Interview with Eben Upton, creator of Raspberry Pi



Facts
  • He graduated from Cambridge University, a bachelor of physics and mechanical engineering, also has a Cambridge diploma in computer technology and has a Ph.D.
  • He was one of the founders of the game studio Ideaworks3D, and also worked at IBM.
  • The author of several articles and even books (for example, the Oxford Rhyming Dictionary, co-authored with his father - Clive Upton).

Destination Raspberry Pi

Schools don't need the Raspberry Pi at all, kids need it. This is the whole idea: this is a personal computer for the child. Personally, I have been programming since I was ten years old and have been doing it for twenty-five years. Lord, how old I am! I'm almost dying! I turned thirty-five, and I still program four hours every night. And it is thanks to this approach that I generally know how to program.

We are closely following the situation with computer science in schools. At best, a school can devote one to two hours a week. No one will become a good programmer in two hours a week. You need to have a computer in your room, and constantly poking around with it. Of course, many today have a PC, but many still do not. In Britain, as in Russia, many people do not have computers at all, but there are televisions.

The idea was to create a machine that can be programmed for four hours in the evenings. It was supposed to be very cheap and durable. Even if you broke it, what's the difference? This is not the same as breaking an iPad. Got a blue haze? Well, okay. You can just buy a new one.

BBC Micro has certainly become a source of inspiration for us. In the 80s, it was an extremely popular computer in the UK. Not only was it cheap, he also knew how to work with the equipment that everyone already had at home. No need to buy a monitor - use a TV. No need to take a special drive - take a tape recorder. This is a very simple and at the same time powerful concept. We understood that if we create a computer worth $ 25-35, which will require additional devices for another $ 100, which you don’t have, then we have failed in our task.

BBC Micro is the ideological inspirer of the Raspberry Pi and the most popular computer of the 80s.
That's why the Raspberry Pi is initially compatible with ordinary mice and keyboards. All that is needed can often be found at all in the garbage dump. The possibilities are huge. Even if you have an old TV with a conventional composite connector, you can use a metal hanger instead of a cable :).

The platform that we had before was created primarily for educational purposes. But then we saw how experienced users fit Raspberry to play videos, run browsers on it, and use Java applications. Often on RPi they launch highly specialized network applications. In general, now Raspberry is used in much more complex and demanding projects than we initially expected when working on an "educational computer."

Now our audience is divided approximately 70/30: 70% of “adult geeks” and 30% are in the field of education. It’s difficult to say more precisely, this is a subjective impression that develops from what we see on the Internet.

If we can call 700,000 users out of a million experienced users, then another 300,000 users remain in the field of education. And much more than we ever planned. It is great that we entered a larger market than we had originally expected, and now we are in an even larger market and remain there.

The good thing is that experienced users contribute. They not only take. In the first three months, we had two or three people offering very trivial 10-line patches, small changes to the source code of the kernel. But later a rapid increase in the number of patches began. If you take the Raspbian image from April 2012 and compare it with August 2012, it will become obvious how much users have done.

The Iron Side

For all time, about six different versions of the boards were released. The first version was based on the Atmel ATmega microcontroller. In 2006, there was also a Broadcom-based version for developers. Then we got the name Raspberry Pi; on it we launched Python.

There was another version, very small, it was shown on video with David in May 2011. This tiny machine can hardly compare with the real Raspberry Pi, but Linux could be installed on it.

Later, in August 2011, we got alpha boards, and they were already cool. From an electronic point of view, it was almost the final version. Only fifty people had such boards; now it is already antique. I have one such. By the way, I can click on the button in Broadcom and make ten more of these fees if I want to secure a carefree pension :).

But, of course, if a year later, looking back, I would have done a lot differently. For example, I would like to avoid all problems with USB. We will resolve this issue soon, but still.

The fact is that many users are limited by the limitations of our USB-stack. So now this is the highest priority for us. It was assumed that the Raspberry USB ports would only be used to connect keyboards and mice, but that was naive of us. We had to work hard on performance and stability, much more than originally thought.

