One Laptop Per Child Talks

OLPC Laptop Physical Design

In October 2006, Mark J. Foster, Vice President, Engineering and Chief Architect of One Laptop Per Child and a leading computer portables expert, presented OLPC technology at the Stanford EE Computer Systems Colloquium.

Due to its length, the transcript of his speech was divided into three parts. Below is Part 1 of Mr. Foster's speech. Please continue to Part 2 and Part 3 for the complete transcript.

Announcer: All right. We'll get started now. This is EE380. The Electrical Engineering Department's, Computer Systems Curriculum. Our speaker today is Mark Foster From One Laptop Per Child, a non-profit organization.

Before we get started with the instructional speaker, is anyone here that's taking this class for credit that was not here last week? OK, So if you'd like to take the course for a one unit credit, all you need to do is watch all ten lectures, and when you watched last week's lecture online, the first five minutes of class, I reviewed the class procedures and have you take the class for credit. So just go ahead and watch that, and if you have any extra questions, you can send e-mail. OK? Thank you.

So, how many of you have heard of One Laptop Per Child? OK that's amazing, that's almost everyone in the whole room. Mark, does that surprise you?

Mark Foster: It does surprise me, I'm afraid, but the numbers keep growing.

Announcer: So how many of you have heard of it before today's class, if you hadn't known the abstract? OK, good, so the numbers are still the same. Just for those of you who aren't familiar with it, let me give you a bit of background on One Laptop Per Child. Since it is a fairly new organization. The association is a non-profit organization in the US that's setup to oversee the children’s machine project. And what they're trying to do is to provide laptops to children in developing countries at a very low cost and sell them to the governments of these countries.

It was first announced at the world economic forum in Switzerland in January 2005. And the Chairman of One Laptop Per Child is Nicholas Negroponte, and their CTO is Mary Lou Jepsen. Mark Foster, who is talking today, is their Vice President of Engineering and the chief architect of the systems. One laptop gained a lot of attention after Nicholas Negroponte and Kofeanon [sp] presented the working prototype of the CM1, which is the Children's Machine One at the world summit at the information society in Tunisia, and that was in November last year. So, just under a year ago.

Now, Mark. Let me give you an idea of Mark Foster's background. He has an MBA from Notre Dame, and he's actually talked to EE380 before. He spoke about SHARK, which was a networking computer. So if you're interested, you could actually find that talk in the archives, if you enjoy his talk today. He's led different projects in portable computing at Apple, at Deck, and at Zenith. He created the first notebook with Ethernet, which was the Z-note. The first true sub notebook, which was the Z-lite. And the thinnest notebook in the world at the time, which was the Hi-Note Ultra.

He also created the first consumer portable which means that he worked on the I-book. And the first high-volume networking product, which is the Airport. He joined One Laptop Per Child in February 2006, as the architect, and he'll be talking to us today about the system they are creating, a lot of the technical details, as well as some of the information about how they are getting these systems to the target audience of children in developing countries. Please Welcome Mark Foster.

[clapping]

Mark Foster: Thank you, and in fact, thanks for also doing the work on finding out a little bit about our background. Because I hadn't actually mentioned the formation Tunisia and all that stuff. so, I'm going to talk about architecture, I'm going to talk a little bit less about electronics or the otherwise specs, because this is a complicated system. It's got lots of pieces. Portables are incredibly painful to do and incredibly fun at the same time. The thing that makes them unique, unlike most other things in computing, is everything interacts. You can't change anything, if a portable is done well, there's no room to add an extra part. A good portable has no air in it. As a consequence though, every simple little thing, that you think is going to be trivial, has a ripple effect throughout the machine like you wouldn't believe.

Now in particular in this machine, we're breaking the rules in a lot of ways. Sliding down the razor blade between hardware and software, to push new kinds of architecture in new ways than the current PC industry. As a consequence, you'll see a mixture of lots of things. The core architecture of course, but also I'm going to talk a lot about mechanical architecture. Power systems design, Asics, Power Management is going to be a major thing a little bit about software and of course, we'll make sure we have some time for questions.

But, let's just start by diving right in and telling you a little bit about us real quickly. As you mentioned, non-profit corporation, make inexpensive laptops for kids, focus on education. That's our mission, trying to give kids, that otherwise wouldn't have a chance, an opportunity, in particular, in the developing world. You can see that's our focus and you can see it here on our long term entries. Here's our sponsors, AMD, Brightstar, eBay, Google, Marvell, News Corp, Nortel, Quanta, Red Hat, SES Astra, and that's not the whole list. We in fact have some companies that are sponsors that don't want to be public.

