OLPC News 2007-12-01

Sat Dec 1 15:10:37 EST 2007

1. Montevideo: We've only just begun!! The first deployment machines
were handed out in Escuela No. 109 in Florida, a rural department of
Uruguay. The second batch was handed out in Escuela No. 24 in Villa
Cardal, which has been a pilot site since May of this year. In Cardal,
we gave children production XOs and collecting their old Beta-2 units.
The OLPC deployment in Uruguay is being run by Miguel Brechner as part
of Proyecto Ceibal (Ceibo is the national flower of Uruguay), a
presidential initiative to equip each child with a laptop. The Ceibal
offices are housed in a Montevideo complex called LATU, or Laboratorio
Tecnológico del Uruguay, which is a public/private sector cooperative
technical lab now responsible for much of Uruguay's technical
certification and quality control programs, as well as serving an
incubator role for various engineering and technical projects. The
OLPC team has put all of their blood, sweat and code into the project
over the past three years because of the unshaken belief that it is
the right thing to do. Now it is real. You can read more about our
first deployment on Ivan Krstić's blog (See
http://radian.org/notebook/first-deployment).

2. Changshu: Mass Production is now very stable. We are using our line at 100%
 capacity. Congratulations to Quanta for stabilizing
 production just three weeks after MP start.

3. G1G1: Every "Day 1" Give One Get One participant (those that
donated on November 12, the
 first day of our campaign) received email on Wednesday informing them to
 expect their "Get" laptops between December 14 and 24. Delivery windows for
 other G1G1 participants were calculated and posted: "Get" laptops
 ordered thus far will be arriving, at the latest, by mid-January 2008.
 Brightstar and OLPC have been working closely to determine when the
complicated logistics of laptop delivery can be promised.

4. AC adapters: There has been a request for AC adapters that are
rotated ninety degrees from the current configuration. In order to
rotate the orientation of the prongs, the
 width of the adapter must be extended (to satisfy the safety requirement).
 As a result, six reoriented AC adapters will not fit abreast in the
standard spacing
 of a six-plug power strip. Mary Lou Jepsen and Fuse are investigating
further; if we can not resolve the issue, we will not make AC adapters
with a
 rotated prong orientation.

5. Schedules:
 This was the week to release "Ship.2", a build to improve upon
network upgrades, wireless problems, and our ability to connect to
T-mobile services. Although the week had its ups and downs, it ended
with successes on all of these fronts and we have Release 649 as the
candidate, barring any last-minute problems in testing this weekend.
The Ship.2 Build connects successfully to many different access
points; we believe we have fixed the "lazyWDS" issue (which could have
potentially caused problems in multi-access-point environments with
other 802.11b/g laptops and XOs); and we are successfully connecting
to T-mobile services after setting an appropriate configuration for
the browser.
 The Roadmap has been updated (http://dev.laptop.org/roadmap); please
send Kim Quirk any new feature/product ideas so they can be scheduled
into upcoming releases.

The release candidate for the G1G1 program is Build 649 and will be
called Ship.2. Final testing is underway, specifically for network
issues that have been reported. The wireless team of Ricardo Carrano,
Michail Bletsas, Javiar Cardona, Ashish Shukla, David Woodhouse,
Marcelo
 Tosatti, Ronak Chokshi of OLPC, Marvell, RedHat, and Cozybit worked hard
 on these issues of both the Linux driver and the Marvell wireless
firmware and worked out a solution, which is now in final test, under
great pressure. Ship.2 has most of the user visible features of the
planned Update.1 release, but has Bitfrost security turned off and
lacking the advanced OHM-based power management with aggressive suspend
 and resume and ebook mode. Both of these major features are now
 operating in our Joyride test builds, but there were too many open
issues to enable Bitfrost or to ship OHM power management.