We have already made a number of changes to USB. Today, most USB devices work correctly. USB mass storage devices work well, network devices too. There were some problems with the webcam and some exotic mice and keyboards, but they almost got over it. Older devices with USB 1.0 also function normally.

We have a bunch of small patches proposed by the community. They solve problems like memory leaks, concurrency problems. There are three interrupt contexts in the Linux kernel, and they all work together, so there are many conflicts and race conditions. The ARM architecture provides two types of interrupts: IRQ, FIQ. On Linux, FIQ is not used. But we learned how to transfer the most time-critical operations to FIQ.

In general, I believe that we initially chose the right chip for RP. It's not just that I work for Broadcom (I'm still a Broadcom employee, they hired me for Foundation as a permanent employee). I'm just not sure what you can find to find an even cheaper chip. It is not a fact that if we chose a different chip, it would work at the same level, certainly not from the point of view of multimedia. So it was a good choice.

By the way, this year we plan to release a slightly updated version of the board. So far we have not decided when exactly. Changes will be small. We want to refine food intake. We have USB connectors on one side of the board, but they protrude too much, so we drown them a little. You have no idea how hard it is to do it, but it will look so much prettier. Some other connectors also protrude; they are difficult to fit into the case. Therefore, we want to “juggle” the connectors a little. In general, we plan to change something this year, but the changes will be insignificant.

But, returning to the issue of energy consumption, I want to clarify. Now the input voltage of 5 V is converted with the help of LDO to 3.3 V. We hope that we will be able to generate 3.3 V more efficiently: maybe using a switching power supply. It all depends on the cost, which we still pay attention to. LDO can be seen on the board, this is a very large and thick LDO (marked as RG2 on the board. - Ed.). If you take an infrared photo, then the LDO on it will blink like a beacon. We want to fix it. It only takes 300 mW, and this is a big difference, because it could have been fed to a USB port and connected to a faster device. Plus, the board will heat less. And of course, we need to learn how to work with cheap power supplies. Now, if the power supply is unstable, this leads to difficulties in the operation of the entire board. Generally, пока мы сконцентрировались на небольших нюансах — внешних устройствах на плате, но не на ее «вну- тренностях», не на архитектуре. Of course, all this takes a lot of time, because this is development, we must find a good technical solution. We have to discuss everything with component manufacturers to find cheap options. To do all this work with nutrition, we have to save on something else. But we have in stock a couple of good tricks to save 50 cents on each board, and we can spend this money on better nutrition.

But still, it would be nice to return on the day of the first launch. If then we knew our current volumes, we would know what to do. I would like to immediately use 512 MB of RAM, because it was not so expensive. It is a pity that we did not put LED on the RJ – 45. Instead, we have an LED on the board, but if you put on the case, it is not visible.

In general, now I would change simple things, and not some frills. I would like to put a large capacitor somewhere else. People constantly touch it with their thumb when they unplug the power cord. I remember the story of Don Cobbley, one of our engineers. In December 2011, we made a small batch of rev 1 boards, they were assembled manually in England. I gave one to Cobbley, and five minutes later he called me and said: "Well, my capacitor has broken off." Although he still has a fee, he has been using it for a year and a half without any problems.


Eben Upton

Open (and not so) source

We work on several open source projects, provide them with financial support, and cooperate with them at a technical level. Examples of such projects: Wayland, Weston (Wayland composite manager), Pixmap, Smalltalk, Squeek. That's all for now, but I believe that a lot will be done around Python, around interaction with other platforms. All of these projects are supported by the Raspberry Pi Foundation, as they are required by experienced users.

Open Source will have to interact very closely. A large audience brought us additional funds, but since we are talking about enthusiasts and hackers, we were forced to return a significant part of the money to the community.

