 Okay, so I'm Jim Gavis, for those of you who haven't met me before, I'm one of the original people who started the XML system years ago, and worried about the web protocols for a while, and helped reform the accident-all-only stuff when X3D6 stopped functioning, which is a tragedy. In any case, these days I'm working on a one-line talk-to-child program project that Nicholas Nicoletti has started. He had been the director of the MIT Media Lab, but all you see is a separate organization that's not part of MIT. The picture here is of a school in Cambodia that Nicholas doesn't cost very much to build a school in Cambodia. And he and other people in the Media Lab have been interested in helping kids learn with computers for about the last 30 years. So one of the things that he did some years ago was give a laptop to every kid in one of the schools. And this, of course, has both expected and unexpected consequences. The first problem was to convince the kid's parents that, in fact, they were supposed to use the laptops when they went home at night, because these computers were worth more than the houses that they were in. And the most valuable thing, by far, was the family cap, right? So the first problem was just convincing them this is the kid's machine and they're really supposed to be able to be used all the time. The next interesting observation was that for most of these kids, this was the first artificial light in their house. And that is not uncommon in many parts of the world. And so that has driven our thinking in a lot of ways, as you'll see. So, you know, our conventional laptops are really built suited for this sort of environment in a lot of ways. And so we're specifically, since nobody built the right hardware, we're in the hardware business. It's not fundamentally what we're interested in. We're interested in helping kids learn. But until we convince the world there's a real market for the uncertain, bullying children in the world, we're in the hardware business. How long does it take to form? I don't know. So our initial goal is $100. We're not going to make it in the first generation. It looks like it would be about $135. The next generation is just going to make $100 a place. There's been tons of newspaper articles. You've all heard about it. I want to sort of explain a little bit about the structure of this whole thing. And the intent is that, you know, the four reasons we'll get to you later, well, to begin with, in most places education is controlled by the government for better or for worse. And there's an observation that in Brazil and China, for example, they're selling about $100 for 12 per year. So, you know, the observation is if you can make something for a board of $100, that in fact the education budgets don't have an increase because that often won't happen. So we really want to get as cheap machines as possible. We get $100 or the next goal will be $50. But the point being that a commercial company wouldn't have the motivation to keep making it cheaper to the world. What they look for is, well, come down and enter a lower price to, you know, get market share. And typically the price stays the same and the features go up. And they may become more profitable over time, right? But we're trying to really push the market. And so we're not a profitable association. Or Fire 1C6 or something like that. And a bunch of companies, not all of which you can see because the display here, the screen doesn't quite seem very right. Sorry about that. I've put out a few million dollars a piece to see us through building these. So that includes Google, Red Hat, 3M, News Corporation, the ones you can't read here, Marvell. You can make communications in a great service. We'll hear more about them later. And we're using an AMD Geo processor. Brake Star, the probably never heard of, distributes something like 70 million cell phones per year. More than 12 million and so on. There are obviously the funding suppliers of various sorts involved. We include a quantum computer. A quantum computer is probably the largest computer company you've never heard of. There is statistically a one in three chance of the laptop in your lap having been made by Honda. They make laptops for everybody. They make 32% of the laptops of the world. So they are the 10,000-ton building and the laptop is manufacturing business. They're based entirely on their manufacturing that's outside of Shanghai. So it's more of the same. The point here is the marketing distribution at this price point would take too much of a price. So we really want to avoid the retail channel. You know, software would be another big challenge. I'm not sure the top graph is correct. I think they would move. But we're trying to get to the point of building 5 to 10 million units next year. The point being that we need volume to make things cheap. We also, you know, by the time you would actually have a machine able to support Microsoft Windows XP for real, you end up with a formula or a class machine you just heard Microsoft and Intel and so on and on. So we're trying to get something cheaper. And we can get there using Linux. From some arbitrary point of view, Linux is a relevant to us. What matters is the kids learning. Okay? Now, we believe it's very important that the kids be able to take the machine apart all the way down to the bare iron. Understand how the figures work. It's a fundamental skill now in our society. So we are very much in the open source free software camp. Linux has the most vibrant community by far. And it's the obvious choice. And it also is able to run on flash and delay. So, here's where we are. Last fall, you heard the big press. I have a new tunerist. You know, we're talking with a bunch of countries. The Brazil and Thailand are the fruits along. We're talking with others. I don't think we'll see real commitments until we have fully running header types around, I guess, August or September with the displays and the like. We're also trying to crack the nuts of China and India. So that's really tough. There are a number of principles out of all of this. It is scale. You can't make the price if you aren't doing this in a really huge scale. So this is an attempt to do something out of scale that is really scarily large. To give you an idea, the 5 to 10 million units is a quarter of the Xbox 360 grand in a year. And they use three manufacturers. We're using one, but it happens to be the biggest laptop in the country in the world. The price, I say it's floating. If the price goes down, or if the price of the components