[Community-news] OLPC News 2007-07-21

[Walter Bender]

1. Testing: We ran battery-life tests on 75 B4 laptops: results will
be reported on the OLPC wiki this week. Next week we will use 100 B4s
for mesh tests. Dafydd Harries and Simon McVittie of Collabora will be
helping us with mesh-scaling and connectivity tests. The current
software builds are progressing well. Most of the Trial-2 features are
in Builds 528+; many activities (Read, Write, Chat, Connect4, Record)
now have collaboration features enabled; the Journal is integrated,
and small link-local mesh networks are working. Thanks to everyone who
has been documenting, reviewing, and prioritizing bugs in Trac
(http://dev.laptop.org); most have been triaged and will be fixed in
the next release (Trial-3).

2. Trial-3: We've started discussions on the software release that
will line up with the mass-production hardware build; it needs to be
ready in early to mid September. This will be largely based on Trial-2
and will include many bug fixes and only small feature changes.

3. Sugar: Marco Gritti reports "really good progress!" He has chased
down numerous bugs—in the clipboard, regarding themes, palette
behavior, and the user interface (UI). He has completed much of work
of integrating open hardware management (OHM) with the UI. Other
details include reworking the brightness and volume key handling for
the B3/B4 keyboards; tracking down the "white row" X11 bug; and
getting the Totem media plug-in to work in the browser. Tomeu Vizoso
also made fixes to the Clipboard; he integrated recent changes in the
datastore into Sugar; implemented lazy (paged) scrolling in the
Journal by caching more entries than can be shown on-screen and
fetching them opportunistically; and fixed some Sugar and Activity
file leaks in /tmp.

J.M. Maurer added a "zoom" signal to the libabiword widget, and hook
it up in Write; added full justification to the Write activity;
implement "buddy left" in Write and the AbiCollab Sugar back end; and
he fixed a bug that was causing Write to crash when opening a second
instance.

Arjun Sarwal worked further on optimizing the response time of the
Measure activity (which turns the laptop into an oscilloscope). He has
made improvements to the UI by adding a grid onto the background and
he has chosen color scheme that improves contrast. New features
include being able to start/stop the display and taking a snapshot of
the currently displayed waveform. Data logging has also added. A
version of the activity is available at the git repository under the
category projects/measure (See
http://wiki.laptop.org/go/Measure_/_Oscilloscope).

4. GNOME Mobile: John Palmieri was at GUADEC (GNOME Users' And
Developers' European Conference) this week, representing OLPC at the
GNOME mobile and embedded (GMAE) meeting. The GNOME Mobile platform is
a subset of the GNOME platform; it represents components that are
currently shipping in the XO, the OpenMoko, the Nokia 800, and other
devices.

5. Salut/Gabble: John was joined at GUADEC by Sjoerd Simons, Guillaume
Desmottes, Morgan Collett, and Simon McVittie. Between sessions they
they worked on minor bug-fixes and updates in Salut (link-local XMPP
connection manager). Sjoerd figured out why Salut sometimes "looses"
contacts on bad networks. Together, they finished the new tubes
specification. Simon also worked on fixing avatar (buddy picture) bugs
in Gabble (the Jabber/XMPP connection manager we use for chat and
video calls).

6. Builds: Dan Winship cranked out multiple builds per day this week.
(We are now up to Build 528.) He made a fix to Pilgrim to accommodate
activities with spaces in their names; and he wrote some tools to help
determine (and document) changes between builds.

7. Mesh activities: Dan Williams debugged and fix various activity
sharing and Mesh View bugs. He fixed Python 2.5 issue that broke
sharing in Record and Read; and he wrote a Jukebox activity, a
gstreamer-based media player with Journal integration. He also did
some Libertas maintenance, patch review; made wireless scans "less
hacky"; and fixed a wireless scanning issue in NetworkManager that
caused access points to drop off the mesh view over time.

8. Power measurement: This week saw a flurry of activity trying to
reconcile the power draw of each XO subsystem with what we expected.
John Watlington, Richard Smith, Joel Stanley and Jim Gettys working on
detailed analyses of power consumption. The goal of the measurements
has been verification we have hardware in place enabling us to turn
off any component at will and that an "off" component really is
powered off. This will mean that as software development progresses,
we will be able to keep any unused circuitry off or at low power when
it is unused. We now have 29 different measurement points on the XO,
giving us fine- grained data on every power rail, covering all major
subsystems.

If all subsystems are on and running at maximum power use, the XO can
consume a maximum of approximately six watts (ignoring external USB
devices). To verify this, we have put the XO in various states of
display, wireless, USB and CPU usage. Normal usage when the system is
idle, or in ebook suspend mode, or mesh only mode is very much less
than this maximum, of course.

