 |
 |
 |
Home | Company | Newsletter | 2006.06.07
 |
|
|
 |
IN TOUCH
THE FUTURE OF EMBEDDED NETWORKING
By Bob Metcalfe, Ember Chairman
Sensors Expo and Conference, June 5-7, 2006, Rosemont, Illinois Keynote, Wednesday, June 7, 2006, 9-10am
Dr. Robert M. Metcalfe: MIT engineer, Ethernet inventor, 3Com founder, InfoWorld pundit, and now Polaris partner and Ember chairman. Recipient in 2005 from President Bush of the National Medal of Technology.
Thank you for coming to hear my talk at Sensors Expo about the FUTURE of embedded networking.
I am EXCITED about networking embedded micro-controllers and sensors. But I am also worried about their FUTURE, or at least about the next hour of that future, which happens to be my talk.
Worried? Having for eight years written a weekly column to a half million InfoWorld readers, making predictions about the future of the Internet, I know that talking about the future of networking is hard --- harder still with my two disadvantages here today. My first disadvantage is that I don’t know much about embedded micro-controllers, not as much as most of you do. And my second disadvantage is that even in what I do know, I am biased.
The kinds of computers with which I am most familiar are quite a bit larger than micro-controllers, starting with mainframes in the 1960s, and then minicomputers in the 1970s, and then personal computers in the 1980s. My specialty has been packet switching among PCs using Ethernet, which was invented in a memo I wrote at Xerox Parc, 33 years and sixteen days ago.
IDC reports that last year there were a quarter billion Ethernet switch ports shipped worldwide -- zero to 250 million in 33 years. In some circles that’s quite a LARGE number, but not here in Rosemont, where the annual shipments of embedded micro-controllers will next year top … 10 billion.
If only more of them were networked.
What little I know about networking micro-controllers and sensors is, well, biased. For five years I have been an investor, director, and now chairman of Ember. For the 10 months ending April, I was interim CEO of Ember. Ember makes standards-based semiconductors and software that we are working very hard to sell YOU. So, be careful to discount anything I say, knowing that I have an interest in one particular embedded networking vendor. Of course, I will try to rise above my biases. But, that’s really hard, especially since mine are the best biases.
SUMMARY
Now, in case you have to leave early, here’s what I'm going to say:
Ethernet is still going strong – up, into, across, over, and down -- so consider the winning Ethernet BUSINESS MODEL for embedded networking.
The embedded networking space is huge, diverse, and fragmented, usefully structured I propose into CSI – control, sense, and identify subspaces.
Standards worked for Ethernet. They are going to work for embedded networking, not only because they reduce costs, but they increase values.
Protocols stacks will matter as much as radios; why not a Google of Things?
An encouraging word about batteries.
Embedded networking CO-PROCESSORS may make more sense now than so-called single-chip solutions for many embedded applications, and so co-processors, especially ZigBee co-processors, will sell in the billions for decades.
And don't worry, I'll be done by 10. I'll try to leave some time at the end so that in case what I'm about to say will prove interesting enough that you’ll have some questions and/or comments at the end -- I hope so.
ETHERNET
The rationale for my speaking here today, despite my disadvantages, is that the history of Ethernet can inform the future of embedded networking.
Ethernet was invented in a memo I wrote at the Xerox Palo Alto Research Center on May 22, 1973, before there were personal computers. Since then, Ethernet has been proliferating and evolving. Ethernet is plumbing for the Internet, which is in turn plumbing for the World Wide Web, which is in turn plumbing for Google.
Today’s millions of Ethernets bear little resemblance to the first Ethernet that Dave Boggs and I built in 1973. So it’s reasonable to ask what the word “Ethernet” actually means. Before answering, let me make it harder by listing some of Ethernet’s PREPOSITIONS. Yes, I SAID “prepositions” – namely the prepositions up, into, across, over, and down, which are where Ethernet is proliferating and evolving today.
UP. Ethernet is going UP as a local-area networking (LAN) technology. It is going from 10Mbps on coax to 100Mbps on twisted pairs to 1Gbps and even 10Gbps on optical fibers. Next stop, 100Gbps. The number of Ethernet ports shipped last year was 250 million; this year will be UP again.
