System Software Teleconference minutes.

[Mitch Bradley]

System SW Teleconference
2007-02-28

Jim, Mitch, Nat (Flipo), Wad, Jordan, Richard, Blizzard, Ronak, Andres, 
Mary Lou

New Action Items:

AI: Blizzard to bug Marcelo to try latest TCP
AI: Andres - look into what it would take to uncache USB descriptors.
AI: MLJ - get new schematics and .brd file from Quanta.

Old Action Items still open:

AI: Jordan and ajax to investigate xrandr segfault

AI: dcbw and marcopg to investigate mesh/avahi not working.
      done, but we're seeing crashing reports



Touchpad choice for B3: Decision - go with Alps

Suspend/resume - DCON_LOAD, PCI_RST#, speed of startup

Several hundred laptops are about to go into the hands of children within
the next few weeks.  There are several blockers against B3 that maybe
should be accelerated.  We must consider the user experience.

We need an ID for children, for backups, IM, etc.  Can we use the "nick"?

Storage view: Wad's proposal (with help from SJ): use a directory model
for now because we won't have the journal in a week.  Each child has a
home directory that is rsync'ed up to the server.   For downward syncing,
every file is available through a URL.  For first path, there is no
way to delete a file on the server.

Need some way to copy a document (since we don't have a file manager) - 
perhaps
a "save as" function in AbiWord.

Further discussion on user experience deferred until tomorrow - separate 
meeting.


USB Performance Problem: Still waiting on the AMD guru, who is studying
the issue in detail.  Preliminary choices (a) leave workaround on and
try to limit the number of snoops, or (b) turn off branch prediction.
(b) is more expensive than (a).  One possibility for (a) is to do descriptor
accesses from uncached memory.

Iperf corruption problem:
  a) The data on the USB bus looked okay
  b) There were always 29 packet exchanges before the hang
  c) It might be a TCP bug - try Dave Miller's latest TCP
  d) netstat reports error counters increasing on every run

Trickle-charge still doesn't work.

Touchpad workaround:  Consensus is to apply the workaround in the kernel 
driver.

239 release notes done.

Tinderbox/power measurement: the vendor has not yet shipped the instruments
despite it having been listed as in-stock.

Kernel boot time is down from 2 minutes to 1 minute, due to work by 
Chris Ball
and Dave Woodhouse.

RandR - Jordan is still working on it.

Avahi - problem is proving difficult to diagnose.  Avahi is starting 
fine, then
crashing, leaving nothing in the logs.