LSE Con Call Minutes from 07/12/2002

1. Daniel Phillips - 2 pieces. 1st early page patch allows you to declare
struct page as the first thing in mm.h. going to submit it formally.
2nd piece builds on 1st introduces new logical addressing model which
replaces the physical model which assumes all memory is linear.
see his original postings on lklm and lwn.net for more indepth 
description. Should be seeing the patches soon. going to pass the
2.4.17 patches to mbligh then martin will get them working with
i386. Want to get it to 2.5 soon but will start with 2.4 to get it
working. The explaination is more complicated than the code.
	How would this be used? To be able to map a little bit of
zone normal memory on each node. currently on numa machines only
the first node gets zone normal. an efficiency benefit mainly.

2. MingMing Cao - need to break the lock into one message per
ipc resource. so you dont need to synchronize locks when using
multiple ipc resources. this should reduce lock contention.
Saw almost 50% decrease on lock spinning with an oracle apps
run.

3. Hanna will send out dcache paper and slides to lkml and Al Viro.

4. kmap patches from Andrew Morten. The global tlb flush doesnt
scale well. akpm's patch helps this. shows 15% improvement in
throughput on an 8-way with dbench and specweb. however, his
4-way didnt show any improvement so more work will be done to
verify the 8-way results are not anomolous and to reproduce
on a local 4-way.

5. Adam Litke - working on porting each gcov, kernprof and lockmeter 
up to 2.5.25. Then work on sending out incremental
patches for the most popular future kernels. John
Hawkes agreed to do a review of 2.5 changes to verify
no functional changes occur during porting. Currently
a couple people are having problems compiling 2.5.25.

6. Randy Whron - Trying to determine which patch causes an
improvement in pipe throughput. Believe it is due to 
irq balancing. However on a networking benchmark Andrew
Theurer saw a 10% decrease with the irq balancing patch.
Randy will try it on different workloads if he determines
it is the patch that causes the improvement.

7. Hubertus ported gcov to a newer kernel version and
added support for another architecture.

8. Daniel asked about seeing results of rmap on a numaq.
Martin got distracted but will get back to getting some
results.

9. Daniel talked about cpu bit map maintenance and other stuff 
I didnt record. Ask him.