INTO. Ethernet is going INTO wide-area networks (WANs), quickly obsolescing and slowly replacing the century-old telephone transmission and switching hierarchy, taking it from electronic CIRCUIT to optical PACKET switching. Ethernet is being deployed at 10Gbps in telephone, television, and Internet core networks. Ethernet will soon be cramming 40Gbps and then 100Gbps INTO each wavelength of dense wave division multiplexed (DWDM) optical fibers.
ACROSS. Ethernet is going ACROSS what George Gilder famously called the “telechasm” between WANs and LANs. Telephone, television, and Internet carriers are now replacing pitiful T1 subscriber circuits with what they call “Carrier Ethernet” services between carrier WANs and customer premises LANs. These services offer QOS guarantees, availability protections, management facilities, and emulation of backwardly compatible time-domain multiplexed services like T1. Clunky old T1 services are delivered expensively today at 1.544Mbps, which isn’t even as much as Ethernet 33 years ago, which ran at 2.94Mbps.
OVER. Ethernet is going OVER the airwaves, which is ironic, since Alohanet packet radio is where Ethernet began 33 years ago. That Ethernet would someday be WIRELESS is the reason we called it Ethernet to begin with. Wireless Ethernet, now called WiFi, took off, with WiMax in pursuit. Cellphones, now being sold at about a billion per year (and incidentally they won’t be called “cellphones” for long) are adding Ethernet-like packet capabilities over cellular infrastructures, and are adding WiFi itself.
DOWN. And Ethernet is going DOWN to embedded micro-controllers. Ethernet is being embedded with wires, but also wirelessly, so for example the IEEE, which standardized Ethernet as IEEE 802.3, has now standardized embedded wireless as IEEE 802.15.4, about which I’ll say more later.
So, with Ethernet going up, into, across, over, and down, what does the word “Ethernet” really mean? Ethernet today is NOT any specific technology, and certainly not the CSMA/CD it was in 1973. Ethernet is probably not even the persistent IEEE 802 standard packet format.
I think “Ethernet” means -- and remember, I’m in charge -- a winning BUSINESS MODEL. The Ethernet business model is not the old IBM vertical monopoly model, or the newer Cisco-Intel-Microsoft-Oracle horizontal monopoly model, or the rapidly evolving Linux open source model. No.
The Ethernet business model is based on:
(1) de jure STANDARDS, not de facto, certainly not NO standards,
(2) implementations of these standards OWNED by companies,
(3) fierce COMPETITION among standards-promoting companies,
(4) market ethics that prevent competing by being INCOMPATIBLE,
(5) EVOLUTION of standards based on market interaction, and
(6) backward compatibility to INTEROPERATE with installed bases.
At the core of each Ethernet business model is a "de jure standard," which is Latin for a standard made by a legitimate standards body, as a matter of law, not to be confused with "standard du jour," which is French for the standard of the day.
There are de jure standards like Ethernet, de facto standards like Windows, and what I call "de ibmo standards," which, like the standards announced by IBM back when it was the monopoly, are not de jure, not de facto, and not du jour. Microsoft makes de ibmo standards these days, but "de microsofto" doesn't sound so good. Anyway, making de jure standards is difficult, but as Ethernet demonstrates, worth it.
The last, sixth element of the Ethernet business model – leveraging the installed base -- has come to be called Metcalfe’s Law. Google it.
I do hereby offer the Ethernet business model to the future of embedded networking.
THE EMBEDDED NETWORKING SPACE – CSI ROSEMONT
Embedded networking, thanks to the relentless onslaught of Moore’s Law, has become what we venture capitalists call “a space” -- a huge space. Again, Moore’s technological imperative has over the decades taken us inexorably downward from mainframes to minicomputers to desktops to laptops to palmtops. Now our computers are so small and cheap that we’re embedding 10 billion of them a year. The future of networking embedded micro-controllers has become a huge, diverse, and mostly wireless space.
In trying to get my head around this huge embedded wireless networking space, I have come up with a simple, tentative, and temporary structuring, which I call, after one of my favorite TV shows, CSI, short for control, sense, and identify. CSI Rosemont.
Above CSI networking is the personal networking space, where WiFi, WiMax, cellular, and Bluetooth live. Personal network nodes each have a human in charge, and they are either plugged continuously into the electricity grid or get their big batteries recharged every day.