6. Testing:
 This weekend Michail Bletsas and Ricardo Carrano are testing 40+ XOs
with a number of access points and traditional 802.11b/g laptops to
"prove" the fix for the WDS problems that have been plaguing us for a
while. Thanks to Marvell for their debugging and test builds to find a
good work around; and special thanks to the volunteers in the office
who activated and upgraded all the laptops in preparation for this
testing: Andriani Ferti, Danny Clark, Adam Holt, Alex Latham, and Eben
Eliason.

7. Support:
 Adam Holt joined OLPC this week as our Support Engineer focused on
tracking, debugging and follow up of customer problems from the field.
He is an MIT graduate who has worked in software  development,
support, and systems integration. Most recently he comes from Jenzabar
in Boston. He was a great help in his first week as we were testing
and configuring laptops for Ship.2 Release.

8. Wireless: Michail Bletsas reports great progress was made this week
in circumventing a problem that we had with access points that were
misinterpreting our mesh frames as wireless backbone frames, resulting
in the creation of erratic ad-hoc wireless network topologies. These
were access points (APs) utilizing wireless adapters from Broadcom as
the ubiquitous Linksys WRT54. We have alerted Broadcom to the issue
(if the standard doesn't change, 802.11s devices will  be rendering
these APs useless unless they address that behavior) and we have
worked around it by  changing our broadcast and multicast mesh frames
to use 3-address frames instead of the 4-address WDS frames that the
Broadcom APs are already using. Thanks to Javier Cardona, a dedicated
team at Marvell, and Ricardo Carrano for implementing and testing this
in time for our Ship.2 Release. Javier will be submitting this change
for consideration in the next standards committee meeting. This change
requires no driver/software changes on the XOs and is implemented
completely in firmware. The important point to note is that all XOs in
a mesh have to be running the new firmware (5.110.20.p42) for them to
be able to collaborate via the mesh.

Michail et al. had great fun at OLPC in the later half of this week.
Somebody was using a wireless device that was acting as a jammer for
the center part of the 2.4Ghz band. Despite all the troubles that it
caused us by preventing us from doing any serious testing of the
laptops, we should be thanking him/her since the extremely difficult
environment helped expose a serious issue with wireless operation in
congested environments which has probably bitten us in the past and
which we had never had the opportunity to properly diagnose and debug.
The problem has been pinpointed to the scan routines in the firmware
and is being currently addressed. Testing with a jammer present is now
a requirement (not your usual WiFi test ;-).

9. Sugar: Simon Schampijer fixed memorize again to release the sound device and
 put the drumgit game back in the distribution which is the only demo
 game we have which includes sound (in current joyride). He added an
 "About this XO" entry to the menu you get when hovering over the XO icon
 in the home view: this brings up a window
 with information about the current build, firmware and serial number of

this XO. With Mako Hill, SJ Klein and Marco Pesenti Gritti, Simon worked on the
 library fixes for Ship.2 which included a fix for Sugar (Ticket #2856).
 Marco and Simon added as well support for the OLPC Root CA into the
 browser. You can test this by pointing your browser to
 https://activation.laptop.org in the current Ship.2 build.

Marco reviewed several patches with small UI improvements
 (to the  interaction with palettes in particular) and bug fixes for
 Update.1  and  packaged them in joyride for testing. Marco also tracked
 down two different issues with the datastore which was causing
 activities to not start. Fixes for both of them landed into Joyride.
 Finally, he changed the default Jabber server for Ship.2.

Reinier Herres worked working with Marco to build a new evince version
 for the Read activity, which will probably be available early next
week. He also
 released a new Calculate with some bug-fixes.

Morgan Collett worked on Chat to fix scrolling issues—he made it not
automatically scroll on new messages if you scrolled up to read the
log

and worked on copying URLs to the clipboard (#5080); he found some
issues with clipboard handling in Sugar. We will land the Chat and
Sugar
 changes when the Ship.2 dust settles.
 He also worked on the Presence Service: preforming lots of testing on
Ship.2 and
 Joyride builds, testing recent fixes.

