Linux Scalability Effort - Call on 5/18/01
------------------------------------------
o Tim Wright:
	Gathering TODO list, currently 54 items, brief highlights:
		MQ scheduler progress
		kernel preemption (Nigel)
		Multi-block buf header
		64-bit PCI support (w/o bounce buffers)
		NUMA
		page stealing
		kernel lock elimination
		TCP stack
	Locking document (Rick) is about ready to be sent out for review
	Benchmarking: comments about std benchmarks
		note: No "DeWitt" clause for DB2 (meaning performance
			numbers can be talked about without vendor permission)
o John Wright:
	John Hawkes will have updated lockmeter/kernprof pathes for ia64 soon
	Nick Pollitt passed around another update to the runon patch
	Kanoj still gathering requirements for the NUMA API spec
	Ulf looking at NGPT kernel patches
o Wookie:
	OSDL needs to have performance tests for developers to compare their
		changes instead of endlessly debating on the lkml without
		hard data
	OSDL plan is to have automated, standard workloads, already set up
		for developers to reserve time on and results to go into
		a results database
	Jwright: key is the workloads chosen, Wookie: it will be a combo
		of realistic benchmarks and interesting micro benchmarks
	Framework is the first step - plan to develop thru the Linux Testing
		Project
o Steve Carbonari:
	looking into dd hang - large dd's hang in the blkdev_put code, they
		grab lock and then go to sync_buf code which walks the dirty
		buf list (which can be really long) over and over
	Asked about rawio interfaces. Jwright: we tried to submit a patch
		last year to make a normal rdsk interface. sct originally
		wanted a similar interface but Linus wanted the /dev/raw.
		So we are stuck with it unless we convince Linus otherwise.
	Possibly setting up a "well-known" DSS workload