[Community-news] OLPC News (2007-03-03)

Sat Mar 3 11:03:23 EST 2007

1. Abuja, Nigeria: A significant milestone was reached when
approximately one- hundred laptops were handed out to children in
Nigerian test school, Galadima. The laptops were received with smiles,
curiosity, and giggles. The most popular feature in the first hour the
children spent with their laptops was the mesh view. As of this
moment, one-hundred families in the Nigerian Galadima community will
have spent part of their family time around the laptops, with the
children proudly explaining how they work.

2. Buenos Aires: David Cavallo, Rodrigo Mesquita, and Walter Bender
participated a series of five half-day workshops for a variety of
audiences. The attending groups included key people in government,
education, and software development, as well as events for the press
and general public. Alejandro Piscitelli and Laura Serra of educ.ar
contributed greatly to the discussions and development of ideas.
Valter Cegal and Rebecca Gonzales of AMD also participated.

3. New York: Sj Klein met with representatives from UNICEF, which is
developing projects for  UNIWiki, an effort to coordinate shared free
knowledge produced internally and by others (e.g., Voices of Youth).
They are especially interested in focusing on projects in developing
nations, with attention to multilingualism, mentoring, and
cross-cultural communication.

4. Washington: The Library of Congress World Digital Library is asking
their network of librarians and curators to join the OLPC curation
efforts, in the subjects and languages that most interest them.

5. Mesh activities: Dan Williams and the Collabora team continue to
work on the Presence Service, a key to developing mesh-enabled
activities. They are making good progress, building out the APIs and
testing the libraries under our framework.

6. Startup screen: Dan also found time to put together a new startup
screen for the laptop that takes a child's picture.

7. Bug hunting: Marcelo Tosatti investigated and located the source of
the iperf-corruption problem we were seeing on some of the laptops
under heavy load. It turned out to be the result of a fix in the
networking driver. Marcelo has also been investigating and working on
implementations to tell activities on the machine when they are
running out of memory and give them a chance to release caches or shut
down.

8. UI: Marco Gritti and Tomeu Vizoso have been making progress on Sugar. The
current builds have a large number of fixes and changes over what
shipped with the Build-239 machines; people will be pleasantly
surprised. Tomeu has been working largely on underlying widgets and
infrastructure and Marco has been busy working on higher-level
constructs, including working with Eben Eliason to firm up design
decisions. The new Sugar has a better default font, moves a lot of the
networking into the home page, includes a Journal demo, and rollover
information for the activities that should greatly help with usability
and first impressions.

9. Music Activities: The TamTam team has been hard at work improving
the music program as well. The new version exposes the track editor
and is much more interesting than previous versions.

10. Trial-1: We have been working toward a new stable build that will
form the basis of our first tests in the field. Andres Salomon
branched a stable tree (http://dev.laptop.org/~dilinger/stable/).

11. School server: The software architecture of the school servers is
starting to come together, through discussions this week around the
networking services
provided and possible scaling mechanisms. For Trial-1, the networking
will remain IPv4, with the school server providing DHCP, DNS, HTTP
cache, and NAT functionality. Hardware for school server development
has arrived in Cambridge, with plans to have a limited prototype up
and running over the next week.

12. Marc Fiuczynky of PlanetLab (Princeton University) visited to
discuss lessons learned by PlanetLab. Herbert Poeztl, who developed
the virtual-server mechanism (Vserver) used by PlanetLab, has been
working at OLPC to help integrate it with our software. (Vserver is a
Linux technology for "containerization" of environments that will be
very useful in the future for both management and increased security.)

13. Suspend/Resume and Power Management: Mitch Bradley has been
working on bringing up resume in OpenFirmware. He has dismantled and
instrumented a B2 with the following results:
* the long delay from power/wakeup to CPU on is down to 12mS (instead
of 500mS—probably a CAFE FPGA turn-on delay);
* we no longer sees random hard-hangs;
* a problem with the DCON wiring with respect to wake-up was found; and
* resume-from-RAM was having problems by is now seemingly reliable.