Personal networking is above CSI networking. I cannot imagine what’s below CSI networking. Nano something?
Identification (I) networking is at the bottom of CSI. This embedded wireless networking subspace is commonly called radio frequency identification, or RFID for short. RFID has its own contending standards. RFID’s network nodes have no batteries and no micro-controllers. They get their power over the air from so-called “readers” and don’t do much but use their simple radios to transmit electronic product codes, or EPCs.
Above identification networking in the CSI space is sensor (S) networking. Sensor network nodes typically have long-life batteries and low-end micro-controllers. They use their more capable radios to form sensor networks of many nodes whose purpose generally is to collect sensor data.
At the top of the CSI space is control (C) networking. Here is where the ZigBee embedded networking standard lives. Control nodes often need their batteries to last years between rechargings. And because they both sense and control, they need to form reliable two-way networks. Micro-controllers and radio mesh networks in control applications are typically more capable than those of nodes that merely sense.
Of course you’ve already thought of ways in which this simple, tentative, and temporary CSI networking model breaks down, especially if I’ve inadvertently pigeonholed your work. ZigBee, which I pigeonhole up in control networking, is being used for sensing applications. RFID, which I pigeonhole down in identify networking, has higher-end active tags that are being standardized to contain batteries, sensors, and full-blown micro-controllers. Champions of WiFi and Bluetooth are not content to be pigeonholed up in personal networking and have been creeping down to control, sense, and identify applications. It’s a jungle out there.
So, what’s really true about embedded computing and networking, much more so than personal computing and networking, is that they are huge, diverse, and fragmented markets. It’s a BIG jungle out there, which is why we are having expos, conferences, and talks like today in Rosemont.
STANDARDS
If we are to apply the winning Ethernet business model to embedded networking, we have to get back to the old topic of standards.
Does embedded networking need standards? Of course. But as usual, the people currently selling proprietary embedded networking will, duh, argue otherwise. They will NOT say that they would prefer doing what they are already doing, which works just fine, and they don’t want their customers running off to competitors. They WILL say that embedded networking applications are TOO diverse to make effective use of standards – horses for courses. They WILL say that efforts to make standards are failing, either because they are not done yet, or that the standards are kludges. Those suppliers that survive will all EVENTUALLY say that their proprietary kludges are actually standards, and so there. I have been listening to these arguments for 33 years. They are all EVENTUALLY false.
Take ARCnet, please. ARCnet was the first commercially available LAN for PCs. It was introduced in 1977 by a leading PC company called Datapoint, and it was a beautiful thing. ARCnet used token passing over a coaxial cable hub bus at 2.5Mbps. When we were forming IEEE 802 in 1979 to standardize Ethernet, I was sent to ask Datapoint Vice President Victor Poor whether he would submit ARCnet specifications to the IEEE so that it might be considered for standardization. Two weeks later, Mr. Poor called me back to say that Datapoint had decided not to standardize ARCnet.
What happened after that is instructive. ARCnet did NOT suddenly die after IEEE standardized Ethernet in 1982. ARCnet blossomed along with PCs because of its commercial head start, because it was cheaper than Ethernet, because it was “deterministic” where Ethernet had collisions, and because at 2.5Mbps it was actually faster, Datapoint dubiously claimed, than Ethernet’s collision-ridden 10Mbps.
I had happily thought ARCnet died during the 1990s when Ethernet went to switched hubs at 100Mbps. But no. I just Googled ARCnet. There is this very day, in 2006, an ARCnet trade association website.
ARCnet, according to the website -- copyright 2002 -- and other Google references, was made an ANSIish standard in 1991 and has sold cumulatively since 1977 over ... 10 million nodes. ARCnet is now no longer touted for PC networking, but has found a new life in, yes, you guessed it, embedded networking.
ARCnet lost to Ethernet in PC networking because of the Ethernet business model. Consider that Ethernet is not just a standard, but a NETWORKING standard. As a standard, Ethernet grows CHEAPER over time because of overwhelming investments, increasing volumes, and fierce competition. Buyers like that. But as a NETWORKING standard, Ethernet grows VALUABLE over time because of, well, Metcalfe’s Law – V~N^2. Standards have advantages, but NETWORKING standards have those advantages squared.
