[Community-news] OLPC News (2006-12-16)

[Walter Bender]

“Our strategic goal is all children should access to communications
knowledge within a frame of equality.” -- President Tabaré Vázquez

1. President Tabaré Vázquez officially announced his and the country's
intention to enter into OLPC and to provide every child in Uruguay with a
laptop within two years. He was joined by the minister of industry and
mines, the vice-president, the minister of education, and the president of
LATU (the national coordinator for OLPC). He invited a group of
schoolchildren to unveil a B1 machine for the gathering. He stated this
would be by far the greatest accomplishment of his administration. He ended
with a story of how when he was a child, growing up poor and attending
public school, when they would see a big car driving down the street, they
knew that they had the opportunity to achieve and even to become a doctor
and then president, as he did. However, he says now these poor children
know they do NOT have that opportunity. It has become no longer possible.
OLPC changes that and restores this opportunity, enabling any child to
learn and to become what she or he desires. The link to the text of the
address (in Spanish) is:
http://www.presidencia.gub.uy/_Web/noticias/2006/12/2006121402.htm

2. Brasilia: The Brazilian government convened a new educational strategic
advisory group composed of the many experienced people who have done
innovative work on computers and learning, some of whom for more than 25
years. Some of them worked with Paulo Freire and Jean Piaget, and one, Jose
Valente, was a student of Seymour Papert and Marvin Minsky at the MIT AI
Lab. The group was unanimous and effusive in its support for the OLPC
concept and is committed to helping the government on all aspects towards a
successful deployment.

3. Håkon Wurm Lie reports that the Opera browser is now running nicely on a
B1 machine (Instructions can be found on our wiki at
http://wiki.laptop.org/go/Web_Browser).

4. This week Marco Gritti and Dan Williams started work on the Journal
implementation, based on the design work that was done during the previous
week. This includes the beginning of the UI and the underlying storage
framework.

5. Wireless: Work continues on the wireless driver.  Marcelo Tosatti has
been fixing more association and scan hangs and is pretty close to getting
the driver upstream.  He's also cleaned out a large amount of unused code.
Once that driver is in a workable state, Marcelo will be spending time
working on the out of memory problems that will eventually rear its head
and how we handle that at an application level.

6. John got a lot of work done on the images as well. We've moved to the
current set of Fedora Core 6 updates and we've got a new kernel in our
builds.  We've also got builds of the new Cairo and Pango packages and
we're going to see if they make a difference in the rendering performance
on the machine.

7. Boot time: Chris Blizzard spent some time learning about the device
manager that we use on the machine. There is a lot of room for
optimization: during startup this segment takes about 15–20 seconds to
start up; it is possible we can get that down to 2–5 seconds. In addition,
Chris has also done some investigation in how much time it takes to start
up the other services that are required on the machine.  In conjunction
with the work that David Zeuthen continues to do, a 30-second startup time
is not unreasonable.

8. The news from the AMD team is excellent process on the X driver; e.g.,
it now supports dynamic rotation of the screen. In the process, Jordan
Crouse extensively cleaned up the driver; he fixed memory allocation,
improved setting the mode, fixed debug messages, etc.

9. Performance is now becoming high priority. Chris Ball has updated the
tinderbox to work with our B-test setup, and added tinderbox tests to log
boot time (tracking times for each of LinuxBIOS –> OFW –> kernel –> init –>
X), and to run Python and Cairo benchmarks, to give us a baseline ahead of
adding Cairo and Python performance improvements.

10. Andres Salomon merged 2.6.19.1 into the olpc-2.6 tree. We don't have it
in an image yet, but it appears to run just fine. He also finally fixed Bug
#520 (No keyboard or mouse input under Qemu or VMWare) and the
i8042-looping-infinitely bug (with a proper fix, not a workaround),
released a new kernel, and had j5 incorporate it into Build 196. He has
discussed the issue with upstream: the proper fix is not upstream yet, but
we're pushing for it to get in now that they're aware that the bug actually
affects people (us), and that it doesn't seem to break anything else.

11. Firmware: Mitch Bradley made lots of progress on firmware startup and
power management, which are integrally related due to the common
requirement of Geode MSR setup. You have to do about 300 MSR accesses in
order to get the two Geode chips configured after any kind of power-up or
wakeup. LinuxBIOS was taking about 7 seconds to get everything initialized
before it jumped into Open Firmware. Mitch has that early startup time down
to much less than a second (too fast to measure with a stopwatch), and has
implemented a firmware fastpath boot scheme that cold-boots into the Linux
startup sequence in about 2 seconds (from NAND). That time is down from 23
seconds.

12. Fastboot requires a special boot partition on NAND.  There are three
reasonable approaches for partitioning; the current front-runner is a
straightforward partitioning scheme recommended by David Woodhouse, the
same scheme used by the RedBoot loader.

13. The MSR init sequence that is common to both cold boot and
resume-from-RAM (or disk) is down to about 300 microseconds.  At this point
Mitch thinks the PLL startup time is going to dominate the resume-from-RAM
time. So far Mitch hasn't found any unresolvable impediment to our goal of
super fast suspend/resume.

14. Mitch also has OFW working in virtual mode (doesn't require fixed ram
addresses) so that, if we choose, OFW can stay resident after Linux starts.
The virtual-mode version is also useful for integrating device-tree support
into Linux.  A friend of Mitch's is working on porting the PowerPC Linux
device-tree support into the x86 world.

15. We released a firmware version with the new A62 EC bits from Quanta.

16. Mitch did some performance testing on the Linux start up. It appears
that booting from USB is quite a bit faster than booting from NAND. Just
counting the Linux time, USB gets to Sugar in 70 seconds, whereas NAND
takes 105 seconds. The slowdown is more apparent after the X startup
initiates, suggesting that X+Sugar is more I/O bound than core Linux. This
shows CAFE in ASIC form in BTest-2 and beyond will also improve start up
time. In its current FPGA form, CAFE is significantly faster than the
Geode's built in flash controller was, but still only a fraction of what
the ASIC's performance will enable.

17. Chris Ball tested/described the upgrade procedure from A-Test using
Mitch's updater. He tested our OFW-over-Ethernet (telnet) path with Mitch,
using it to debug a board with a bad DCON; this will help for other
failures, since we cannot use serial at the moment.

-walter

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