Contributors program archive: Difference between revisions
(→Goals) |
|||
Line 88: | Line 88: | ||
Our goals will vary as the hardware matures. For the A-Test boards right now, we need the most help on: |
Our goals will vary as the hardware matures. For the A-Test boards right now, we need the most help on: |
||
* |
* power management in the device drivers: for us, every joule matters, and a simplistic "oh, we mostly have most of a chip turned off, maybe" isn't good enough. We want to know that every possible power savings has been realized, and that suspend/resume is rock solid and blindingly fast. |
||
* fast suspend/resume: We |
* fast suspend/resume: We must go beyond the current state of the art as discussed at the [http://lwn.net/Articles/181687/ power management summit]. |
||
* modal operation: if certain applications are full screen, the system should automatically suspend and resume whenver idle for more than a short period. |
* modal operation: if certain applications are full screen, the system should automatically suspend and resume whenver idle for more than a short period. |
||
* display driver: we must convert from an XAA to EXA driver for the X Window System. |
|||
* variable speed display driving: (aka: mode change on the fly), again, to save power. |
* variable speed display driving: (aka: mode change on the fly), again, to save power. |
||
* DCON support |
* DCON support: this includes transferring responsibility for screen update to the DCON, even while the processor is still on, in order to allow us to disable the Geode video drivers and save power. |
||
⚫ | |||
* fbdev driver for the machine: fbdev exists for the GX1 and LX, but we need an LX2 version. AMD has started work here. We'll need to finish this up, and also support the DCON chip. |
|||
⚫ | |||
⚫ | * input driver: we have a novel, dual mode pointing device, which needs support in the window system and applications. And besides the built in pointing device, we'd like to ensure that common USB HID input devices work well. |
||
⚫ | * file system: for first shipment. Dave Woodhouse is improving on its existing performance and memory usage. Additionally, there is work underway at some large computer vendor for a third generation flash file system, though it is unlikely to be ready in time for initial shipment |
||
* wireless: we will be deploying mesh networking. Serious experimentation in this area is in order, to shake down the drivers and to gain experience in its behavior in differing conditions (e.g. rural areas with low noise characteristics; busy metropolitan areas). We understand that to do serious tests, more than a single board will be needed. Please be realistic in your expectations: two boards is not interesting; two hundred boards can't be provided. |
* wireless: we will be deploying mesh networking. Serious experimentation in this area is in order, to shake down the drivers and to gain experience in its behavior in differing conditions (e.g. rural areas with low noise characteristics; busy metropolitan areas). We understand that to do serious tests, more than a single board will be needed. Please be realistic in your expectations: two boards is not interesting; two hundred boards can't be provided. |
||
* compiler optimization: if you are a compiler wizard, we understand that the Geode lacks a specific back end code scheduler, which limits performance, particularly FP performance. We'd love to see work go on in this area which would help everyone. |
* compiler optimization: if you are a compiler wizard, we understand that the Geode lacks a specific back end code scheduler, which limits performance, particularly FP performance. We'd love to see work go on in this area which would help everyone. |
||
* codec optimization: we're sure there is room to be gained on performance here that may be Geode specific. Remember that CPU cycles and memory references (particularly memory references to main memory) translate to power consumed. |
* codec optimization: we're sure there is room to be gained on performance here that may be Geode specific. Remember that CPU cycles and memory references (particularly memory references to main memory) translate to power consumed. |
||
* tickless operation: |
* tickless operation: there are patches out of tree about to be integrated into Linux that allows Linux to function without periodic tics; we've been experimenting, and while our tick rate is probably the lowest yet observed on a Linux system, it can and should go further. Nothing in the system should poll! |
||
* power management desktop interaction: applications need to be better aware of their power usage and requirements, and communicate this better to the system. |
* power management desktop interaction: applications need to be better aware of their power usage and requirements, and communicate this better to the system. |
||
* The OOM (out of memory) killer is naive, to say the least. |
* The OOM (out of memory) killer is naive, to say the least. |
Revision as of 01:15, 16 February 2007
OLPC Developers Program
Background
BTest Systems
We have had the two builds (of three builds) of full OLPC laptops built, for "BTest". The BTest-1 build has been distribution: the BTest-1 Release Notes describes the state of the first build of machines. The BTest-2 build is starting distribution as of February 12. The BTest-2 Release Notes describes the state of the second build of machines.
There is typically one or two builds of systems internally before an external beta test, so our BTest-1 is really beta test for the electronics, while alpha test for the new screen, touch pad, industrial design, and keyboard; this is earlier in the development cycle than systems are usually made available, and is correspondingly rougher.