Memorize: Muriel Godoi reports that Memorize was launching under
Rainbow, but wasn't saving games in datastore; after some chats with
Michael Stone, they realize that Memorize is creating new sub-folders
inside their instance folder using tempfile library under permission
700 denying the access when trying to write into it. Changing the
folder permission fixed that problem. Muriel also fixed the mime-type
icon file location and added ogg support.

Muriel also reports that Food Force had some UI improvements, such as
a message bar where contextualized educational messages to the player
will be displayed. The code will be posted to his public_git next
week.

10. Etoys: For Ship.2 release, Etoys team packaged a new version that has the
 progress made since Ship.1. Bert Freudenberg has been carefully
 working to adapt changes in Sugar while keeping the compatibility with
 old Sugar so that the latest work can be back ported.  In this week,
 Yoshiki Ohshima and Bert improved the sharing experience and Takashi
 Yamamiya fixed a bug in script editing interface to make the version
 for Ship.2 be very comfortable. In the meantime, Scott Wallace took a
 pass on making an IDE for a traditional text file based programming
 system on Squeak, which may be useful for making an IDE for XO.

11. Localization: Xavi Alvarez and Sayamindu Dasgupta have finally put
the GIT integration
 of the Translation Infrastructure (Pootle) in place in a fully working
 form. Translators can directly commit to GIT now via the Pootle web
 interface.
 We also have a new project in Pootle, called Update 1 (Core), which
 tracks the update-1 branch for a few core modules of the Sugar
 environment. From an average translation coverage of 23% at the
 beginning of the week, we have 33% average coverage now. Languages with
 more than 80% of Update 1 (Core) translated are
:

French
(100%)

Chinese (Taiwan)
(100%)

German
(96%)

Arabic
(90%)

Polish
(85%)

Spanish
(84%)

Quite a few of the other languages are catching up quickly.
 A number of Indic language translation teams started work this week:
 Malayalam
,
Marathi
, Punjabi
, and Tamil
. A Japanese project was also started this week. Waqas Toor and Salman
Minhasreport that more than 85% of Urdu localization is completed.

Xavi and Sayamindu also managed to fix a few build breakages
introduced by some issues
 in Pootle's PO generation code. Things should be running smoothly now
 (Sayamindu has also put in a logging framework in place that monitors
 the interaction between Pootle and GIT so that we can investigate
 potential problems easily later)
.

12. OLPC Pakistan: Waqas and Salman have initiated developing two
e-Books and are busy working on an English to Urdu Glossary Project
for OLPC.
A community list for support in Pakistan is up and running (Please see
http://groups.google.com/group/olpc-pakistan and visit the
#olpc-pakistan IRC channel on irc.freenode.net).

13. System programming: Chris Ball worked on OHM early in the week,
Ship.2 testing towards the
 end.
 Bernie Innocenti has been squashing the last few input, localization and
 configuration bugs. Mostly in olpc-utils and xkeyboard-config and
 finished moving the rest of the Xorg packages to Fedora's
 repository and now has no packages built locally. He also helped SJ
Klein apply blanket changes all over the dictionary to fix
links and encoding errors. And he did some transformations for the start pages.
 Bernie also finished building a geode-optimized package of glibc 2.7
 that looks very promising, but still needs testing and started packaging
 X 1.4.99 from Fedora development, which seems interesting for the
 various EXA improvements. There is some PCI rework needed in amd_drv
 before we can start doing some benchmarking. He is hoping to offload
 the rest of this work to Stefano "aleph" Fedrigo, who in the past
 has been doing some neat X performance analysis for us.

Dennis Gilmore spent some time helping John Watlington and Michael Stone
 with some issues with the school server live image. He submitted a
 patch to upstream yum adding geode support and worked on a patch for rpm
 adding geode support should be done today or tomorrow. Dennis worked on
 some code to enable subscriptions to koji.  This will let us import new
 builds into our own koji when we get it.