go down, then the price of the machines to the governments will go down as well. There's a component, so it's cost plus thing. We have full access to quanta's actual expenses. The way OEMs usually work is they basically build a particular machine at a particular price point against other OEMs, and then they work like hell to figure out how to make the machine cheaper and cheaper to keep the difference. That's the business model they're usually in. This is a little bit different. They're working with us on a cost plus basis to be able to see what the prices are for all the components. That doesn't mean I can tell you the prices of all the components. Some of them are special deals that people are willing to cut with us. The fixed percentage. They're taking a smaller percentage they would normally, but the only way a company like that can afford to do this is if there's a real return to their stock holders. We want a really open system, and we mean open, and so that means that if Microsoft wants to put some flavor of windows on the machine, what power to them? This is an open system. Since the economics of this are driven a lot by what the cost of textbooks are, we care a lot about being able to use it as a book. It doesn't work well as a book. You have a real problem in the economic story. If you don't get your economics, this thing can't scale. It's all about the scale in many ways. It's why scale is what the problem is. There's a building of kids out there. So much as 5 to 10 million sounds like a big number. Understand the steady state once you've equipped everybody is probably in the order of 200 million minutes per year. Okay? So we're still talking about it in their first year being a small thing, even though by any other scale. I'm not sure if we're going very good or we're going the way universities is. I started with the talks with Marguerite Jackson and some people put together. Do you have questions at the end? I can't take questions at any time. You said that the green microsoft windows in the laptop, you're fine with that? Yeah. How can we stop them? It's an online system. But would you distribute that with the microsoft windows inside? We aren't going to, or we can see it as not. Okay. So if they want to put it after, they have the answer. Ultimately, ultimately, there's no way to stop, you know, nor would we stop other things running on this machine. My opinion is that microsoft is between a rock and a hard place and doesn't have appropriate software for this kind of hardware. Yeah, but my point is that if Microsoft wants to put it after, after you have distributed and given it to the tonic. Well, we aren't directly the distributors. Remember distributors are things of the ministry of education. Okay. And so ultimately, we aren't in the final control about what goes on this machine nor can we be. I believe that it would be sad to try to create an open system that the kids can, like, disassemble until the end to the bare metal and not have an operating system that they can use. I can't. But it can force a situation. You're asking people to do something which is literally the emphasis of what open systems is about. Okay. And yeah, there may be some consequences we don't like about that, but are we going to sacrifice our own principles? But it would be nice to convince the government that at least one of the governments really would prefer to ship something other than Microsoft. Yeah, software. Will that happen all over the world? Well, you know, probably not, but that's the way it goes. Linux ultimately has to succeed on its own virtues. Let's make it succeed, okay? Yes. We're not going to do it by trying to control it. Okay? I mean, all that would do is, you know, have Microsoft more unhappy than they already are about this project. Fundamentally, we don't have to write software right now. You know, that's, you know, Windows CD is partly insecure and regular Windows XP is too big for this class machine. So it's, you know, we have a lot of experience with being able to run Linux on, you know, hand-held, remote-recorded, small environments. This is actually a lot more machine than we've had in PDA, so we'll see. Okay, so part of this whole game is how do you get to anything approximating $900? Some of the skepticism that you've heard in the press has been that there are lots of people who know what the pundit prices are. Oh, let me step back a second. I missed one important point. The other piece of this price which is floating is that Flash and DRAM are true commodities. So while we can get contracts on other parts, we can't control the price of RAM and DRAM. They're literally sold. They're future markets, the whole of all the web. So when we talk to governments, we can say, you know, what's happening is basically we're sending it to governments. You'll commit this much money. We'll get to as many machines as we possibly can. We can guarantee that everything you sell Flash and DRAM will cost no more than this, right? Flash and DRAM might cause it to be higher than our estimates. Flash and DRAM will not be next year, but as you know, once in a while, there are, you know, few interesting fluctuations in the markets. They're gonna take a time off. They're gonna take a time off. For example. That's a real example. Or Japan also, because you have problems like that. Okay. So one of the interesting questions out of all of this is, well, how do you get to something approaching $100? Because a lot of people are knowing what the prices of things are. Okay. And if you figure out, if you just think you're building a day, a regular laptop, and you're going to come up with a number that's at least a word or $200 to manufacture it, the largest single component of that is the flat panel display. Okay. Even if it's something like a 7.5, 7.5 inch flat panel display, that's where, like, you know, the cost would have to go. So Mary Lou Jepsen was noting that in fact, the costs in a conventional flat panel are not mostly the LCD, they're mostly elsewhere in the flat panel. They're in the, you know, polarizers, back lights, regular ICs, printed circuit boards, all that sort of stuff. So we're connectors. And she's really one of the world's experts at the display technology. And so she's invented a new form of LCD. It's a dual mode display. These numbers, by the way, are, as always, I never quite keep up. I should have reviewed this last night. We're actually going to, we finally have nail-bound to manufacturer of the display. And it turns out that for the plants that that manufacturer has, the