They just want to help, which we think is pretty cool. As I mentioned, we sell to governments and then the government have to agree to donate the systems to kids. And we have actually a great many more governments than we can handle right now. So in reality, and I know this thing may exist, many governments have expressed interest in this, we're focused on a very small number, that are some of the toughest situations. We want to solve the hard problems first in our launch countries, initially those are Brazil, Nigeria, Thailand, and you will see new announcements about these, as this continues to grow very soon. But, in these environments, we're going into some tough places.

Sitting here in the geek center is one thing, this is where the kids that we're going after sit. This picture was taken just a couple of weeks ago in Nigeria. There's no power in that room. Most of the time, they're not even in there, we were fortunate that this is a school that actually has a room with a roof. A lot of the time the kids, in fact, are outside in the rain, that's how they are taught. Hopefully, it's not raining too much, but that's how they're taught.

Often small groups, certainly teachers in many cases, that have a lot less education or experience than, far less than you do, I was going to say than your kids do, but most of you of course are too young. The teachers there, it's a tough environment, and it's such a tough cycle to break. And that is what we're going into, we know that with our eyes wide open, this is what we're facing. Challenges? Absolutely.

Power, connectivity don't exist. The physical environment stuff, the political uncertainty, I'm sure, if you guys are actually paying attention you probably noticed that there was a revolution in Thailand, one of our core launch countries, and interestingly enough, the Thai's have just sent a group of people to work with us all month long, they're continuing to participate in the project. But as we're moving around, in developing this machine, it's an unusual world out there, we're dealing with a lot of real parts of it that you wouldn't see otherwise. It's central though, we've got to navigate these currents, we've got to develop long term relationships with the governments which is exactly what we've been doing. And get them out there, and effective distribution is certainly key, that's the big part of what we're up to, I’ll talk a little bit about distribution, but it's certainly a major factor.

Another one, and you're going to hear this a lot from me, is inefficient software. Folks we have run up... I guess us old farts like to talk about the good old days right, as I suppose that's what I'm doing, but the reality is, back when memory resources and CPU speeds were much more constrained, the situation is considerably different than much of the software we see out there today. Personally I'm going to be an assembly language hacker, That was how I got started in everything. But the reality is, they actually had a, I'm not picking on Linux, I'm not going to pick on anybody in particular, but to give you a little example. The Gnome clock applet, somebody just measured it's memory utilization at 60mb on their desktop the other day.

That is for a little display about this big. To say that that's nutty would be obviously woefully under exaggerated. And of course then we worry about cost, but surprisingly it's on the bottom of the list. The reality is we know these are tough environments. One of the reasons we got this is because we know it's not an easy task. The certain cost reduction. How do you transform cost dramatically? We're going after a very inexpensive system. The numbers are all over the map that are out in the press, just say this, I've seen them at $100 on the bottom, I've seen them at $150 on the top, we're somewhere in between, but the fact is that's one heck of a lot less than any other notebook out there.

How do you do it? How do you begin to approach those funds, here's some of the ways that we can begin to do it. Obviously the architecture low cost, as you'll see, we're designing a couple of custom ASICs that are in the machine to help us do that, we've used consumer interfaces instead of traditional PC interfaces. That helps, a lot.

Large volume is key. We are building, basically a single configuration. That machine you saw at the start. And then that same machine goes into the different countries. The only difference for those countries are the software mix that we put on the machine, and the keyboard. Everything else is identical in those boxes. That means you can build them on an economy scale, which no other PC manufacturer today is coming close to. Clearly our projected volume our first year will produce the highest volume PC ever. If you even want to call it a PC. It's something different, it really is.

Direct distribution is important. If you look, certainly anybody that has taken any marketing classes, will realize that an awful lot of the price of the machine goes into the distribution chain, so it comes out of the manufacturer, often you'll even have a driver who's exporting it, they'll take a cut, get it to the country, a distributor will take a cut, at least one, then you've got a retailer, and the reality is, that will easily triple, double or triple the price of the machine depending on the specific distributors involved.

That's a big chunk, not having that, which is obviously significantly reduce cost by going directly to the government. Next source is Open source software, no software royalties is certainly a nice thing. More importantly, open source software let's us build exactly what we want. We are not trying to run Fedora, you can run fedora, you can install fedora, but that's not what we want for this machine. It's a start. But the actual product we want is different, and we're changing a lot of software in there to target it at kids, and you'll see.

Optimized software, as I mentioned. Inefficiency in software is causing all of you to spend more money on bigger faster CPUs and more memory than the task in fact should mandate, and again, I could stand here the whole time and talk about software inefficiency, but it would be a waste of time. I hope you'll just go do a PS -aux on a Linux box or pull up the task manager in windows and take a look at the memory footprint that these applications are using. And then go look up the specs of the memory sizing on the original Macintosh. And then ask, wait a minute, are our resources really that small? And then finally power management.

