Interconnect Software Consorium Socket API Extensions Working Group.
--------------------------------------------------------------------

Meeting:
	Thursday 21st March 2002
	21:00GMT (04:00p EST, 03:00p CST, 01:00p PST, 22:00CET, 06:00JST+) 

	Toll Free (US & Canada):          1-888-742-8686
	International:                    +1-303-928-2600
	Conference ID:                    3378790

--------------------------------

Attendees:
	David Edmondson, Sun
	Jeremy Harris, Sun
	Vish Subramanian, HP
	Jack McCann, Compaq
	Annie Foong, Intel
	Masanori Itoh, Fujitsu
	Satya Sharma, IBM
	Juliana Hsu, IBM
	Jeff Messing, IBM
	John Kasperski, IBM
	Louis Laborde, HP

--------------------------------

Agenda:

1. Agenda bashing.

2. Review action items.
   See below.

3. Next meeting.
   28th March is very near Easter.  Do we want to have a meeting on
   that date ?

   Yes.

3.5. How to complete the requirements ?
     dme: can use HP text with additions.
     Vish: need problem statement.
     
     ACTION: dme and Satya to discuss the details of producing the
     requirements spec and report back.

4. Areas of interest discussions.
   * overlap with POSIX AIO

   Vish: POSIX AIO.
	 Walk through the recent posting, discussing various issues
	 (and wandering off somewhat along the way).

	 Vish: not clear about connected UDP sockets.
	 Jack: is there something about async sendto/recvfrom that
	 doesn't make sense ?  Is that why POSIX didn't specify the, ?

	 Vish: because they were disk oriented.
	 ???: completion is more interesting than initiation

	 Vish described the completion implementation in POSIX AIO.
	 The stateless completion mechanism likely has scaling issues,
	 as with select().

	 A debate about correlators ensued, but it mostly appeared to
	 falter around a lack of common understanding about what a
	 correlator actually is.  In some people's minds it is an
	 object which is used to synchronise across async initiations
	 and completions, where one initiated action corresponds to
	 one correlator.  To others a buffer address (in conjunction
	 with socket) is sufficient to act as a correlator.

	 Satya: There is issues with boundary marking when considering
	 IO completion, i.e. in stream transports there are not
	 defined pdu's.
	 dme: Whilst the stream doesn't have distinct pdu's, the
	 buffers used to receive the data are still distinct.

	 Jeff: what about coagulation of received events ?  This would
	 appear to break the "single event <-> correlator" mapping.

	 Satya: maybe this is a buffer issue ?

	 Vish: per-event is important.  IB solicited marking is one
	 solution which has been adopted by IBTA.

	 Jeff: per-id requires per-blob correlator, which is wasteful
	 as we need to keep creating, destroying and tracking the
	 correlator objects.

	 Jeff: what will use cancel ?  For example, what does it mean
	 to cancel an async write ?
	 Louis: cancel can fail

	 Jack: it may be useful to cancel async reads, as a way of
	 removing the relevant buffers from the pool.
	 Vish: DB vendors would like cancel operation

	 Jeff: in IB, not possible to map receive buffers to specific
	 receivers, as they are per-QP

	 Kill all receive buffers is more common than kill specific.
	 
	 dme: connect() is a possible async operation that doesn't
	 have a IO buffer address.
	 Jeff: fd is fine as a "correlator" in this instance.

	 John: Applications need to be able to deal with completion
	 events on "stale" objects.
	 
   * HP requirements paper next week.

5. AOB.

--------------------------------

Outstanding Actions:

2002-03-07/03 (dme)
	Product a summary of requirements extracted from experience
	with /dev/poll.
	dme: Aiming for next week.

	NB: Both Linux and HP have a /dev/poll like interface.  The
	HP-UX interface does not appear to suffer some of the
	limitations apparent in the Solaris interface.
	DONE: 2002-03-21

2002-03-14/01 (dme)
	Agenda item for next week: discuss AIO interfaces
	specifically, including how they relate to POSIX.
	DONE: 2002-03-20 (see above).

2002-03-21/01 (dme/Satya)
	dme and Satya to discuss the details of producing the
	requirements spec and report back.

--------------------------------

Next Meeting:
	Thursday 28th March 2002
	21:00GMT (04:00p EST, 03:00p CST, 01:00p PST, 22:00CET, 06:00JST+) 

	Toll Free (US & Canada):          1-888-742-8686
	International:                    +1-303-928-2600
	Conference ID:                    3378790

--------------------------------

Contact Details:
	email:	icsc-socketwg@opengroup.org
	web:	http://www.opengroup.org/icsc/member/sockets/