sweet spot for number of, what happens is you have a piece of mother glass that you made an LCD out of. Just like the integrated circuits, the question is how many do you fit on one piece of glass? And so there's a sweet spot determined geometrically by how many pieces you fit. So it turns out we're going to end up with an 1.5-inch display of 1,200 by 900 resolution. I think that's actually closer to the room under DPI. In one mode, it's black and white, the grayscale. And it's reflective, which means that it's sunlight readable. We'll be able to take this out into the white sunlight in music. This is not for the most of existing laptops. It is really needed in the environment in which suspicion is likely to be used. You don't necessarily have a dark room to work in. The other mode is a lower resolution color display, which is transmissive. But it doesn't use a conventional backlight. It uses an LED. And we hope it's going to use this as fuck doesn't have some brightness and brightness. So the basic scheme is a diffractive scheme where you can print plastic to become a diffraction breathing and end up separating the colors from a white-white LED to illuminate the appropriate pixels in this interesting pattern. And you end up with something which will work out in order. You need three pixels for every one, obviously. So it's a lot of resolution in color. It'll be something higher than 640 by 480 in some ways. It isn't just the trivial 3-to-1 computation that you get from alone. This is before we finally set what the final resolution would be like last week on this slide or maybe it was a week before. So these numbers are slightly wrong. The basic principle is correct. The power consumption in great scale mode is less than 1 watt. Running the backlight will consume up to 1 watt. The conventional laptop with the conventional backlight is often consuming 6-way watts. So again, we're going back to the fact we want a human-powered machine. We have to have a human-powered machine. There's lots of people who do not have electricity. Lots of kids who do not have electricity at home and often do not even in schools. Okay, so here's the basic machine. Modulo, I need to update the specs again. It's a GEO which is an x86 architecture machine. 128 mega-gram half a gigabyte of NEM flash. If you're running the F2 on that, which goes data compression, so it gets a lot of... For software and typical data and non-compressed stuff, it gets about a factor of 2 compression. So one way to look out of this and you can probably shift about the amount of software to get on the lights and be with space left over for content. So the display, as I said, is a little better resolution than the one on the slide. There are three external USBs with power for only one. There's USB 2 so it's possible to plug a USB 2 described into it. I'll get to that a bit more later. We're using a Marvell wireless chip. You may never have heard of Marvell because they're not typically being used in that. They've mostly been used in embedded systems. The reason why we're interested in this is that this has an outboard R9 processor and that needs to pull by some brand. What this means is we can put a routing cable and forwarding software and run that outboard while the processor is suspended, the main processor is suspended, we can have the machine still able to put forward packets in the mesh network. Okay? Marvell thinks about this machine. This is one of them, okay? The display is the other really big Marvell thing about the machine. Sort of everyday codec, a capacity and analog devices, 1888, you know, blinding out, you know, microphone in, stereo speakers. The same number of memory keyboard that actually we found in some samples that seem to work okay. That's because of dust and dirt and water, okay? A conventional keyboard probably would not survive long enough to be to be tolerable. The touch pad on it is going to be, you know, sort of a, it's intended to be a kid's machine. So think of it as, you know, we're really trying to make it look something other than perfectly grand. So we're thinking, you know, care about the industrial design. It's nimble, though, hybrid batteries with an optional upward generator. We had to give up on the internal generator idea, which nimble really wanted to do. The realities of how much power a kid can actually produce and the ergonomics of what a kid can produce in using what sort of muscles at this power level makes sense. We intend to ship Linux BIOS on the machine. So you all actually go to see what's going on. This was just one of the early prototypes or, you know, early industrial design prototypes that pointed us that we intended to be usable as essentially a conventional laptop or as an e-book sort of thing, or I think we're just playing games on, with some buttons on the front, that sort of thing. This is another physical model that was made just to show the kinds of things. We actually now at the final industrial design chosen, I don't think I've been rendering of it. These were some recent industrial design samples. It'll look something like the machine on the left in terms of having a handle and stuff like that. Okay, so in January we got really kicked off the engineering of this project and we were staring at the reality that the machine might take between two and a half to three and a half watts of power from the run. That doesn't sound like much. Your laptops are probably consuming kind of clients as much. But the reality is that a small child can generate a lot less than or generate only 5 to 10 watts. And not for less power than they hope for less time. So we have to get a reasonable return of the effort. We don't want the kids spending all their time cranking or generating electricity. We want them to spend most of their time learning. So our goals have always been 10 amps. So there were a couple of obvious things that occurred to us. We really want the mesh network on all the time. And because we don't want people to be affectionately effective, you need to be able to forward packets for others as much as possible. And I think kids not confident there's going to be battery left when they use it. They're going to try to try to disable the wireless. And even if you don't make that easy to do they'll figure out how. It's only a matter of when. Right? So we really need a way to have the mesh able to be on and then the rest of the machine off. And that's what we're doing. So Quanta introduced us to Marvell. We didn't even know this chip existed. And the Alpharos folks weren't even going