[Community-news] OLPC News (2006-12-09)

[Walter Bender]

1. Mechanical system: Mark Foster and the hardware team have been working
hard responding to field feedback on the B1 machines. It is always very
satisfying to ship out the first build machines, but now comes the serious
work of transforming the OLPC machine into a production-worthy system. This
enhancement process will impact all areas of the system. The mechanical
engineering team, for instance, is now focused primarily on increasing
robustness and longevity, fit and finish, and on improving production
efficiency.  Special thanks to Bret Recor for his incredible efforts this
week assisting the mechanical team—his contributions were both impressive
and sincerely appreciated!

2. Electrical system: The electrical team is particularly focused on signal
integrity. Thanks are due here to AMD, for their advice and suggestions,
which will definitely improve the robustness of the system. In addition,
some power-system issues have been reported in the field, so the system's
front-end has been significantly improved to handle the wide variety of
operating environments that we are encountering. Fortunately, some of these
issues are controllable by the built-in firmware in the system's embedded
controller and so may easily be updated in the field. Special thanks to
Mitch Bradley for creating a super easy-to-use update tool that will make
it easy to rapidly update our systems in the field with the new code!

3. Software: Work continues on the wireless driver. Marcelo Tosatti has
apparently tracked down the scan hang we were seeing. We'll see a fix for
that in our builds at some point. We are getting pretty close to an
upstream merged
of that code.

4. Dan Williams, Marco Gritti, and Robert McQueen have been discussing
changes to the underlying protocol we have been using on the local mesh
network to communicate between the laptops. This is required to set us up
to be able to scale up to server-mitigated networks using Jabber, the
underlying protocol used in many chat networks, including Google Talk.

5. Robert also did some testing using the gstreamer framework and reports
that video conferencing looks entirely doable on the laptop. He says that
video streaming works smoothly with FFmpeg's H263 codec at QCIF resolution
(about 176×120) at 15 frames per second, taking up about 40–50% of the CPU
to encode and about 10–15% to decode. Increasing the resolution takes the
CPU usage up considerable, but it might be possible to do QVGA (320×240).
More testing and performance work is desirable. Audio streaming works fine
with PCMU (one audio encoding) but he wasn't able to test Speex (another
audio coding) because it appeared to be broken. We'll have to do some more
testing.

6. User interface: Dan, Marco, Ivan Krstić, and Eben Eliason held a
day-long charrette at Pentagram. The primary focus was on the journal
design and a set of first features to implement has been selected. Also
moving along has been activity-bundle implementation and the beginnings of
how the clipboard will work.

7. Performance: As Nicholas has repeatedly pointed out, for the last 15
years, software has become larger and often much slower. Ian Piumarta (and
others working with Alan Kay) has been working on an extensible, dynamic,
object-oriented language system that holds the promise of radical
performance improvement on languages such as Python and Javascript, now
heavily used by OLPC. This work seeks to both radically simplify building
such languages and achieve much higher performance. Building an
implementation of Javascript or Smalltalk in this system can be as little
as a few thousand lines of code; a simple Javascript implementation can be
done in just over 400 lines of code, yet be faster than the carefully
crafted conventional interpreter orders of magnitude larger. It is entirely
possible the technology Ian has developed will not only be used for
languages of key interest to OLPC, but even long before then, it may result
in greatly increasing the performance of graphics on the OLPC and the Linux
desktop in general.

8. Chris Ball started serious exploration of Python performance on OLPC,
and established a baseline of the Cairo graphics library performance on our
hardware, and confirmed the great performance improvements seen elsewhere
in Cairo. By linking Python differently, it appears a 40% improvement of
performance is easy to achieve. Chris also worked with Mitch on completion
of an extremely simple system update procedure. We can now update the
software entirely on an OLPC system in about 3 minutes elapsed time and 20
seconds of human time.

9. Mitch and Richard Smith will shortly release a new version of the BIOS,
containing fixes for some of the battery-charging problems that have been
reported.

-walter

---
Walter Bender
One Laptop per Child
http://laptop.org