Summary list of ARs from this meeting: AR for Mary/Bryce: update the website to have a link to the test cases and scripts. AR for Mary: who is the TOP maintainer? AR for Natalie: Gather as much data as possible to determine why systems don't boot w/ new CPU patches. AR for Martine: Generate a list of items to be used as a starting point for the memory hotplug testing plan. ################################# Attendees: George Mann, individual contributor Joel Schopp, IBM Mary Meredith, OSDL Mark Wong, OSDL Bryce Harrington, OSDL Bruce Vessey, Unisys Natalie Protasevich, Unisys Martine Silbermann, HP * CPU regression testing Bryce ran the script for test case #2 (process migration from offlined CPU) in a loop. It ran for 85 hours and produces 164,348 iterations of CPU on/off-lining. This test was done on a ppc64 w/ 4 physical CPU running SMT to make it look like 8 CPUs. Kernel 2.6.11 was used without any additional patches. The question of what CPU the process gets migrated to came up in terms of choosing the "most idle" CPU. Joel explained that it's irrelevant since the load balancing code will take care of redistributing the work load appropriately. Here are some options that were suggested for future runs: * run script under full workload (maybe w/ database) * modify script to randomly choose what CPU should be off-lined We also discussed what tree should be used for CPU testing and -mm seems the most appropriate. Because of limited resources it was suggested that only functional testing be done over short periods of time on a regular basis and only occasionally do a real stress test over a long period of time. AR for Mary/Bryce: update the website to have a link to the test cases and scripts. * Memory hotplug regression testing. Bryce tried to compile and boot the latest memory hotplug patches but wasn't successful. It turned out that the base kernel didn't boot either so it might not be an issue w/ the patches. The base kernel that Bryce used was 2.6.12-rc5 which seems to boot fine on Joel's ppc64 so further investigation is required. Bryce will try to boot w/ the next kernel rc and if the problem persists he'll post it to the list. * Memory hotplug regression testing plan. At the F2F in Paris we discussed the need to have a testing plan for memory hotplug with a well defined set of intermediate milestones. Jim Wasko took the AR to ask the IBM developers to develop this test plan. To get the effort started Martine offered to generate a list of items that could be used as milestones. AR for Martine: Generate a list of items to be used as a starting point for the memory hotplug testing plan. * CPU hotplug patches submissions. Natalie tried to boot her systems with the latest CPU patches. Only 2 out of 5 systems actually successfully booted. The systems were of different architectures types: i386, x86_64 and ia32. The x86_64 which is an 8 CPU box didn't come up SMP, Natalie suspect there might be a scheduler issue. She will gather more information and post her findings on this list as well as lhcs. AR for Natalie: Gather as much data as possible to determine why systems don't boot w/ new CPU patches. Ashok Raj sent a written update: - IA64 BSP offline support was sent to ia64 list for inclusion. Most likely will be scheduled sometime in 2.6.13, after 2.6.12 is released. - x86_64 patches are being discussed, after initial review, send slightly modified patches for -mm inclusion today again. It seems to pass my stress tests for more than 2 days. If no further comments are received, very likely to be part of base in 2.6.13 timeframe. ACPI changes for x86_64 platform CPU hotplug support is next in line. - i386 patches also getting more solid, Shaohua (ACPI suspend resume) took my stress test script and identified two more bugs. - One was fixed earlier, but got re-introduced in -mm, Vatsa sent a fix again. - Variable overflow bug in head.S in i386 startup, which Shaohua has fixed now. Now he reports he can run for over 24hrs of stress test. * Update on use cases: - Dynamic partitioning use case: Silverster hasn't had a chance to start on the Dynamic Partitioning use case but he has Mary's template and should provide a first draft by the next meeting. - Virtualization use case: Martine has gathered all the information sent by the Virtualization team and put it in the use case template format. The team is currently reviewing it. After a couple more rounds of reviews we'll share the use case on this list and take your input before publishing it to a broader audience. * Others Bryce will investigate further the specifics in terms of memory size/type of the itanium 2 system currently in the lab. Mary asked if the patch that Virtual Iron submitted for TOP to avoid that TOP crashes when CPUs are offlined had been accepted in mainline. Apparently not....more investigation required. AR for Mary: who is the TOP maintainer? Next meeting is scheduled for June 21st at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST Thanks for your participation. Martine J. Silbermann