Power management is a surprising one. A lot of folks don't realize, when it's well done, power management can be an excellent key to reducing cost. If you just imagine for a moment that we do a good job on power management, all of the sudden, the battery doesn't have to be as big. The power supply in fact, internal, ACDC converters and such can be smaller. So can the AC adapter. And so, if you really do a good job on power management, you can reduce costs throughout the system. So these are some of our key means of accomplishing these low costs, but the one that I really got to say is we got a little help from our friends.

The reality is that the strategic partners that I mentioned at the start are absolutely essential to this. Much of the companies believe in the same mission, they want to help kids, and are contributing to make this happen. And as a result, we're getting very favorable costs, we're getting a lot of help to pull this off, so this is just by no means a little PC. It's a very large team effort. At the core of the machine is a relatively traditional box. AMD GX2-500 CPU.

The difference between this and the CPUs you see today is that this thing draws a lot less power. The power is about five watts for the GX2 as compared to lets say the Pentium fours that you've got sitting in your machine, which is anywhere between 50 and 100 watts. Starting with 100 watts, we certainly can not achieve our mission. And our mission is to make sure we can have this thing to use in a very tough environment. So what we did was we targeted a power envelope and then went and found a processor that did it, and that's not bull, because I know you can read that AMD is on our board of directors, but when I joined, I had a list of 315 different CPUs to find out what was most appropriate.

What has the best power consumption with floating point? Which was crucial. For 32 bit core in the GX2, was just the right part. Nothing else was out there. There were some interesting power pc chips, not bad, but they had the completely wrong peripheral set though, so legitimately, this is not a PR step, this is the part that I thought was best out of the market. Go out there, graduate, and make better ones because we're always looking for lower power and more pep. 148mb of ram with 1mb of serial interface flash, that's our bios, bios boot loader etc. diagnostics, conventional USB 2.0 ports. And SD card slot is a fairly recent addition to the machine, we'll talk about that in a little bit. We do have integrated wireless which I will go into in considerably more detail. Audio and video support, the usual speakers, line-in line-out microphone.

And then the main storage in the machine is the 512MB LPC NAND flash storage device. So single chip, no rotating media, just the flash card. and then a process is being developed, or I should say specifically extended for this machine JFFS2, the creator of that file system and some other folks are working to make it much better and that's what we're using for storage. That gives us 2X compression on the fly on the primary storage on the box.

Let's talk a little bit about the wireless mesh. Everybody is playing with it, I mean you've got it here. We start with 802.11G that's great but we want to do more. And in particular, the kind of infrastructure you've built up here at Stanford with access points everywhere, nice high convenient pre-planned antennas and everything else doesn't work. We're going into jungles, we're going into deserts, we're going into places that you just don't have that infrastructure. And again, you often don't have power.

So one of the primary thrusts of the machine is to get very good connectivity. We do that through this 802.11G and we're working through the community to produce this new standard which is 802.11S, which is the wireless mesh standard. And that enables you to transform the current hub and spoke model, if you will, for the wireless environment into a truly distributed system that automatically establishes routing between machines. So that's why you don't see an access point there in the machines we've got up there, every single pc notebook is a wireless router.

And it's not just a wireless router, it's a wireless router 24 hours a day, even after you turn it off. There are some exceptions to that if you happen to be in a very power challenged environment, we'll give you some examples of those coming up. Anywhere there is power available, if you can charge your battery during school in the day, you can have a 24 hour a day wireless router and that has very profound ramification. If you go back to the little history, the protocol for this was first developed at MIT called roof net, and in fact long before any of the rest of us had 802.11, that's how they were connecting all of their buildings.

They were geographically distant, and they did it with antennas that were sitting on the roof, directional antennas. What they wound up doing was developing a pretty darn cool system that was resistant to rainy phase, etc. etc. Now we move forward a few years, advance those protocols, and now what's happening, is we're crating this wireless mesh that's a much greater challenge, where the machines are dynamically establishing the mesh as they're turning on and turning off.

In this highly mobile environment, it's a trick. It's very hard, and the wireless mesh routing is really interesting. It's based not just on distance and hops, which of course is crucial, you know, what kind of signal can I get from you to me, and who are you nearest to, and relaying things the right way, but also based on power. Because these machines are battery driven, some are going to be dying, so in fact, when a battery starts to get low, we're not going to be sending 90% of the traffic through that node, we'll automatically redirect around through other machines.