But, mentioning Wayland, you can not say about the X Server. Of course, it would be nice to have acceleration X, GLX and other convenient things for X. On the other hand, Wayland is a movement in the right direction. If you don't need network transparency, Wayland is what you need. Plus, we must take into account that we have limited resources, that is, we have to focus on one thing. Again, we have few graphical software with which we must maintain compatibility. There is no “five year X program” that RP users should have.

In a word, we do not have old software, and resources are limited. If we had two options and we could afford to invest only in X or only in Wayland, the choice in favor of Wayland would be much more logical. The people who work with us in this area are really the best. These people have a huge impact on the project (Wayland).

I recently saw screenshots and hope to get a demo this week. We will add Wayland to our new SD card ... Wayland will generally have everything that works so that people can understand what we are doing. For now, it will be just a demo for technical users. I hope that in the next 3-4 months we will be able to make it work. We will do everything we need - 3D in the application window, 2D in the application window, improve the interface. We’ll select the top 10 applications that people use, and we’ll work on finishing them, up to Minecraft. Achieving stability, and in the end, Wayland will become our standard desktop.

We are testing only on the version of Raspbian with optimization enabled for our mathematical coprocessor (hard float). The rest of the distributions are brought in by the developers of these projects. Obviously, the two largest projects are Fedora and Arch.

Of course, I cannot but say about DTS. I think we will have DTS, but I don’t know for sure and I don’t want to think about the future. I am discussing this issue with the DTS, but we have not yet reached an agreement. I hope we get it. On the current chip, we have a DSP for decoding, and we have already done some work with it. Indeed, similar chips are supplied in many mobile phones, and all of them are needed for decoding DTS. So we will work. We have a 250 megaton DSP that can decode a 20 megaton DTS. That is, this is the largest multi-stream DTS that can be found. And ... yes, this situation is quite annoying.

So far, DTS is simply disabled because we do not have a license. We are working on it with all our might. We can say that media centers lack only this small detail. Although, it would be nice to improve the search on the track. Now you can only jump back and forth. But, in general, when we have DTS, everything should go much better. The fact is that we want to do everything legally. Although you can now get devices that play DTS, but we suffer from the fact that we want to do everything “in white.”

Regarding the graphics core driver, I will say this - this is a Broadcom solution. I continue to argue and discuss this with Broadcom, and I hope that I do it for good reason. In general, reverse engineering projects already exist, so there are practically no secrets left. Yes, and I do not believe that something threatens Broadcom's intellectual property. However, this is a Broadcom solution, not mine. They do what they see fit. I hope we can somehow influence them.

When we released the source code last year, the reaction was mixed. We have seen both a very negative reaction and an extremely positive reaction from people whom I respect. But on the whole, it all demoralized us a little. Personally, I thought: why am I doing this, why am I working so much?

Broadcom worked with many volunteers in the evenings and weekends to complete this release. A lot of people spent their free time, but as a result, most of them were left unhappy. We are doing everything in our power. Therefore, whenever I meet with a company, I make a fuss about it, maybe for nothing, I don’t know. The technology is not so new, in general, this chip has long been used. I hope that over time we can do something about it. The Pi Store, you can say, has not yet met expectations. However, we always knew that it would take some time. Now I just think that it will take longer than we expected.

Of course, we are interested in becoming a great platform for free applications. If you look at the Pi Store and some Google Play or App Store, they move in different directions. For some this is a commercial area, for others it’s free. We will stop for free. If children want to have gaming software, you need to give them that opportunity.

We are looking for ways to add more things to the Store. We plan to develop some kind of compatibility layer with one of the mobile platforms. Then it will be easier to create mobile content. Hope this helps. We hope more people will make open source software packages and add them to the Store. Then you can simply double-click and get the program. I would be very happy if we had thousands of applications, but so far there are only 60 or 70. But we are working on this because we want children to have more choice.