Tim Simon runs Golden Power Manufacturing in Hong Kong. Tim is a sophisticated proponent of ZigBee/802.15.4. He doesn’t just want access to standard ZigBee components for his Chinese factories from fiercely competing suppliers. He is also working to convince his customers, do-it-yourself retailers around the world, to offer a variety of interoperating, multi-vendor ZigBee-branded products, say thermostats, lighting controls, HVAC, security devices, and garden sprinklers. Tim aims to become the world’s first ZigBee Original Design Manufacturer (ODM), leveraging not just the cost benefits but also the value benefits of the ZigBee standard.
Like Ember and Tim Simon, I am a proponent of ZigBee/802.15.4 embedded networking standards – 0dBm 250Kbps 100m wireless mesh networking at 2.4GHz. Am I going to tell you that ZigBee should be THE only standard for the future of embedded networking? No. I am telling you that non-standard networking is not the way to go, and that there will be more than one standard in CSI networking.
There will be more than one CSI standard because embedded networking is a huge, diverse, and fragmented space. It is very unlikely that any one standard will adequately serve all the many applications in that huge space. There will also be more than one standard because technology advances and outgoing standards need to coexist with incoming standards. There will also be more than one embedded networking standard because bare-knuckled commercial tribalism is not dead.
Well, I have to admit that there are many standards under the Ethernet rubric these days, hardly any of which I invented. It’s just that for 33 years the winning networking standards have been called Ethernet.
So, let me mention in passing the standards battle now starting up between ZigBee and Zwave, in the control portion of the CSI space. This battle resembles in many ways the historic battle between Ethernet and ARCnet. Zwave has a two-year commercial head start and is currently cheaper than ZigBee/802.15.4. Zwave is also less capable than ZigBee, for example operating at 9,600bps in a single channel around 900MHz versus ZigBee’s 250Kbps spread spectrum around 2.4GHz. But most importantly, Zwave is a proprietary PRODUCT of Zensys, while ZigBee/802.15.4 is a de jure STANDARD, with all the committee meetings that entails.
ZigBee versus Zwave is deja vu all over again. The history of Ethernet and ARCnet tells us that BOTH ZigBee and Zwave will prosper due to the rapid growth in embedded networking, especially since embedded is a much larger and more fragmented space than PC LANs. History should also tell Zensys not to fight a rearguard action waiting 10 years to get Zwave turned over to a standards body.
Before leaving the topic of radio standards, let me mention in passing two old “E” technologies that threaten these two new “Z” technologies. Threatening ZigBee and Zwave are Echelon and Ethernet. Echelon, founded to do control networking in 1988, has a considerable head start on both ZigBee and Zwave, whose technologies Echelon says, duh, are not very good. And although mostly wired, some of Echelon’s technologies were accepted as ANSIish standards around 1999. And ZigBee and Zwave had better watch out for Ethernet itself, which has its greedy eyes on the embedded space.
THE EMBEDDED INTERNET
Analogizing ZigBee to Ethernet is not quite right. ZigBee’s adopted IEEE 802.15.4 radio standard is analogous to IEEE 802.3 Ethernet, both at the two lowest layers of the ISO protocol reference model. ZigBee above its radio is a protocol stack, just like the Internet’s TCP/IP is a protocol stack, both at the top five layers of the ISO protocol reference model. Zwave, like others in embedded wireless networking, has its own radios and/or stacks too.
During the 1970s and 1980s, Ethernet carried packets for a wide variety of protocol stacks. Among them were the Arpanet’s NCP, Xerox Parc’s Pup, IBM’s SNA, DEC’s DECnet, Xerox’s and Novell’s XNS, AppleTalk, Microsoft’s MS-NET, ISO’s OSI, and the Internet’s TCP/IP, to name a few. What followed was a fierce battle above Ethernet, which TCP/IP eventually won in the 1990s. What stack will win in embedded networking?
I am expecting that ZigBee standard stacks will win in embedded control networking, but I’m pretty sure there will be others in the many nooks and crannies of the CSI space. The battles will take decades.