to commit their next generation for what we needed. So it was a pretty non-brainer choice. And the Marvell folks have probably been doing a great job. The driver there's a big issue for the driver. Right now we can't the current firmware still has some that loads into the Marvell chip still has a little embedded operating system yet. So we can't yet release the law of firmware. However, you know our intent is and we love how it is to replace that the whole thing with something like I mean I'll ask which is open source at which point the firmware would be open. But that would probably make very well need to help the community to actually see done. I say that because reality will be that by the time you get this machine done we'll be having to worry about the next generation machine and it won't be an issue there. So you know, it may be help me in order but right now to get out the code that goes into the Marvell chip itself you have to sign an NDA to get access to it. But it's not where we or Marvell want to be in the future. The driver itself is in a gift tree on Dave Whitehouse's machine at the moment so you can actually get out of that service then. The other observation as the CPU is is usually refreshing the screen or forwarding packets. So, you know, the CPU most of the time is idle. What do we need a CPU for? If we can have it splay on and we can have the wireless chip on. So, necessarily being another dimension we're going to turn off the CPU of most of the system and leave the wireless chip there's a chip we're building we have to build some sort of a chip anyway called the DECOM chip which drives the LCD there's, in some sense, architectural things aren't great for low power things you'd really like the frame buffer to be outwards so you don't have to have the processor on to read out the frame buffer So, the way the DECOM chip works is that it's got a small RAM chip and it takes the pixels and stores them in that RAM chip and so we will be able to power off the main processor though this has been to RAM and leave the screen running Same thing goes for the wireless So, we expect that there are a lot of things like just looking at something which just happens a lot of the time like most of you right now are not actually taking on your displays that will be in half to one and a half watt fans range so depending upon the screen the wireless fans are used so that gets us into a much better wonderful ratio It takes about half a watt to run the wireless the wireless chip you might think, oh it's going to be when transmitting but the wireless chip is the signal processing I've received on the receiving side of the wireless chip is significant so the wireless chip does data work a lot even if it's just listening Okay, so here's the sort of good and bad on the heavily Turning off the CPU gets us roughly to our power consumption goals where the DECOM and our level wireless chip makes this work and it takes about 150 milliseconds to wake up again which is unfortunate it's slow all the way through to many other chips Have you put any other chips in DO or some other or maybe so Here's the issue the experience and the buy-offs can work on handballs and I was one of the people who started the handballs.org effort on handball devices but if you see handballs on any sort of PDA or phone thing it all goes back through the handballs.org in some sense we did that with softwares and people took it over The experience is that if you don't have an FPU you have a lot of porting effort, particularly from multimedia stuff with open-source software but it's sort of tolerable when you have a very limited number of applications you sit down and you re-write the individual applications it's not tolerable on a general purpose machine so one of the requirements has been having a non-conventional porting point unit right now it turns out that the GEO is the best team in town that won't be true in your room now but it is today Mark Foster, who is our herder lead architect and surveyed 320 chips or something in January the thing that came closest to what we needed to do that to where the GEO was was a power PC which actually did have an FPU but then had a gigabit internet embedded on the chip too so nobody you know this isn't we're happy about the GEO it happens to be the least evil alternative right now and so we're calling all the people and sending it up to industry what we want and over a year it's actually possible for them to take it these kinds of highly integrated chips or chips work is that they have pre-existing designs they plop down an internet so with a year or more advanced warning you can actually get a chip that will meet our needs so second generation will certainly be lower power and I'm not going to predict what architecture I'm not going to predict what it will be and you're not even as good a sponsor we're going to have to live out to our needs and meet what we need if they're going to play in the future too just the way I guess Google has done similar things with Intel being on their board and they used to play in Egypt so in any case the bad news of course is that we can't turn off the processor between keystrokes and I guess Wednesday happens to be working on these machines but any substantial pause in your machine we intend to make the machine go to sleep we've arranged that the Econ will be able to turn the system back on almost continuously so if you've suspended the processor and the screen is off as soon as you hit a key your screen will go live again even if it takes another 200 milliseconds to get the processor done what is this one that's power management jargon for a state in which your processor is not doing any instructions but you still have your system power applied in various places S3 is actually where you turn off power to the processor and other parts of the peripherals so is it necessary to have CPU running at full speed is there some way to get it up to something that may not be full speed but it's enough to process a keystroke it's well with low power in the first place but there's no frequency there's no frequency or scaling in the geo there are on some of the other lower power embedded chips it won't execute any instruction before that time geo is a very slow back it's space locked loose stable before turning on the box its PLL stuff isn't great today that's what else happened yeah well it couldn't do but PLL just takes the external frequency just skip the PLL step no it doesn't it doesn't work doing that transition is just too hard okay so we intend to shift our