David Bren, a member of the Raspberry Pi Foundation and creator of the Elite game with an early prototype Raspberry Pi

Raspberry Pi Foundation

The Raspberry Pi Foundation is five trustees, something like an organization’s board of directors. It employs people and owns 100% of the business. That is what lawyers told us to do. In Britain, if you have a non-profit organization that is engaged in something other than collecting money (in our case, trading), then you must create another company that will belong to this NGO. This is our structure.

Our charitable organization has five trustees, the company has five directors. The Raspberry Pi Foundation employs two people, four in the company. They have a common office, that is, a total of six people.

Plus we have several freelancers. There is a guy named Ben Evison, he is a very talented assembler programmer (author of the RISC OS port on the Raspberry Pi. - Approx. Ed.). We hired him as a freelancer. There is another guy named Tim Relange from Vancouver, Canada. He is working on the Small Talk dialect, Scratch. On it, children are taught programming. Tim worked a lot on Scratch (to be precise - on a virtual machine), which is used there.

Also, several people work for us on a contract basis to maintain Google+ and Facebook. Another special person connects us with people on Ebay who violate copyright. Because it is very important to protect your brand. I know how this happens in Russia, but in Britain, if you do not protect your trademark, they will simply take it away.

Interestingly, both of the last specialists are women. They both have young children, and they combine work with caring for them. This is a great way to work with very, very smart people: you need to give them flexible working conditions. Otherwise, these people would never work for you, because they should look after the children.

In a word, seven to eight people work on freelance at any given time. Now, maybe eight or nine, because we have a couple of additional employees who came to us from our business partners.

Sales

The main thing in RP is that it weighs about 42 grams. So the RP box weighs two kilograms, so it’s very easy to move, right?

About 1.1–1.2 million Raspberry Pi have already been sold to date. Somewhere in February there was a million. More than a month has already passed since then, and we sell at 100-200 thousand per month.

Our largest market now lies in the USA. Traditionally, sales were distributed as follows: 1/3 in North America, 1/3 in Britain and 1/3 in the rest of the world. But now America dominates by a wide margin. There it works according to the viral principle: people buy RP, show to friends, like it and friends - bam! Plus, we are widely represented in the media, and RP is in many hacks. (Hereinafter, data for only one of the two distributors, Premier Farnell. - Ed.)
In January, we sold approximately 3,000 RP in Russia. And in total, at the end of January, 900 thousand devices were sold. Of the 100,000 sold later, about half went to Europe, about a third to the United States and Canada. As a result, the United States and Canada give us approximately 300 thousand sold RP. Europe - about 400-500 thousand.

We have some progress in Russia, this is obvious on a map that shows where all the RPs are now. Many of them are in Russia. Although, for example, in Italy we sold twice as much.

In general, we did not expect to sell a single RP in Russia. On the other hand, when you look at how many people live in Russia, what are the traditions of computer research and technical development in Russia ... Probably, we should just sell you millions of RP :).

Now we are going to "go" to Japan, we want to gain a foothold there. We are not very strong there yet, they have other guys. But in Japan there was interest in us, so you need to at least develop a little in this territory. We try every couple of months. We want to sell not only in Western Europe and North America, we want to sell in Brazil, Russia, the Philippines, everywhere.

We always rely on local distributors to buy the RP and distribute it locally. One of the problems in Russia lies precisely in the fact that when you buy an RP, it is often delivered to you from another country, but is not sold by a reseller within the country. We are interested in more such resellers, say, as in the USA. In the States, this is the case: you can buy RP from dealers, or you can from resellers, which makes the market more competitive. These guys cannot overcharge because they are competing with each other. We need to look for ways. In electronics stores, the price will be twice as high as ours, which will fill the gap, but it will be bad for us. We will definitely do something with this, but it will take some time.

One of the great features of working with our distributors: they are global companies. Even when everything is not perfect, when you have to pay extra, they are still available. After all, we have a kind of business: we do not make computers, we develop them, and then we license production.

Subscribe to Hacker



P.S. Можете поделиться знаниями и интересными идеями, написав для ][? Дайте знать :). Мы платим гонорары, но это не должно быть главной мотивацией.