Speaking of battles, TinyOS is frequently mentioned as an alternative to the ZigBee stack, but that’s not right either. TinyOS is an embedded operating system with which a lot of R&D has been done on non-ZigBee mesh stacks. But TinyOS can also be ZigBee-enabled, and has been. Ember is working to make its new EM260 ZigBee co-processor available to TinyOS developers who want to use a standard rather than develop an experimental mesh-networking stack for their applications.
Speaking of standard stacks, there’s TCP/IP and especially IPv6. Why not use IPv6 on 802.15.4 radios instead of using ZigBee? That’s a similar question to, Why not use IPv6 on the Internet instead of IPv4? The answer is that IPv6 (and IPv4) are generally too heavy for embedded networking.
The resources available to embedded networking nodes just can’t lug around TCP/IPv6 packets with their 128-bit addresses. Of course on the high end of embedded networking, this might not be true. We can expect TCP/IP stacks to penetrate some of the higher-end segments of the CSI space, but in most cases, if an embedded network is to be on the Internet, the TCP/IP stack will be running on a gateway node. Embedded network nodes running lighter weight stacks more suited to radio meshes, like ZigBee, will hang off Internet gateways.
Bob Heile, chairman of ZigBee and a denizen of IEEE 802, thinks that many embedded networks will be gatewayed onto the Internet. He refers to the CSI networks built with ZigBee as being part of the “Internet of Things.”
Heile’s embedded Internet vision resonates with that of TI’s Don Sturek. Don was CTO of Figure 8, which became part of Chipcon, which became part of TI, in one of the pioneering roll-ups in embedded networking. Don not only sees ZigBee/802.15.4 embedded mesh networks on the Internet, he asks, When will we have a World Wide Web of Things? Heck, when will we have a Google of Things?
A World Wide Web of Things would enable easy publishing and spontaneous viewing of control-sense-identify information. A Google of Things would interconnect CSI data to enable powerful search and inference.
Just look at the emergent surprise after we managed to hook up so many of the world’s documents using Tim Berners-Lee’s World Wide Web. Along came Google, and all our IQs jumped 100 points. Imagine the surprise when we manage to hook up many of the world’s embedded micro-controllers and their control-sense-identify data? Very exciting.
Another way to say what I'm saying here about protocol stacks, gateways, and the Internet, Web, and Google of Things, is that embedded networking is NOT just a chip business, NOT even just a software business, but a NETWORKING business. Embedded networking suppliers are wrong who think the road to long-term success is giving away non-standard software with their non-standard chips.
BATTERIES
And now an encouraging word about batteries.
Recall one of the challenges of embedded networking applications is battery lives measured not in hours but in years. As ever MORE powerful micro-controllers and their radios require ever LESS power, the CSI application space, already huge and diverse, expands rapidly. MULTIPLY this by the increasing availability of many new energy sources.
As a venture capitalist, I’ve been looking at the battery space for a couple of years, and there’s great stuff coming. For example, in Golden, Colorado, Infinite Power Solutions is building fabs for volume production of thin-film batteries you can buy by the cm^2 – solid state, postage stamp thin (<5mm), stackable, flexible, almost indestructible (<-50'C to >+180'C), energetic (>.2mAH/cm^2), powerful (>10mA/cm^2), safe, infinitely rechargeable (>100,000 cycles), long shelf life (<1% self-discharge per year), low cost, and laminated into circuit boards and eventually integrated onto semiconductor substrates.
Thin-film batteries, printable batteries, and various forms of energy harvesting will make feasible applications for billions more embedded micro-controllers in coming years. Not just for HVAC and remote control, but credit cards, digital pens, active shoes, medical implants, and tags for asset tracking, smart labels, and security. The list goes on.
SINGLE-CHIP ZIGBEE SOLUTIONS
Let me close with a twist on the inexorable integration of embedded micro-controllers, sensors, and their networking.
It’s great for embedded networking that TI, Freescale, and Ember each introduced last year what we all call “single-chip” ZigBee solutions – for example, my favorite, Ember’s EM250 ZigBee SOC, shipping in volume.
And there are OTHER so-called single-chip ZigBee solutions. At the ZigBee Open House in Milan last March, in the booth next to Ember’s, I saw another “single-chip” ZigBee solution. But, this single chip didn’t have a micro-controller on it, or RAM or flash or DSP or encryption or stack. It was just most of the parts of a radio transceiver on one chip being marketed as a “single-chip” solution while missing a lot of stuff.