money's bios on it right now we're using inside bios inside bios folks work really really I want to thank them because they had no economic motivation but they were kind enough to even though we aren't going to shift their bios they spent a couple of days to get their bios up on our hardware so we have the bios in day one getting one ex bios up on this machine has been harder than we expected it is now more or less up because of about the last week it doesn't get to the point where we just switch over to what I expected over the next month or so it obviously allows us to complete control and with the kind of power management we want to do we care a lot about that we would also like to do the boots and the saws over the wireless network experience from the handheld world is that even when we got the iPad installed down to a contact flash card that you plugged into that did a personality transplant the physical having to have that device simply going through the steps for each machine is really boring which is really like a machine that you turn on and that's the right thing the first time needing something physical to plug into the machine is another logistical headache we don't want to do that there's nowhere out there so we'll save a small amount of money there but also kids can see how computers will work probably down to the Ironman for the kids to care most kids don't care about something new and you want the kids to be able to understand how computing is a big piece of most people's lives and if we can start building $100 or $50 machines will be a big part of everybody's lives all over the world okay so this is up now on the GX2 pretty much as we will ask in a few days there's a lot of work we need help from memory consumption a lot of the current code is just slogging it's slogging both in design and bugging when we saw Linux is suffering from bloat maybe not quite as badly as Windows but it's really a problem from doing some investigation certain key applications are like Firefox are literally hemorrhaging memory the cache management code is currently effectively broken any time you use tabs on Firefox if you look at it you will see that if you consume 3 to 500 megabytes in a day's use it never is getting memory back people will finally realize there's a problem here but if you wonder why our desktop has unpredictable performance a large part of the reason is that we have a number of applications which are memory logs and or leak memory that are causing things to get caged out and so rather than getting instantaneous response at a context which you're waiting for something to page out or page in and things are slow so the work that needs to happen for applications to be useful to all of you see is work that will help everybody to about the 90% level at the Linux and Linux environment level the amount of stuff you need for all of you see is actually very small so this is all for your benefit to become much more aware of you will all benefit a lot if you go help analyze and fix these problems our Federico Menterra has started doing a lot of analysis of this phenomenon environment and making good strides even so we'll obviously avoid or won't be able to use things that are huge pigs and one of the nice things we have about the reality of our Linux is in some sense the huge number of choices so we will obviously make some choices so I'm trying to make a plea to everybody here that if you want your applications to run fast not only should they not leak but memory is accessing memory is very slow for every memory reference you make that has to go all the way around you have many many many instructions you can often make your code run two or three or five times faster is why reducing the memory footprint of your data structures so this is serious stuff folks a lot of people grew up in the era when memory was the ratio of time required to reference memory versus instructions was a lot closer to one to one than in this today it's not like can be as bad as 300 to one kinds of numbers right more typically 20 to 50 to one if you want your code to run fast make it small so there's lots more work to do here there's the amount of footprint we have on this just write your heavy fragile expensive so the three strikes are out so this is a flash based machine you got half a gigabyte this year that's $7 conventional this would be something like $1 or $30 and power another couple of watts the other problems so data compression helps some but things like huge amounts of space on the disk is an issue by the way all this sort of stuff will help the LPSP people now and that's what I would add on that this is not a particular question for one LAPTA for child product is we have probably we have schools municipalities that run thousands of think lines and the measures I get with upgrading to of school earnings was that the turn of the swap and then the think lines went black and it's not one think line it's half of the think lines and you do an upgrade and this is and the answer from the community was that well you can just buy new equipment for God's sakes but this municipalities their choice with their money is that they haven't the 38 thousand euro to do this upgrade or they haven't the time spending to add on 64 ram to the machines so this is exactly the same problem as one LAPTA for child are into also with half thick or diskless workstations and the thing is that we are competing with pupils that are used to using Microsoft in our restaurant so they are hugely disappointed because they have at home 3.2 gigahertz processor the equipment at home is just too cool so on the other side you have the municipality that won't have the money so if we could give the same user experience for the users also the machines run a whole lot better it will help everybody here thank you there's bunch of stuff to do user land like flashing cursors or updates themselves there's some stupid applications like the burn power or the emanations that go on forever flash drive will get installed we can't get the macro media folks to do something more same I told them this too there's lots of work to do and full kids that's top standards that sort of thing we'll probably want to have really know when we're doing something like a full screen ebook reader because then you know just spending all the time makes perfect sense all this sort of stuff is also currently for LTSP in the bedroom and most of this work will help just things in general there's some additional model of power tricks we intend to do we intend to drive the flat