[OLPC-devel] Minutes of OLPC software teleconference, 6/13/06

[Jim Gettys]

Minutes of teleconference, 6/13/06, started about 10:35EDT after Jim's
fumbling with the OLPC's telephones.

Attending: Jim Gettys, Dan Williams, Marco Gritti, David Woodhouse,
David Wahl, Jordan Crouse, Steven Goodrich, Marcelo Tosatti.

Discussion of day of week of this call: it was noted that we should
really do this on the mailing list since by definition, those who could
not make it were not present.  Again, please let Jim Gettys know if
Tuesday can't work for you (7:30PDT/8:30MDT/9:30CDT/10:30am EDT/14:30
GMT).  The time of day is chosen to get all of North and South America,
and as far east into Eurasia and Africa as possible.

Maybe starting next week we'll do an (evening U.S. time/morning far east
time) call (alternating weeks) to pick up the far east/Australia).
Please send me mail if you would attend such a call.

Introductions all around, so we can start learning people's voices.

General status: 485 boards arrived at Brightstar; 50 arrived in
Cambridge this morning.  Antennae are catching up to the boards; they
left Quanta at the end of their work day today in Taiwan. Well over 100
boards will likely go out within the week to the FOSS community, and
another significant number will be going to launch countries in the very
short term.

We started with a discussion of what boards are needed by whom for
testing: we need to get ATest boards to Dave Woodhouse, 2 to AMD in
Colorado, and one to both Ron Minnich and Marcelo Tosatti to replace the
pre-A boards they have now, so that BIOS installation testing can be
done. If the pre-A test boards are not being used, please return them to
OLPC.

Action item: a pointer (link) to both "believed to be working and
somewhat stable" full Fedora Rawhide and to the OLPC distribution needs
to go into the directories, so that those who need to demo have a known
set of bits to grab, rather than daily builds which will inevitably be
more "interesting" ;-).

Action item: Aaron Burton, an intern here at OLPC, is doing fresh
installations from scratch of both full Fedora and the OLPC distribution
with actual ATest from Quanta kits, and will be working on
debugging/clarifying the hardware and software installation directions.
We hope this will reduce questions as boards arrive in people's hands
around the world. Aaron has started on this task. My previous experience
is that such instructions is more difficult than it naively sounds,
particularly due to language issues.

Action item: jg to send Dave Wahl the alpha blending performance
question again (done).  Dave, let me know if you don't have it now.

Ivan Krstic reports that laptop.org will be able to support source
repositories (in particular git trees) in the next day or so, and will
announce to the devel list.  There is 100megabits to a switch that
connects to multiple gigabit long-haul ISP's in the MIT co-lo center
where laptop.org is housed.  That 100mbits could be increased if
warranted.

Ivan has also installed Trac as a bug tracking system, and will be
announcing it generally too.

There was discussions of code management systems: we noted that both the
kernel and now increasing parts of the X Window System are using git,
and therefore we know git will be in common use on significant parts of
the work we have to do. Dan noted he hates mercurial, and will evaluate
with the rest of the Sugar team if they will use something git or
something else

Action items for those who want accounts on laptop.org for hosting
projects (e.g. for hosting git repositories): send your ssh2 key to
Ivan.

We had a discussion of the pros/cons of running multiple partitions on
flash.  This is not a resolved topic and will be revisited. There was a
discussion of the new flash layer being made available by IBM called
UBI, and the relative merits of its wear leveling across partitions, and
how it may interact with compression in JFFS2, and how paging works.
Dave believes UBI will be stable enough for us to use. Also noted by
David Woodhouse is that garbage collection in jffs2 requires 5 reserved
flash blocks for each file system (the blocks are 128K on our flash).

Jim noted that on swap-less systems we've observed useful paging
behavior in the past, where the paging system figures out which pages of
code are in actual use in the running set of processes; the out of
memory point is when most of your RAM is in use for data, and you
finally don't have enough code pages left to get work done.  Paging
performance therefore matters more than one naively might think (as you
page back in code pages you need), and this is why the flash performance
problem previously noted affects more than just application startup.

AMD continues to investigate options around the flash performance
problem; no guarantees of success.  Dave reported mail that Marvell will
also be looking at the flash controller on that part and be getting back
to us.

Dave reports that Jffs2 is believed reasonably stable at this point and
not a barrier to installing Linux directly on the flash of the boards.
He continues performance work.

AMD reported (Jordan?) that fbdev is up and running on the GX2 with a
few patches from the version that went into Linus' tree recently. This
eliminates the VESA code dependency out of LinuxBIOS, and we should also
start using it in our full up Linux systems rather than the VESA driver.
Jim noted that we have to deal with support for the DCON Xilinx version
by sometime in July, so that we have code for testing the screens.

Ron Minnich is at a conference until Thursday; Ray Tseng has provided
sufficient information about how to reflash the serial flash so we can
do so from Linux; this may have already been integrated into the
LinuxBIOS flashing tool.  So we're pretty close to being able to cut
over to LinuxBIOS, and do so from Linux, rather than going the Freedos +
Quanta flash utility route.  This will clearly need careful testing.

There was some discussion of schedules to bring people more up to date.

Action item: Jim to make sure Mark saw the fact that there is a known to
work with Linux flash controller logic design available.

Adjourned 11:32AM EDT.

VOIP: Jim will be ordering a small server; we have a volunteer on the
devel mailing list to set up Asterisk so that we could support standards
based VOIP call-in, requested by some people on the list.  It will be a
bit before this machine gets ordered, is installed and up, but the
intent is there.

My apologies if I missed something. Taking notes and running meetings
isn't something I do very well. And I don't know everyone's voices yet,
so my apologies for any lacking or mistaken credits as to who said what.
                         Regards,
                                 - Jim


-- 
Jim Gettys
One Laptop Per Child