Summary list of ARs from this meeting: AR for Joel (repost): Find out if IBM is willing to share their user memory remove tests and in what time frame. AR for Mary: 1. Update the memory hotplug testing plan to include Christoph's suggestions. 2. Send mail to lhcs mailing list regarding content of /proc after CPU is brought back online. 3. Find the person that originally communicated w/ Bryce on that subject and ask him/her to validate the script. AR for Martine: 1. Contact Ric Wheeler from EMC to get the right VMware contact. 2. Send a reminder mail to DCL TWG that it is important to get the memory patches reviewed immediately (as pointed out in the hotplug report - almost yellow item). Attendees: Mary Edie Meredith, OSDL Bryce Harrington, OSDL Bruce Vessey, Unisys George Mann, Indivual Contributor Martine Silbermann, HP * Report from Joint Initiative Face 2 Face in Beijing Martine gave an update on the SIG's activities to the technical Working Group. The slides presented are in the attachment. There were a few comments and suggestions from the participants: * address/involve the consumer product people like VMware that might be interested in participating in the hotplug memory effort since currently they need to shut the image down if they want to add/remove memory. AR for Martine: Contact Ric Wheeler from EMC to get the right VMware contact. * address the desktop and/or laptop communities. The infrastructure of the hotplug IO could be tested at a much higher rate by a much bigger number of users. The SIG involvement in hotplug IO has been minimal due to lack of resources and interest in participation from that community. However getting those communities involved whenever possible to test different features is an excellent suggestion. * someone from the DCL TWG (Technical Working Group) brought up the question of relevance of hotplug now that virtualization is becoming more common and that we have the option to migrate VMs to allow the hw to be freed up and serviced. The consensus was that the software part of hotplug is still needed to handle VM operations. We might want to point out in the use case relative to reliability this specific scenario. * Memory hotplug common regressions - testing plan. Martine contacted Christoph Lameter to ask about tests/scripts for memory fragmentation. He suggested adding page migration testing to the test suite (it's unclear at this point whether he offered to share some existing tests or asked if we could develop some on the subject). He also suggested that it would be helpful to have some tests that show where the memory is ending up after the memory migration (i.e are the mems_allowed of processes obeyed?). AR for Mary: update the memory hotplug testing plan to include Christoph's suggestions. * Update on memory hotplug patches. Joel couldn't attend the meeting but sent in an update on the memory patches: Memory hot add - still in mm, still expect it to go in 2.6.15 Fragmentation avoidance - expect a revised set of patches this week, we really need to push for quick acceptance of these patches in order to get them into distro enterprise releases, which we expect to base on 2.6.15. If the patches can't go into 2.6.15, which is very probable, we at least need to have them in -mm to have a shot at distros picking them up from there, or backporting them from 2.6.16. Memory Migration - Christoph Lamenter from SGI is pushing these. There is a series of 7 patches we are submitting to go into -mm. The same distro considerations apply to these as to fragmentation avoidance. He made a plea again to have as many developers as possible from all OSDL member companies who have an interest in memory hotplug succeeding make sure their developers comment on these two patch sets over the next 3 weeks. AR for Martine: Send a reminder mail to DCL TWG that it is important to get the memory patches reviewed immediately (as pointed out in the hotplug report - almost yellow item). * Last meeting's ARs. Regarding the value in of /proc for the number of interrupts handled by a given CPU after that CPU has been offlined then onlined, we need to contact the CPU hotplug community to report the problem. AR for Mary: Send mail to lhcs mailing list regarding content of /proc after CPU is brought back online. * Others. Bryce being the only person at OSDL currently working on writing scripts for the memory regression testing the question of asking for more resources came up. He's still working on trying to resolve the issue of being able to automatically revert to booting a known good kernel if/when the current kernel doesn't boot on ppc64. It was suggested that we start by writing a script for memory-add; as mentioned in previous meetings the idea would be to boot the kernel w/ less memory than available then to add the memory and test that the added memory is present and usable. Bryce and Mary feel that the script for test case 1 for CPU regression testing might not be adequate, we need to get feedback on the validity of that test from the CPU community. AR for Mary: find the person that originally communicated w/ Bryce on that subject and ask him/her to validate the script. Next meeting is scheduled for November 8th at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST. Thanks for your participation. Martine J. Silbermann
