functional ideas. I’ve even called them by different names: control programs, tools and networking applications.
What you really want is to convey what the network can provide to the applications and the applications exploit that. You don’t want the applications to have to care about port numbers or IP addresses. So that’s why you will virtualize the network in ways you could not do today. The FlowVisor is a slicing tool that came out of Stanford University that lets you run multiple sub-networks on the same physical infrastructure without colliding.
Why would you want to do this? Think of the regulated industries that have to separate parts of their operation for compliance reasons, such as finance, healthcare and others. Commercial banking and investment banking, right now they’re running completely separate networks. We know some of these people who are building common physical infrastructures, separated only logically in a way that is compliant with the law and is auditable.
These are the programs and tools that determine what your roots are. How do you know what your roots ought to be if for the flow paths? It’s based on policy. It’s based on security. It’s based on traffic engineering. It could be a mission  control. It could be energy usage. If you’re in an enterprise, it might be that. If you’re in a data centre it could be energy and cost. If you’re a carrier, it could be reliability, robustness, it could be latency. If you’re in a home network, it could be parental controls. We’re going to see this as a way of finally being able to manage a home network. If you do not have a teenaged son now, you cannot administer your home network – we know that. I have one.
Control networks, ad hoc networks, sensor networks – these are all going to have different operational tools here. Some of these are going to integrate with the operation support systems and the BSS of the carrier networks. There’s a lot of work to do in bringing this into the interfaces that we have for those carriers now. They’re starting to do that. We are exploring what these things ought to be, what their interfaces might be. We are not rushing into standardizing these interfaces up here, because they are mostly software. They are going to be determined quickly by the industry, and they don’t lend themselves to immediate standardization by a committee.
So these are just some of the things we’re working on. Conformance testing so you can claim compliance with OpenFlow and the vendors have that confidence in advertising, and the purchase have that confidence in purchasing.
We’re studying in great detail what we call hybrid switches in networks for the environments where you can introduce OpenFlow into a network that already has legacy components. Now if you look at the carrier world, they’re experimenting with OpenFlow in green field opportunities, like cloud services and data centre services.
It’s going to take them a while to figure out how they’re going to introduce these capabilities into their production networks in the edge, aggregation and core. We’re going to talk about this on the panel a little bit. Peter Feil is here from Deutsche Telekom; he can tell you what their plans are for that.
These topics have different use cases and applicability, and that will affect the timeline of when they are adopted. So the SDN abstractions, that’s these things up here. Forwarding models, we’re kind of tied into what looks like Ethernet frames, IP packets, MPLS and VLANS, and some of the descriptions of the forwarding models imply that you need TCANs at every table, at every phase of the packet inspection.
We don’t want to have that kind of specificity. So we’re looking and more generalized forwarding models.
We are standardizing the protocol itself, because you need to. And we’re looking to figuring switches. We haven’t yet talked about interfaces between different controllers or network operating systems. We will certainly study that with the people who are building them. Our goal now is to get people out there trying stuff and experimenting, and not be constrained by saying you can only do it this way.
This is a movement that I’ve described in kind of general terms. It’s going to have different flavours for the use cases and environments in which it plays. But I think these are some of the things that it will lead to. I’m not very good at predicting the future – I’m not even very good at predicting the past. But bear with me. I think networking is going to become an integral part of computing in a way that makes it less important, because it’s less of a problem. It’s not the black sheep any longer.
And the same tools you use to create an IT computing infrastructure of virtualization performance and policy will flow through to the network component of that as well, without special effort.
I think enterprises are going to be exiting technology – or exiting plumbing. They are not going to care about the plumbing, whether it’s their networks or the cloud networks that increasingly meet their needs, and the cloud services. They’re going to say, here’s the function or the feature I want for my business goal, and you make it happen. And somebody worries about the plumbing, but not as many people who worry about plumbing as today. And if you’ve got this virtualized view, you don’t have to look at the plumbing. So they’ll be exiting plumbing.
The operators are gradually becoming software companies from internet companies.
They are bulking up on those skills. They want to be able to add those services and features themselves instead of relying on the vendors, and doing it quickly for their customers. It gives opportunities to operators that they didn’t have before of operating more diverse services and experimenting at low cost with new services.
IT might actually have greater business value and be valuable in the enterprise for positive goals, not negative goals. Negative goals is ‘don’t screw up’. Positive goals are to improve the bottom line.
And finally, for the young generation of game players, programmers and hackers, I hope to see them back in the networking world. It’s been a wasteland for about 12 years. And even back then, if you were a networking start-up, you were building a switch fabric and it was going to take $50m-$150m of venture investment.
If you look at the world of the Web 2.0 start-up companies, that’s what this is going to become. It’s two guys and a dog. That’s what we say in the Silicon Valley. What is the dog for? The dog is the HR department. You can’t give them anything but love.
Dogs give love.
We’re going to see a plethora of new companies offering these software applications, tools and control modules in the upper layers of software-defined networking.
They’re going to make it easier for enterprises and service providers to customize the networks toward business ends. That’s our goal: to improve business. To improve the contribution of networking to business, with fundamental abstractions, a separation of control and forwarding to enable ordinary software to be used to program the network. And that’s why we call it software-defined networking.
Thank you very much.

Manek Dubash
Dan, thank you very much for that fascinating presentation. It looks like the whole networking business is going to be shaken up big time. I’ve got a couple of questions.
If you have a couple of questions to put to Dan immediately, put your hand up, please, and we can relay those on. There will be a panel discussion after this, so there will be another chance for questions, if you want to hold on until then.
My only question is, you haven’t mentioned performance yet. With all this control data going across the network, with all the things running on standard processes, as opposed to dedicated processes, you haven’t mentioned how good the performance is going to be.

Dan Pitt
Performance is one of those variables that different people have different requirements for. It’s going to be a cost-performance spectrum. Both the vendors and the customers of those solutions will occupy different places on that spectrum. We have done a little bit of study into the performance required for a controller in an enterprise or campus environment for how large a number of end systems it can support, and how many flow request per second it can support. The study we did showed that, with a four year old Dell PC, you would have more than enough processing power for a campus with 30,000 end points. So that’s just one data point.
It will be critically important in a number of environments. But there’s no one answer that fits all for what that performance ought to be.

Bob Emmerson, European Editor, TMC Group
My main questions will come for the panel, but there’s one I just don’t want to forget.
You mentioned toward the end sensor networks. Are you talking about end-to-end type sensor networks?

Dan Pitt
Machine-to-machine? Yes. One of the challenges of sensor networks is that many nodes are both sensors and relays, and they are very, very power limited. So how do you reduce the amount of intelligence those things need to do anything? And this is one potential tool for doing that.

Manek Dubash
I’d like to thank Dan Pitt very much for a fascinating presentation.


valorar este articulo:
Your rating: None

Post new comment

Datos Comentario
The content of this field is kept private and will not be shown publicly.
Datos Comentario
Datos Comentario