Minutes from 11/02 LSE Con Call

Hanna Linder, Jack Steiner, Shailabh Nagar, Bill Brantley, Paul Jackson,
Paul McKenney, John Wright, Ray Bryant, Steve Carbonari, Steve Henkels,
Krisna Swami, Paul Dorwin, Pat Gaughen, Jess Barnes, MingMing Cao, John Hawkes,
Duk Vianney, Terry Prickett, Hans Tannenberger, Ruth Forester, Matt Dobson,
Martin Bligh, Richard Griffiths, Gerrit Huizenga, Bill Irwin, Dave Olein

* indicates to do item

I. Hanna Linder- Should we move the con call time 1.5 hours ealier?

	That would make the new time 9:30am PST, 11:30am CST, 12:30pm EST
	Nobody opposed it. Will enable people in India who chose to stay up
	until 11pm to call in. 
*Hanna*- will send out the info about an International number later.

II. Shailabh Nagar- Raw IO update on LSE web page.

	Basic idea is to split IO into 3 parts. First two parts
	aligned on page sizes the third chunk is not page aligned.
	Reason explained on lse web page.  Seems to work fine, 
	working on code improvements. Such as IO not more than a page 
	in length will not use this mechanism.

	Ruth has tested it and seen the time of application cut in half 
	(only tested with sequential reads).

*Ruth*- make sure there is no read/write corruption occuring. 
	Andrea Arcangeli reported problems with file system creation or 
	read/write. 

III. Paul Jackson-

	Published latest version of CpuMemSet design notes on lse.sf.net.
	Paul McKenney has been on site at SGI and they have been discussing
	it in person.
	If someone wishes to offer feedback, please do.

IV. Russ Weight- Known Resource Limits in Kernel

	Looking for known limits that could be a problem with
	scalability. Arrays are just a starting point.  Many
	other resources would fall under this task.

*Russ*-Read paper from last year's ALS.

	Jesse Barnes-When I try to compile kernel with 64 bit pages there 
	are places it breaks (on ia64). 
	There was a patch for it but it might be old now.

**- Send changes to Jay Mosberger again.

*Shailabh*- Ask Steven Tweedie or Ben LaHaise regarding 512kb limit.

V. Matt Dobson- Launch Policy Patch work

	If set launch policy for self all children will 
	run on only those cpus. Off Andrew Morton's CPU allowed patch.

	Paul McKenney- Could use a launch policy directory for
	 similar memory policy.

	Martin Bligh- Might cause problmes with renumbering.

	Jesse Barnes- posted a patch to help with debugging this 
	sort of thing. Modify prctl tool with user level tool 
	to control it.

	Nick Pollett- Did a runon patch to allow you to set launch
	policy prior to forking.

*Matt*- Ask him to resend it and possibly update the kernel version.

VI. Dave Olein- 

	Thread code in the Linux kernel now isn't posix compliant. Dave wrote 
	a patch to make threads posix compliant. Where should he send it?
	Gerrit Huizenga- To posix threads group and lkml as rfc patch. 

*Does anyone have a P4 SMP system to test Dave's first patch on?
	Apparently these proccessors dont do the round robin 	
	schedluing like other apics do. 

VII. Gerrit Huizenga- LSE Rollup Patch 

	Working on creating a single patch that includes many
	of LSE existing patches. Would make it easier to compare 
	ongoing changes against best recipe instead of just checked 
	in patches. Also will make applying the patches easier. 
	John Stultz is working on setting it up via CVS tree on LSE sf site.