Paul Jackson, John Hawkes, Terry Prickett, Steve Carbonari, Rick Lindsley, Shailabh Nagar, Bill Hartner, Peter Wong, Johnathan Lahr, Martin Bligh, Bill Irwin, John Stultz, Jack Vogel, Hanna Linder, Ruth Forester, Paul Dorwin, Dave Olien Bare Bones Summary: I. Shailabh Nagar talked about the results they saw with their patch that increases block size running the readraw. The results seemed counterintuitive: Performance decreased with increased block size, even though fewer requests go out. See Shailabh's email on lse-tech. Also the question was asked about what are the best benchmarks to run for this stuff? Dbench and Aim7 were mentioned but are there others? II. David Olien and his IO APIC routing patch. People at Intel and IBM tested it using webbench and netperf on 4-way systems and so far have seen little performance improvement. Bill Hartner's group will try it on SpecWeb99. David is considering where to go with this. III. Hanna Linder working with Maneesh Soni on his dcache_lock patch. She tested it on an 8-way system running dbench with 70 clients and is seeing little performance improvement so far. She and Maneesh are analyzing the lockmeter results, more on this next time. Bill Hartner has seen low dcache_lock contention in his testing as well. His group will try it on specweb99 later. IV. Jonathan Lahr and the io_request_lock. He is exploring minimizing the scope with a new patch. Jack Vogel noticed the discussions so far have remained high-level and do not point out specific locking problems. V. Rick Lindsley and the Locking Document. He will add a new page to the lse locking web site (lse.sourceforge.net/locking) to cover all the versions of the kernel for which he has updated the doc. Expect to see up to 2.4.9 in a few days.