[Tinyos Core WG] Meeting: September 20

[Kevin Klues]

Here are the meeting notes (on wiki as well as here)

http://tinyos.stanford.edu:8000/TinyOS_2.x_WG/9.20.2006

------------------------
= TinyOS 2.x WG/09.20.2006 Meeting Notes =
------------------------

= Agenda =
------------
 * Tutorials
 * TEPs 101 + 109
 * TEP 117
 * TEP 103

= Participants =
----------------
 * Kevin Klues, Wustl
 * Phil Levis, Stanford
 * David Gay, Ben Greenstein, Intel
 * Vlado Handziski, TU Berlin
 * Phil B, Jonathan, Gill, Arch Rock
 * Prabal, Berkeley

= Action Items =
-----------------
 * Everyone needs to start wroking on the tutorials they signed up for
 * Phil B. will look at meeting notes from last time, Henri's email, and
talk with Jonathan to work on revising TEP 117
 * Vlado will look at gcc compiled code to determine inefficiency of
returning struct values for ADC configuration
 * Gil wil start chagning implementation of sensor drivers to work top of
new HIL interfaces once Vlado is finished with his evaluation
 * Jonathan will get stm25p flash implementation up to date with new Storage
interfaces on a branch that David will create

= Next Meeting Agenda =
------------------------


= Discussion Notes =
--------------------

== Status ==
Phil: Tutorials have been signed up for, but some remain
Phil: Easy to debug and don't take as long to write
Phil: Would engourage peole to sign up for them
Prabal: Which still need to be written
Phil: Six tutorials remain that need to done, 3 not yet volunteered for
Phil: The three are concurrency, SDK, and storage not yet volunteered for.
      * Prabal - Storage
      * Kevin - Power management
      * Henri - Platforms and Network Protocols
      * David - SDK
      * Phil - Concurrency
Phil: Usually takes about 4 hours or so as an estimate

In regard to TEPs, we need to provide summary of changes, and not just
send along new version.  We could alternatively do a cvs diff if you provide

the version in cvs it should be diffed against

== TEP 117 ==
Updated to reflect newest I2C interface and associated flags.
Changed result_t to error_t

Vlado: Wanted USART discussion included as well
Phil B: Thats not in there
Phil: Talk to Jonathan, and lets defer it till next week then
David: Henri's email needs looking at (He says TEP 109, but he really means
117)

== TEP 101 ==
Was out for review and TEP now addresses these comments.
Four major changes
 * Removal of HAL2 and intro of HIL for ADC
   * Has platform independent data interface
   * Pass suitable configuration struct that is platform specific
   * Read interface remains uninterpreted and conversion functions all on
next level
   * Plan to submit additional TEP for how to convert into engineering units
when connecting to sensors
 * Usage of DMAs to implement fast version of reaadstream
   * Required changes to semantics of read operation
   * Until now, read took confiiguration as parameter
   * Once device granted to read, it would configuree it, do conversion and
give back result
   * Now dma needs configuration as well, so takes too much time
   * Solution: Provide configuration command to user rather than including
it in read as a parameter
   * User needs to call configure command before before using ADC -> used to
limit delay on configuration
   * Before was paramter of read, now have separate command so you can
better control timing
 * Removed long discussions of roundrobin arbitration
   * Points now to TEP 108

Phil: appendix C should refer to HIL, not HAL, just one letter change
David: inefficiency of returning a struct, does gcc produce code storing
everything on stack or in registers?
Vlado: wanted to ask about this actually, could potentially return a
pointer...
David: Have you looked at the code thats produced?
Vlado: No, we should definitely check though
David: Enough inlining could make this all efficient anyway, without needing
to return pointer
Phil: Just look at the code and see what is actually happening, it may work
great, so lets not speculate yet on how to "fix" it

Phil: Do chagnes sound reasonable given TEP 109
Gil: Haven't actually implemented sensors on top of new interfaces, but
seems reasonable and will most likely work
Vlado: Its not really much of a change in implementation, just a change in
abstraction organization

== TEP 109 ==
Clarification of MAY, SHOULD, MUST
Make comments about providing unit conversion interface is now SHOULD not
MUST
Virtualization at sensor HIL is now capital SHOULD, just suggestion (but
strong suggestion)
Naming convention is now a MUST
MUST for breaking up sensor specific drives s
Examples still don't reflect chages to TEP 101
Just need to make these small changes and check in new version

Phil: why would one never not to virtualize a sensor?
Gil: Look in appendix C, split phase, so no need to virtualize
Phil: I see, that makes sense.  What time frame do you have for implementing
109 based on 101 changes?
Gil: Maximum two weeks
Vlado: ADC code not yet in tree, but will be soon
Phil: Lets check out assembly and make sure that the new code is reasonable
to drive implmentation discussed in TEP 109

== TEP 103 ==
Small changes to table giving example of each chip, can look up things in
data sheet now for more detail
Major change - Section (Storage in Tinyos-2.x) rational for architecture
 * follows lots of suggestions given by David Moss
 * leaves door open for using both low level, and high level implementations
Fixed some typos
More careful about use of SHOULD and MUST
Interfaces not actually in the TEP, but just pointer to them (reference with
html link)
Might be worth making Block storage more general based on David Moss's
comments
 * provide lower level abstration (allows erase, bunch of consecutive writes
not to same bytes, sync, check crcs)
 * removal of verify/commit interface, need to read crc explicitly and check
it yourself
 * can do more low level operations now
 * take a look at my followup email to last meeting as its more coherent
than whats said here

Vlado: need to fix table in TEP to be proper docutil table
Jonathan: Changes seem reasonable, would still work with Deluge. Verify not
really necessary, sync more useful.
David: can you do implementation of smt25p to use new code, I'll put it on a
branch for you to check out
Jonathan, sure, should be trivial, can get it done by end of the week.

Phil: interface signature should be appendix rather than pointing to tree
directly, useful to see them right there
David: I agree, I'll do it that way

== Misc ==
Meeting time change:
 * Preferences for wednesday has won out!!!
 * new tinyos-2.0 WG meeting time is Wednesdays at 9:30am. PST
HIL meeting discussion:
 * Lets just do it over email
 * Vlado will send out initial email
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.millennium.berkeley.edu/pipermail/tinyos-2.0wg/attachments/20060920/b190436b/attachment.html