The wattage drawn from each power rail was carefully reviewed and
audited to make sure we knew where each watt was going. We were not
able to account for every watt of the system, but we are really
close—close enough that we feel that all the components are
functioning as specified. But "as specified" in a few cases it still a
bit high. WLAN and DCON (display controller) are a bit problematic. We
have promises from the WLAN team that there are lots of knobs to tweak
to reduce power consumption and that they will now start tuning those.

The word from HiMax on DCON though is that what they measure for
sleep-mode power draw is the same as what we measured for sleep mode,
approximately 150mW. This is not acceptable. When we go into sleep
mode or low-power WLAN-only mode we will need to drop the power to the
DCON. This means that the kernel will soon have to sprout knowledge of
how to init the DCON from power up.  Previously the kernel could rely
on OpenFirmware (OFW) to have initialized the DCON. (We have verified
under OFW that we can drop power to the DCON and then bring it back to
life.

Most other Linux power-management controls has been verified as now
working correctly, e.g., audio, camera, etc. The surprise discovery is
that our processor's power supply is not as efficient as it could be.
On the more positive side, we shaved 30mW from the power required when
the laptop is not turned off and another 35mW whenever the screen is
turned on.

9. Performance: Chris Ball changed some of our power behavior; we now
turn off the screen while suspended via the power button or lid close,
saving power. Chris noticed that resume from suspend becomes unstable
if we are stressing the wireless radio at the time of resume. This bug
has been passed on to Marcelo and Cozybit for investigation. Chris
also reported a problem with game keys and power management events
being seen as power-button presses to AMD; this is putting us into
suspend when we don't expect to be going there.

Chris is investigating our memory use, which has crept up recently—we
are running more Python processes than we use to without having them
share libraries, and running a greedy DHCP server (taking 12–20MB!)

Chris and Richard wrote a Python script to calculate the watt-hour
value reached by the battery tests we are doing. This lets us make
accurate calculations on the duration we'll get from measured power
draws.

10. Firmware: Quanta released the 2nd test for what will turn out to
be the PQ2C20 firmware. Richard Smith reviewed all the changes. The
highlight feature of this is the removal of code supporting Atest
(dropping ATest code removed 672 lines from the code base.)

Andres Salomon worked on a number of embedded controller (EC) fixes,
including helping Richard debug EC issues by providing proper kernel
support. Mitch Bradley got Forth running on the 8051 processor as a
possible base for the open EC code and worked on a firmware release
with activation support.

Lilian Walter got IPv4 and IPv6 dual stack ping and finger working.
Lilian is researching on how to implement Teredo tunneling (Teredo
tunneling is a protocol designed to grant IPv6 connectivity to nodes
that are located behind IPv6-unaware NAT devices); this involves Ipv6
features such as router solicitation/advertisement and DNS AAAA
packets.

James Cameron has started work on an open-source firmware
implementation for the EC.

11. Kernel: Marcelo Tosatti did more parallel suspend/resume testing;
found an unnecessary call to mdelay() in the CAFE driver; and
investigated more Libertas suspend/resume problems and failures under
high traffic. Andres Salomon worked on another Libertas merge, dealt
with getting code upstream, got the wakeup-from-keyboard work
properly: lots of fixed bugs and hopefully fewer introduced ones..

Remaining for completion of Linux kernel power management is powering
the DCON down when not in use (saving 150mw in suspend, as noted
above), and debugging of the DCON patch to enable the use of the DCON
use when the window system is idle (saving more than 0.25 watts while
running). We also need to explore how much power is saved by turning
off the Dotclock PLL in the processor when it is not needed.  These
will improve further our battery life.

12. School server: Daniel Wyatt made the first installing,
pre-configured school-server live image (See xs-dev.laptop.org/xs and
the scripts in the git tree
http://dev.laptop.org/git.do?p=projects/livecd-data;a=summary ;  the
configuration files are at
http://dev.laptop.org/git.do?p=projects/xs-config;a=summary and the XS
"call home" scripts are packaged at
http://dev.laptop.org/git.do?p=projects/xs-callhome;a=summary).

13. Content bundles: With input from the community, Lauren Klein and
Kathy Paur helped flesh out the content-bundle specification. Now we
need to implement a mime-type hook for .xol (the extension name being
considered) and a script to unpack it and regenerate a navigation
templates in /home/olpc/Library. We have had some community members
already uploading .xol files to the [[Library Grid]] on the OLPC wiki.

14. Content Jams: CC-Taiwan is one of the largest Creative Commons
chapters; and they are very excited about running a content jam around
Wikimania. Another Jam is being planned around linuxconf.au.

-walter

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