panel slowly whenever possible to save power this is not a model of us if you look at cell phone chips and panels are doing likewise it should be easy to add this to acts we will then intend to dynamically change the refresh rate of the LCD so that if you're doing things that require lots of changes to the screen while we're wearing a high speed otherwise slow it looks like to us that we need to go on day one of IPv6 the the development world has few IPv4 addresses in India there are places known to be 5 levels deep in that there are universities in India who have a simple IPv4 address now this is not for any same reason I talked in Surcoma's topic and he says that there are cultural problems applied here too that a lot of these people a lot of these countries are very would be embarrassed if they were turned down on their allocation requests in fact they wouldn't get the address allocation authorities would give them the addresses that they would add but they haven't been asking but the real point is that that multi-level having NAPL over the place will make it almost impossible to manage a large scale system or if you can't remotely diagnose a machine then you have to have people on the ground and at this scale finding enough people on the ground would be in a situation where people all over the country all over the world can potentially help diagnose problems and so we want connectivity for that reason for collaborative applications as well because things like Skype now are standing on their head to manage to avoid the current crops that have crept into insufficient addresses we're still figuring out exactly what we need here but I think we're likely to die in six system management is a huge challenge I think about this from the point of view of expertise the field expertise may not exist in many places that we need to deploy so we need much better really better than even project Athena which is where X has developed and today there are like 1700 Athena systems with three system managers which sort of eats the pants out of Microsoft and Uranus and LTSP has similar stories to tell we have to have more flexibility because we don't have a huge this the sort of solution that LTSP or the Athena environment has is to put everything into your file system because the kids have to go home at night and there's only a half a gigabyte of that and we also have a different principle we want our machines to be easy to fix rather than hard to break the reason for this is if you want the kids to actually be able to mess around with their computers and learn how to use them then they need to be able to screw them up well hopefully they don't screw them up they'll learn not to but we need it to be easy to fix rather than hard to the whole issues around security, authentication, privacy and obviously since you can't see this is the bottom bullet on my slide just the scale of the climate so you know we're really reinventing a laptop or something else I don't know what you care for but it's a very this is not just stripping out stuff out of a laptop this is a fundamentally different machine we're trying to get $100 for about $135 it is much lower power than a conventional laptop we're going to worry a lot about things like ego breathing and be able to use it in this place you use a book right so we're really in some sense following the iPod Revolution man flash being cheated some might be able to display which is also fundamentally different human power generators to recharge the batteries if you need it you know our cable sort of mode and mesh networking because we can't install access points every place we want three kids to be able to sit down underneath a tree which may give them a little bit of shape and work together with no infrastructure we want the machines to actually be usable in that case so things like zero content mvns and all that stuff our technologies which thankfully are more or less able to be used now and we intend to use them heavily we will be using them heavily so here's the first board looks like this the wireless is a little square in the upper right hand corner but it will end up underneath a can the the marginalship itself is just the left of that that's the output amplifier and the receiver I guess the GO itself is the big thing in the middle there's an AMD chip there's an extra flash chip right now the real flash machine will be booted from eventually there's a serial flash down in the lower right hand corner which is like six pins the EME if you see the EME upside down it's an embedded controller for dealing with power management there's 128 the four chips on the left have 128 made of RAM the blank space there is the quantum memorial we've never built a machine with serial flash before that's a bit they would not believe we're not comfortable with a building machine which didn't have a flash controller chip they put down the pads for a flash controller chip well Jeff, that's too his mouth so we turned on about a month ago and two days after power on Kinovics booted off of a CME so we'll see and Fedora have all run on this machine there are some changes that need to happen they're pretty minor so if somebody in the devian installer team wants to talk to me I'd love to talk to them because we have some experience about where the installers hail multiple of them that's okay, the booted installers just as broken or the wrecked installers just as broken we're basically the same reasons very few it probably has two more layers than it will in the production version but I don't remember how many they're giving another spin of this, we're going to build about 500 of these boards in the next three weeks I'll talk about that more later so here is just an early shot of a machine running with Linux in the background on the screen so there you go this was the first one of the first screen shots the camera version of it so we're building about 500 boards this summer lots of work to do on software Red Hat is doing a lot of heavy lifting right now Marcelo Tassati has joined Red Hat and it's working on all these things and it's going to be working a lot about the base system and how to do the install stuff they're also putting together a fedora derivative for the core of the machine you can run stock Debian or what not on a conventional disc the point is when you have a half a gigabyte flash you can do work right so the screen should be ready at the end of August we have time for an extra cycle of screen development so actually it looks like testing will start in September we took about a month's