The current time from power-reapplied to completion of the wakeup
procedure is 27.6mS. Mitch knows an easy way to knock off another
4.5mS , to bring the core wakeup time down about 23mS. It is
conceivable that he might manage to shave off a few more milliseconds,
but probably not much. It is very tightly coded as is, and the "long
poles" are hardware delays like PLL startup and ROM access time for
early instructions. This time does not include video subsystem restart
time, so additional time will be needed for that.

14. Performance: Chris Ball and Dave Woodhouse worked on a
booting-performance problem. The JFFS2 kernel-thread-speed problem is
resolved. Booting a current build on a B2 machine takes 2 minutes 2
seconds, but drops to 1 minute 24 seconds with the work around for the
USB branch-prediction problem (reported last week), and now drops to 1
minute 6 seconds with a fix to the scheduling for the JFFS2 kernel
thread. This scheduling fix should go into the build soon. Chris Ball
also added graphing of Python performance over time to the tinderbox.

15. Network: Michail Bletsas setup a 14-node mesh testbed at OLPC and
spent last week with Cozybit and Marvell debugging the mesh firmware
and the wireless driver. As of Friday night all of the major problems
have been addressed to the point that the mesh functionality is now
usable:
* in-mesh multi-hop multicast support;
* link-loss detection and route tear-down with RERR messages (This
improves route-restoration time);
* mesh transmission rate is done a the highest-available rate for each
hop (The rate for each hop is determined when the route is
discovered);
* deferred route discovery (Route discovery is now done by a lower
priority task, which reduces the variance of transmission time);
* WDS problem workaround (Wireless interface will accept WDS replies
from WDS-enabled access points. This will only work with APs that have
a different MAC OUI than the XO's [Ticket #901]).
* fixing flow control for the mesh interface on the libertas driver
alleviated the problems with high-data-rate TCP flow corruption
(Ticket #915).

16. IPv6: In preparation for integration of laptops and servers, OLPC
has started working on our IPv6 implementation. Chris helped Dave
Woodhouse with setting up "tubes," the machine running our IPv6
testbed. All XOs and other systems able to support IPv6 in the
Cambridge office now get public IPv6 addresses by default, thanks to
Dave Woodhouse and our new intern systems administartor Daniel Jared
Dominguez.

17. Power management: The power-rail measurement system arrived this
week. We now have in house equipment that can measure the current on
all power rails of a B2 board to an accuracy of about 1mA. Richard
Smith will start taking measurements on each rail and testing the
suspend-resume code. In addition to current measurement this equipment
has can control several relay contacts. These can be connected to
power switches on the laptop. All of the measurements and outputs can
be controlled remotely by Ethernet, serial, or USB. We therefore have
the ability to build a new tinderbox that can automatically test nand
image builds and firmware upgrades, while taking power measurements on
every power rail during the entire process. Previously we could not
test firmware upgrades automatically because we did not have a good
power-cycle method/restart-after- flash method.

18. Kernel: Andres synced up our kernel with the 2.6.21-rc2 release.
Andres also fixed a bug in the interaction between the rpm spec file
and dynticks that was breaking system tap. With that fixed, tinderbox
should work with 2.6.21-rc2 properly and can properly benchmark jffs2
and the SD driver. There is a pending bug in the SD driver for which
he has prepared a patch but cannot test until he can actually
benchmark. Andres "thinks" the dynticks bug is fixed (Ticket #954),
and the jffs2 bug (which adds 30 seconds to the boot time, as
discovered by Chris Blizzard) is still pending. Zephaniah Hull and Jim
Gettys "fixed" the touch-pad bug. Jordan Crouse of AMD has been
developing driver patches for suspend and resume and is further
investigating the USB performance problem. Andres also did some
investigation of the USB performance problem; he and Jordan discussed
how the hardware would implement uncached memory and poked around the
kernel code for ways to easily do it.

19. From the community: Andrew Clunis reports that a reasonably
functional version of the Develop activity is now available in
sugar-jhbuild (http://wiki.laptop.org/go/Develop). It provides a very
basic "IDE"—a file TreeView and text editor, currently provided by
GtkSourceView.

20. Policy discussion: A new mailing list, aop at laptop.org, has been
set up to host community discussions about policy decisions:
everything from the OLPC security model to our position regarding
FOSS. You participation is welcome.

-walter

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