So it's really a quite sophisticated routing algorithm that's interesting. But the crucial thing is this on-chip ARM9 CPU and ram. This is a Marvell part, Marvell corporation, the guys who developed this part, but they were doing an awful lot of special work for us. What this does is, so we've got the CPU and 96k of ram that are on their own power plane. So we can shut down the rest of the system, and this ARM9 continues to run at very low power consumption, and can autonomously route packets without the main CPUs involvement. If someone addresses a packet to this machine, the wireless chip is of course capable of automatically waking up the machine, to respond to that packet that was addressed to it. So it's a very cool capability, it's very important, and as you'll see, there's a lot more you need to do. I told you we're in an environment where there isn't power. How do bootstrap your way. This is how we get started.

And then as we continue to develop the system, obviously a lot of these are going to be islands of connectivity. They won't have, if you will, backbone internet connectivity. And if you don't, you still want to be able to deliver educational content to kids. How do you do that? Well, one of our sponsors is SES Astra. You may not know who they are. They are one of the world's 3 largest satellite distributors of information. And SES has joined our board, and we're actually going to be able to broadcast educational content via DDBS, digital data broadcast satellite.

And so, what we'll do is will then take the place that you can usually get power, is in the school. The school server itself will have a little teeny box. Where ever there is power in the school, we're going to drop that server in there, it's a little teeny thing, low power consumption, just like the system, that has this DDBS receiver. So it can act as an in-point or cache for the educational content we're broadcasting from satellite. So it allows folks that are in an environment where there is no way to get connectivity, for us to be able to deliver kids that information, and give them an opportunity to learn more.

And obviously we want to do more. Where you have connectivity, we can certainly exploit it, and in fact as we start to get a lot more connectivity, we want to improve the scaling of the network. Scaling becomes the limiter in the mesh network, the management traffic increases with the number of machines. So what you want to do is, you want to solve that problem, and the answer will be both solar powered access points, as well as segmentation. Segmentation means I can just group you guys into groups. I can point one antenna over here, and one over here. That's easy, fine. We just doubled the scalability. You can do the same thing with the frequency domain, so I can say from there to there, you're channel 3 and channel 6 are separate. so we've now got four times the traffic in the same physical space. So, the point is, these are necessary things as the system grows up. You start with the bear minimum and moving into a place where there isn't power, isn't connectivity, and then we want to make sure that we provide a path for very high density appointments as time goes on.

One of the really important things that I didn't mention yet is that our minimum order quantity for a government is a million units. No less. There may be exceptions for countries that have identical configurations. But it's basically a million units. Why? One of the key reasons is the mesh. There's a number of reasons behind it, one is certainly the impact on the educational culture.

We want to make sure that we improve education, we want to make sure the teachers have the tools and the training. On the same token, an interesting side effect is with the mesh, with a million units, we get such density for there to be a truly rich connectivity fabric. So that's one of the reasons we're starting with that requirement. There are many other reasons for it as well, as I mentioned, quantity determines economics. The ability is hit a very large volume is crucial, and certainly a major factor in that as well. But, it certainly has other benefits.

Another thing that's really unusual about our machine is the LCD display. Our CTO, in fact, has created something that is really special. What this panel does is unlike anything else I've ever seen. I've seen reflective color, I've seen transflective color, I've seen transmissive panels, I've never seen anything like this. This panel is truly a reflective monochrome panel. No backlight, you see a 1200X900 dot per inch seven and a half inch LCD. And it's dense. At 200 DPI, we're talking very close to laser print quality. Certainly from the original laser printers. It's really nice.

And then, magically, you turn on the backlight, and you see color. Really unusual, pixel structure, it's all 100% Mary Lou's invention, and this is a really neat part of the machine. Completely new text, I'm really delighted to have this component in our box, because it gives all kinds of cool benefits that we can exploit. In particular, it's inexpensive, straight, but also very low power consumption. And this ability to instantly go between a monochrome, a very high-res monochrome mode with a great reproduction of text or whatever it may be, and immediately flip to color mode when you want to is totally cool. I wish I had brought one with me today, and show you, my apologies that I did not.

But, it's running, it works, and in fact, just this last week, we did some [xx] on the panel, that doubles it's reflectivity. We actually measured, and the goal of double that reflectivity worked. So it's a really neat trick, and again, there's no other system out there like this. And it's something that is invention purely of OLPC. Not that someone came and told us. Mary Lou created this, pushed it into the LCD manufacturers and made it real. Really neat stuff.

Due to its length, the transcript of Mark J. Foster's speech was divided into three parts. Above is Part 1 of Mr. Foster's speech. Please continue to Part 2 and Part 3 for the full transcript.