scope on the display so November or December is when the production testing starts and we hope to be in full production by February of next year first time we'll have packaging machines as opposed to very bored to be sometime in September the production early production units is this November or December date that's when we get back to our tool and the ejection plastic molds that sort of thing there's lots of work that you can do for these joint projects we've done lots of different projects there's this work everywhere memory usage and leaks there's criminal work to do power management out of memory behavior we really don't want to swap onto a page on the flash it really does not it turns out NAND is a lot better than NOR but I don't need that 4 megabytes a second to write to the flash it's not a good packaging device even if that flash wears out it's not going to be good in lots of pages there's obviously going to be something UI working unit for power management and making sure things work well on every scale screen there's huge amounts of work on localization there are I will note that the current localization strategy that we all have will not scale from the entire entire planet there are 347 languages in use with more than a million speakers we really have existence proofs of pretty near complete translations for smaller populations than that we need better internationalization how we do that is a non-trivial technical problem I don't need it next year I need it the year after in other words we wanted to play enough languages to really badly break things next year the current processes will work they won't work probably the year after so this is a plea for somebody to really worry about it remember the translation also needs to be able to be done offline as well as online some countries out of country network access is very expensive so nice as far as that is and I applaud you and two folks for doing it we also need offline capability also note there's no way that we're going to be able to hold up releases until we have translations of hundreds of languages so we have to be able to do language, you know update languages independently of basic systems we have to split the process it's work okay until now sort of to try to synchronize it all I claim we go another back to year 5 and a number of languages we're going to use and we need to start thinking about it obviously educational software for kids I now still have to be able to close projects at laptop.org very happy to do so we still don't have a big bit of a second link so it will be a lot more nicer than me because of the source board actually there's five big bits for second links leaving that building so different networks so it's the MIT Coal Center so I like people to work together this is a huge way that independent of what the underlying distribution is let's really try to work together on the user environment side for kids we're building a Python GPK based collaborative system the intent is is to which is using all the zero-concept and MDNS kinds of facilities so that kids can cover each other and start chatting and or be able to work together and or a teacher might like to lead people through the web so you want that sort of collaborative environment there is a headache which is I'm not convinced that it's going to be feasible to run more than one if you will toolkit environment at a time given the amount of RAM but some people of course are very attached to their desktop environments I will point out that for a preliterate child the conventional desktop is the one answer kids just learning to read means something else we're trying to build that something else please come help one of the beauties of the environment we have is way more flexible and build better and obviously Debbie and I don't think has to be told this but I always think please work upstream in the projects patches downstream often never get merged upstream right so I like the work the help done in lots of different projects that need help so as I said we're going to have several hundred developer boards in the Junits industry and we expect 100 of those will probably scrap out because they won't be worth trying to fix so we don't have production tooling equipment test equipment the fallout rate will probably be pretty high I'm primarily interested in people who are interested in helping on the issues that require the special partner right now you can do all the UI kinds of stuff on existing desktops right so this that's not a call for why you need the hardware we're saying there may be certain exceptions to that for people putting the other distributions and for political reasons we'll have to give some for the countries more directly just in part the stuff is real the Red Hat folks have put together a fedora environment that's running under QEMU to actually see what libraries and so on it's still very early days and lots of feedback is needed on what we need in the base system we need to throw not have a lot of old junk I don't want XT on this machine and motif and that sort of thing we need something that's somewhere between a conventional desktop and what we've done in these days that's sort of the end so there you go I'm done questions which URL can we reuse? laptop.org somebody was very kind to give us that name we said you like laptop.org he said sure we don't we don't we don't we don't please don't melt this is Jeff does Jeff has to scale up to this sizes of flash it turns out it does just they went house and the issue is that Jeff has to currently has to essentially see on the flash all this internal data structures those structures and memory what? those structures and memory yes it does the amount of memory it looks like a few megabytes which is not unacceptable it's not ideal in the longer run there's a new system that's under development but it's not mature enough to think about the plane the amount of time is looking about 8 seconds there's a partial full flash system so Jeff has to looks like it will scale just fine to this size it wouldn't scale to 4 gigabytes of flash you'd have to wait minutes right? yeah so yes it does that was one of the things we were worried about I was talking to Jeff yesterday he was saying this and yeah this is the other which doesn't matter really we've got some data compression so if your file system doesn't fit you can screw it in and use all this RAM so that's the trade-off jfx2 has a lot of corner cases where it takes lots of RAM so in normal life if you have a really fresh jfx2 image it doesn't take much