Record of LSE discussion Dec 13 2002

I. Martin Bligh - 

	The -mjb tree is a scalability tree against 2.5 with 
SMP & NUMA stuff.  He is trying to collect whatever patches people 
want to use, as long as they seem mergeable into mainline at
some point. There are 20 patches in there at the moment and they 
dont conflict too much. Separate patches are maintained, but not
published yet.
	If you have anything that improves benchmarks or real workloads 
send it to Martin and he will try to collect them. He isnt doing too much
experimental stuff (leaving that to akpm for the moment). Generally
his tree is focusing on 4 cpu or above or more than 4gb memory.
Espeically large ia32 systems but large systems in general. 

II. Bill Irwin-

	His tree is both more and less radical than martins. 
With a different kind of scalability.  Things such as task 
scalability and vm things that there is no maintainer for currently. 
hugetlbfs stuff will probably stay in his tree before sending 
to akpm. One patch will help desktop users is the sped up /proc. 
	All his patches are broken up and separate. He is trying 
to avoid overlap with any of the other trees. However, he is looking 
to stick to things that cant be found elsewhere in here. 
	It mainly contains patches he wrote but will also accept 
things from other people, espeically it if it interesting to Bill 
and no one else wants it.

III.  Making ps or top not suck -

	Martin asked if anyone is working on making ps or top
to make it suck less. He said they both open about 5 files
in /proc for every single process. It is bad that it touches the 
inode, dcache and file caches for every thing and it shouldnt
need to do a syscall for each file, that is fixable according 
to martin. bill said there are 3 problems, syscall traffic, 
never guarenteed to keep up so have to scan incrementally, 
kernel side has to not fill directories.
	Martin is going to look for someone to work on it.

IV. ideas for next year.

- Start bringing in lkml discussions instead of just
	lse tech talks. 
- More talks about what people are planning on doing 
	instead of what people have already done.

- Remind people of one benefit of these calls is 
	allowing people to talk virtually face to face 
	and people can have real conversations.

V. Next year

next time we will talk about scheduling issues in NUMA.
This meeting will be on January 10 and there will be 
new numbers and passcodes. 

VI. General talk about stuff with John Hawkes about
kernprof and supporting 2.5 kernels and other cools
things coming soon from him.