The BTest-2 hardware is continuing beta test for the electronics, beta test of the new screen (this time with a diffuser improving it futher) and touch pad. The industrial design is only somewhat improved from BTest-1; most of the learning from BTest-1 on mechanical improvements could not be incorporated in time for BTest-2 and so BTest-3 will be significantly more rugged than BTest-1 or BTest-2.
The focus for the developer program BTest-1 machines will be software development on GUI related projects that need to understand the screen, the touch pad, and/or the camera in the system, along with wireless testing, which has been difficult to do sooner due to the cumbersome nature of bare PC boards.
BTest-2's focus will start the process of testing the mesh network, and we're also working on suspend/resume, though as of this writing, suspend/resume is not yet running.
Machine are also being allocated to launch countries and do not come under this program, which is aimed at individual free and open-source developers or research organizations interested in contributing to furthering OLPC's goals.
ATest Board Types
We've had approximately "ATest" 500 developer boards built, to jump start serious development in the free- and open-source software community and the initial deployment countries. Quantities of this generation of boards is limited as we do not have production test fixtures. Note that these are bare printed-circuit boards. At this time, we still have a limited number of boards available, and some more may become available as BTest systems replace many uses of ATest boards.
The hardware specification of these boards is set.
There are a number of ATest board types in the wild:
- 30 Pre A-Test PCB, which have been built and distributed.
- 20 A-Test PCB's, which have been built and distributed.
- 485 A-Test PCB's, which have been built and are being distributed through Brightstar. These can be distinguished from the first 50 boards in that they have serial-number barcodes on them.
These boards lack DCON chips and instead come with the standard VGA connector you'd find on the back of a desktop or laptop computer. We expect that production boards may ship with pads for VGA connectors, but not the connector itself. Additionally, the Pre-A-Test boards have populated mini-pci connectors; on the A-Test boards, these mini-pci connectors have been left off; the pads are included and will probably be included on production boards.
The A-Test boards include a socketed ROM chip for BIOS development; this socket will not be on production boards. The sockets are empty; the BIOS is stored in the serial flash chip that is interfaced via the embedded controller. The process for updating the serial flash under Linux is not yet available and at the moment involves booting DOS and updating the flash chip using a utility from Quanta. Unless you are directly involved in BIOS development, you should stear clear of BIOS updates unless and until instructed by responsible people working on behalf of OLPC.
At this time, the hardware is all believed to work (having been tested), but not all drivers are working properly under Linux. The wireless driver is being completely revamped at the moment.
See: Notes on using the OLPC developer boards.
Setting Expectations
Do not expect:
- Monitor or flat panel
- disk, DVD, CD or USB drives
- keyboard or mice
- powered USB hubs that may be needed for use with some peripherals
- USB ethernet adaptors
- other input devices
We expect you have or can acquire these locally.
Setting Expectations for BTest Machines
Please carefully read the BTest-1 Release Notes or BTest-2 Release Notes and the BTest-1 Demo Notes.
If you get a BTest machine(s), expect to get (a) box(s) with:
- one or more BTest systems, with localized keyboards if available
- one AC power adapter for each machine (to the extent possible by our logistics and crystal ball gazing) which will have the right plug type for your country if available
- one battery pack for each machine
- factory or OLPC pre-load of some version of the software: you should plan to immediately update the software to a current version upon arrival. This has been made extremely easy with the Autoreinstallation image.
Setting Expectations for ATest Boards
Expect to receive (a) box(s) with:
- one bare OLPC A-Test board, in a static protective bag, with static warning, with serial number both on the box and on the board.
- one power supply brick (U.S. plugs), 15-Watt capacity. Note that Sony power bricks should also work fine, and that if you can find the connector (the Sony appears compatible) you can use many different DC voltages.
- RS-232 cable adapter to DE-9 male connector for serial console use and debugging
- a small plastic bag with standoffs for the board, along with a diagram showing where they should be inserted
- one pair of 802.11 antennae, which will need to be connected to the board before use.
Project Hosting
If your interest is primarily on doing some systems level or on applications level coding, then join one of the projects on our Hosting Wiki. We have much more flexibility, bandwidth and CPU available than alternatives like SourceForge, and your project won't be as lost among thousands of other projects unrelated to OLPC. If your project has aspects related to OLPC, but is primarily part of some other project (e.g. GTK+, X11), we're also happy to provide more limited OLPC related facilities, such as bug tracking and our wiki.
Hardware Schedule
In this first generation of boards, which we call A-Test boards, the hardware is fully functional except that video is VGA out, rather than using a flat panel with the DCON chip which appears in the BTest systems.
Packaged BTest-1 machines were built in late November. The BTest-1 systems are fully functional, but use an Altera FPGA in place of the CaFE ASIC which is present in later builds, for NAND flash, camera, and SD interfaces. This FPGA has lower performance and consumes much more power than the CaFE ASIC does. Another build (BTest-2) occurred in late January using the completed CaFE ASIC. The third BTest-3 build, in larger quantities, will sometime in the spring. This final BTest build is the first point where working with children starts to make sense, as the software matures.