RAM once you start fragmenting it might come really terrible the problem of why you might find using the compression when you pack up a whole thing it carefully fits it doesn't files into the gaps so you get really good compression once you start editing the files and changing them compression doesn't get as good so if you just like install an image you have to remove a few things because that can set it's now yes, the best alternative right now is far from my deals I'd say we'd like to use a file system that's underdeveloped and we'd like to use but it's not been sure enough yet jfx2 has had lots of problems in the system the file system period in 1938 256 is already a nightmare it actually has to go through the whole 560 yeah, well that's what I'm saying it takes 8 seconds to mount after a gigabyte that's measured now running on this days we're doing work to make things run faster it's better than it was there also needs to be fixed for the issue of corrupt file systems because the file system will have the jfx2 code scanned the entire disk and it can find any valid node anywhere and that takes ages yeah, well it's the best thing in town that doesn't mean anything we really understand the shortcomings what we're relieved about is not having to try to stabilize the new file system in the next 6 months it's a matter of resources for people working on an extra-generation file system that looks like it should be a much better thing but it's not so yet another thing I've learned just yesterday because I have a USB wireless stick a netgear one there is actually a project working on an embedded firmware for arm-based wireless solutions based on the prison chipset would be possible to share a bit of code with them conceivable what we're likely to put out there for the mesh networking is a little bit different than what we'll put out there at the convention they have their own firmware and they have a small stack of wireless I'd like to know more about it because I don't think that Mikael's is worried about the networking piece of the project is aware of it so please give me pointers if you will prison54.saucecoach.net I'll remember could you please scribble it on the back it doesn't start what are the policies that you install we're not aware of them I'm assuming that we want them to be installed in any place we weren't trying to install continue to do that we're just trying to put it to death who used to do that there are two problems and this is true for Debbie under America I don't know whether we'll do it or or one of them would be a cool install but the resulting images in it are being did not have USB modules in it and so we're trying to boot off the USB device the other problem actually there's a current issue as well which is that there's no way right now to know when a disk is spun up on an USB box properly so you have to put in this wall you have to put in this wall with the light to wait for the disk to be ready before you can actually run from it so both are the issues we saw them in both and one makes this for us because we're all flooring that's what we do at work and it still sucks so these are the problems we've just met as we've done this for you it sounds good this for you yeah I like this collection I was also wondering if you had any ideas of this it's also an issue that the iBase has it's 6-7 languages now it's one of them all for RAM and it's a problem it's a problem scale it's a problem so for example the strategy you put in all languages is that it's fairly broken it's already the case that translations might be bigger in the next few years for sure I doubt it it's one official script with 300k translations right I cannot afford to use languages that aren't going to be used on my flash so the moon food is done so but they really what I really want to be able to do is say I want exactly the following three lines remember kids made it to more languages so we need more than one language to install I want English and French right it's like that languages it's amazing that's the only solution without scaling problems it does that needs to be refolded and release I'm going to go to the public they are they have two years of language what we have is going to use breakage already that's going to be a very slow breakage it's a little more even it's just so full that's the rule already going I don't have I want to be able to put more, you know, another, you know, more of the Wikipedia on the machine to bring language, not with the translations for languages that won't be used. Does anyone even have a proposal? Well, there are some stuff to be learned from the London Postal Service, but I don't think it goes even far enough. And there needs to be fundamentally new thinking about how the language is there and how the release process is, because you don't want a whole lot of release or it might be a small language, you know, that... Well, in practice, we never try to synchronize, but you do try to synchronize. But for most languages, people don't follow the new district, whatever translation they're on today. Right. So the two can be separated. Remember, remember that in a school environment, at best it's going to be able to be updated in the beyond security packages of order once a year, at best. So they're going to want security during school. So this fundamentally means that the translations have to be able to come along when they're ready. I mean, you want to be able to add that translation as soon as it's available. You may have had to commit to a more recent version of the system and be missing translations for three of your favorite applications. Right. Now, that will also incent people to help. It's a two-way street. We have to make it very easy for people who are non-preparities to help in the translation process. And it needs to be able to be done not just online, but also offline. There's some offline tool whose name, I forget, but yeah, there's nothing to progress. But, you know, editing.po files in Emacs is not going to be an answer for some teacher in the leads. This gentleman can tell you that that's the case. Right. We need real user-friendly tools for this, but we really want to make that happen all over the worldwide basis. And the whole processes around the review of translations, right, we want to make that scale up. There are a lot of issues that come up in translations. Right. Think about Chinese. Right. They keep it. Yeah, so, you know, there's actually a big problem, I think, much bigger than most people post on Peripheral Translating the Word 3. Yes. Yeah.