This made me see the TI, Freescale, and Ember ZigBee chips in a different light. They sit on tiny circuit boards surrounded by discrete passives and sometimes power amplifiers, plus batteries and sensors – not really single-chip sensor networking solutions either.
So, just how integrated will the future of embedded networking be, when?
Going back to the 33 years of Ethernet history, I remember network interface cards (NICs) and routers, both billion-dollar examples of prolonged networking non-integration. For about 20 years after the first so-called single-chip Ethernet solutions, people chose to buy Ethernet for their PCs on hundreds of millions of separate boards called NICs, rather than having Ethernet chips soldered onto PC motherboards. And, you can confirm with Cisco, it’s been 20 years, and people are still choosing to buy routing software in separate boxes called routers, rather than integrated onto a single board or chip.
So, even though we talk about chips the way we used to talk about boards, still the relentless drive toward deeper and deeper integration takes time and is delayed sometimes for good reason. So it will be with ZigBee chips.
This week, after getting into volume shipments of our single-chip ZigBee node, the EM250, Ember is seeming to go backward by introducing its EM260 ZigBee Co-Processor. Ember CTO Andy Wheeler, VP Engineering Skip Ashton, RF Director Nick Horne, and their teams came up with this co-processor idea. After five years of working with application developers, they think many developers will ZigBee enable embedded applications, NOT by porting onto one of these amazing single-chip ZigBee nodes, but by adding a ZigBee co-processor to their current application processors.
Why would a developer want not a "single-chip" ZigBee node, but would instead choose to hook her application micro-controller to a ZigBee co-processor? Let me count the ways.
First, the developer would want to avoid porting her existing application software to any new micro-controller. Second, the developer would not want to bring the ZigBee stack into her application software space. Third, the application micro-controller would not have to handle the transmitting, receiving, checking, encryption, acknowledging, retransmitting, storing, and forwarding of mesh packets handled by its ZigBee co-processor.
But wait, there’s more. Micro-controller suppliers, of which there are many, each with its own broad family of micro-controllers, would not have to add a low-power embedded radio to each and every one. This is good because advanced CMOS radios are still quite a bit bigger than micro-controller chips, they require semiconductor processes with significantly larger feature sizes than micro-controllers (now 180nm versus 90nm), they are progressing more rapidly than micro-controllers from a late start, and they cost quite a bit of money to port, say $5M to $10M for each micro-controller.
In 1984, as my company 3Com was going public, I predicted that Ethernet cards would ship in the millions for decades. I now predict that ZigBee co-processors will ship in the billions for decades.
CLOSING
So, in summary:
Ethernet is still going strong – up, into, across, over, and down -- and the winning Ethernet BUSINESS MODEL is something you should consider for the future of embedded networking.
The embedded networking space is huge, diverse, fragmented, and maybe usefully structured into CSI – control, sense, and identify subspaces.
STANDARDS worked for Ethernet, and they are going to work for embedded networking, EVENTUALLY, not only because they decrease COSTS, but because they increase VALUES.
Protocol software STACKS matter as much as radios, so why not a Google of Things?
Great new power sources -- THIN-FILM BATTERIES, FOR EXAMPLE -- are coming available that will further open up the CSI application space into billions more embedded networking applications.
Embedded networking CO-PROCESSORS make more sense right now than so-called single-chip solutions, and will in the billions for decades.
EXPO BIOGRAPHY
Bob Metcalfe is a high-tech venture capitalist at Polaris Venture Partners in Waltham, Massachusetts. He serves on the boards of Polaris-backed start-ups including Ember, Narad, Paratek, Mintera, SiCortex, and GreenFuel. In 1973, he invented Ethernet at the Xerox Palo Alto Research Center. In 1979, he founded 3Com Corporation and took it public in 1984. During the 1990s, he wrote a weekly Internet column in InfoWorld for over 500,000 information technologists. Metcalfe graduated from MIT, got his PhD from Harvard, taught at Stanford and Cambridge, and was elected in 1997 to the National Academy of Engineering. He received the National Medal of Technology in 2005 and is a Life Trustee of MIT. |
| |
|
|
|
 |
|