Goals
Our goals will vary as the hardware matures. For the A-Test boards right now, we need the most help on:
- power management in the device drivers: for us, every joule matters, and a simplistic "oh, we mostly have most of a chip turned off, maybe" isn't good enough. We want to know that every possible power savings has been realized, and that suspend/resume is rock solid and blindingly fast.
- fast suspend/resume: We must go beyond the current state of the art as discussed at the power management summit.
- modal operation: if certain applications are full screen, the system should automatically suspend and resume whenver idle for more than a short period.
- variable speed display driving: (aka: mode change on the fly), again, to save power.
- DCON support: this includes transferring responsibility for screen update to the DCON, even while the processor is still on, in order to allow us to disable the Geode video drivers and save power.
- input driver: we have a novel, dual mode pointing device, which needs support in the window system and applications. And besides the built in pointing device, we'd like to ensure that common USB HID input devices work well.
- file system: for first shipment. Dave Woodhouse is improving on its existing performance and memory usage. Additionally, there is work underway at some large computer vendor for a third generation flash file system, though it is unlikely to be ready in time for initial shipment.
- wireless: we will be deploying mesh networking. Serious experimentation in this area is in order, to shake down the drivers and to gain experience in its behavior in differing conditions (e.g. rural areas with low noise characteristics; busy metropolitan areas). We understand that to do serious tests, more than a single board will be needed. Please be realistic in your expectations: two boards is not interesting; two hundred boards can't be provided.
- compiler optimization: if you are a compiler wizard, we understand that the Geode lacks a specific back end code scheduler, which limits performance, particularly FP performance. We'd love to see work go on in this area which would help everyone.
- codec optimization: we're sure there is room to be gained on performance here that may be Geode specific. Remember that CPU cycles and memory references (particularly memory references to main memory) translate to power consumed.
- tickless operation: there are patches out of tree about to be integrated into Linux that allows Linux to function without periodic tics; we've been experimenting, and while our tick rate is probably the lowest yet observed on a Linux system, it can and should go further. Nothing in the system should poll!
- power management desktop interaction: applications need to be better aware of their power usage and requirements, and communicate this better to the system.
- The OOM (out of memory) killer is naive, to say the least.
We're sure you are brighter than we are and have seen what we're missing in the above list.
You can contribute in many areas which do not require hardware. For example:
- memory usage: many applications and toolkits waste and/or leak memory. Fixing these will help everyone and most easily done on conventional systems.
- performance optimization: fixing memory usage will usually result in faster code.
- toolkit adaptation: the display's effective resolution will change from grayscale to color mode and back. Toolkits and applications need to be able to adapt, and themes that work well in both modes verified.
- UI: most of the user interface work can be done today on conventional Linux desktops. But our system will also have an e-book mode, with dual 4-direction keys and enter. Key applications will need updating to work well in this environment (e.g. evince, web browser). Testing application's behavior under grayscale conditions and making whatever changes are needed would be helpful.
- Applications: goes without saying. The "Sugar" environment under development can be run on conventional desktops.
- IPv6 support, and service discovery, which are very important to us.
- Security: SELinux may be a way to protect against Day 0 attacks; as a large ecosystem of similar machines, it is something worth seriously worrying about.
We will give preference for ATest hardware to proposals that require access to the OLPC hardware to make progress; BTest requests should either require mobile use of the systems or be focused on GUI and applications. At this time, we do not know exactly where the line will be drawn, so we are interested in seeing what everyone is interested in doing.
Qualifications
We're looking for people able and interested in helping in development. The qualifications needed depend strongly upon where you are interested in working: for example, people working on BIOS/boot paths should be seriously "friends of the electrons", and not scared of JTAG and similar kinds of debugging.
Most driver work takes normal driver debugging skills, though getting power management right can be more challenging than most driver development.
Window system development requires X experience, and so on.
Idle machines
If you no longer have time to contribute to the OLPC effort, please be so kind as to return your board for redistribution.
How to apply
Please send mail to the developer at laptop dot org email alias with the following information:
- Name
- Email address
- Employer (if any), University/College
- Shipping address and instructions
- name,
- address, (cannot be a post office box)
- city,
- postal code,
- country,
- telephone number, (yes, we really need this for the shipping companies)
- any special instructions
- Description of your plans for the machine(s)
- Quantity of machines desired
- Description of your experience, both with hardware and software
Presuming your request is approved, a mail message will be sent to you with shipping information, or a regret. Note that some requests may be more feasible and applicable later in the project, when we more machines available.