Andres Salomon worked on the xorg evdev driver (*cry*), setting up

general geode testing infrastructure, some participation in the
 libertas nonsense (*cry*^2), and some dcon debugging.

14. Presence service: Guillaume Desmottes tested, modified and improve
Dafydd Harries' patches for #4965
; he did more ejbbard/Erlang investigations
; reviewed Sjoerd Simons's Salut fixes
; started to evaluate Openfire as alternative Jabber server
; debugged and fixed a PEP problem in Gabble breaking sharing with
Openfire (#5223)
; improved friends roster synchronization (#4965)
; fixed a Gabble bug when requestion lists channels (#5164)
; debugged and fixed a muc properties bug in Gabble breaking sharing
with Openfire (#5224)
; and modified telepathy-python tubes examples to work with latest Salut
.

Dafydd Harries worked on a more scalable XMPP protocol for activity management
. Sjoerd Simons tested Salut for Ship.2 and fixed all outstanding Ship.2
 issues it had.

15. Multiple-battery charger
: Tooling has been ordered; first prototypes will be available at the
 end of
 December or early January. The PCB for the production system is done.
 The question is how many to build for our first run. The current
 plan is
 to build twenty-five, but this may be increased if there is more
 interest from
 trials.

Lilian Walter is progressing nicely on the battery charger firmware.  She
 has been
 working on implementing a LiFePO battery charging algorithm this week.
 She has charged a battery, and so far only damaged a single battery.

16. Firmware:
 Mitch Bradley and Richard Smith have released Q2D05 with
miscellaneous bug fixes for
 Ship.2. This contains EC firmware with the software workaround for
 most of
 the suspend/resume problems and the now controversial (?)
 blink-power-LED-on-suspend. Since suspend/resume is disabled for
 Ship.2,
 these changes will not be apparent to the user.

Wireless support from within OFW is still not totally robust.
 Mitch is unsure whether it is the WLAN firmware, the 1CC RF jamming,
 or an OFW driver problem. (Over the past week we have been unable
 to use WiFi channel 6 at 1CC due to constant, high level non-WiFi signal
 from an unknown source, the aforementioned jammer.)

17. Touchpad: Richard Smith is now pretty convinced that our touchpad
problems are caused by
 the auto-calibration feature of the touchpad. The two
problems—undersensitivity and jitteriness—are opposite results of a
bad recalibrate.
 By forcing a calibrate to happen with the touchpad in various
 conditions he
 can recreate our touchpad problems.

"Go to a corner" and stay is caused by under-sensitivity. Duplicating

this is fairly easy. Do the recalibrate (the "four-finger salute")
 with as much
 of your thumb on the touchpad as you can, pressing quite hard.

"Jumpiness" is caused by over sensitivity.  Duplicating this is bit
 harder. The best Richard has found is placing a large chunk of thick
 rubber on the touchpad while the unit is on battery power and then
 recalibrate.

A recalibration while the touchpad is in use causes under-sensitivity but
 we're not sure how over-sensitivity happens the field. Nor do we yet
 understand why some laptops are so much worse than others. We
 are arranging a conference call with ALPS to discuss the issue.
 The only "fix" Richard proposes is to disable auto-calibration. It seems
that auto-calibration can't ever be made safe without some method of
 insuring that the touchpad is free and clear for the recalibration.
 It is
 unclear whether we still need to auto-calibrate, or if this was only
 needed
 for the problems seen with the B2 build.

18. School Server
: School server development has restarted.  It was discovered that
 previously
 distributed installers will no longer work due to a reliance on a now
 missing
 Fedora server.  The build system also needed repair but seems to be
 behaving
 again. A new build is being tested and will be released over the
 weekend.
 This build will have no new features but will contain the latest
 wireless mesh
 firmware and drivers.

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