Summary list of ARs from this meeting: AR for Bryce: merge the information between your web page and Mark's testing plan. AR for Martine: 1. Organize a call to discuss OpenHPI control vs monitoring issues. 2. Ping Silvester for status on dynamic partitioning use case 3. Get/integrate feedback from virtualization team and post an updated version ################################# Attendees: George Mann, individual contributor Joel Schopp, IBM Mark Wong, OSDL Bruce Vessey, Unisys Natalie Protasevich, Unisys Randall Loomis, Wind River Martine Silbermann, HP * Discussion on the hotplug CPU test cases (update/revisit). There are currently 2 separate versions of the testing plan, one generated by Bryce: http://developer.osdl.org/dev/hotplug/documentation/cpu_test_cases.php and one generated by Mark http://developer.osdl.org/maryedie/HOTPLUG/planning/hotplug_cpu_test_pla n_status.html The content of this 2 documents should be merged (AR for Bryce). For example test case 6 which checks that management tools such as sar and top still work when CPU are taken off/on-line should be part of our testing plan. For test case 3 Mary recommended that we alter it slightly to span 5 processes per cpu configured so that the likelihood of process migration to the newly on-line cpu will be higher (than if there is one per cpu). However the issue isn't that much the number of processes as it is the activity level of the CPUs. In order for the load balancing code to reassign processes to the newly on-lined CPU the workload has to be such that the other CPUs are busy. In her status report Mary also wrote "Test Cases 1 and 5 do not agree", since no one on the call could understand why, this issue needs to be revisited during the next con call when Mary is on the call. We discussed the effect of taking CPUs offline on the CPU affinity set. When a process is bound to a CPU set and all the CPUs in that set eventually get offlined the CPU set is "extended" to all online CPUs. When the CPUs that the process was originally bound to comes back on line the original information of favoring that CPU for the process is lost. The consensus was that this current behavior is good enough and that keeping track of original CPU sets might introduce excessive overhead that is not worthy. * Discussion on OpenHPI's ability to control in addition to monitoring. Since no OpenHPI representative was on the call we decided to address this matter at a latter time in a smaller forum. If you are interested in participating please contact me directly, currently the list of participants include George Mann, Randall Loomis and an OpenHPI person (TBD). I'll send an invite to the hotplug SIG w/ logistics whenever it'll happen. * Status of memory testing plan. Mark and Bryce put the ppc64 in the OSDL lab in a rack. No booting/testing activity. Joel reported that a lot of the activity on the development side in the past couple of weeks has been focused on sparsemem extreme which allows handling of larger NUMA platforms. The memory hot add patches are still being evaluated, the expectation is that they make it into 2.6.13 * CPU hotplug patches. Natalie is still having trouble w/ the latest -mm tree on x86_64. She suspects that the hang that happens at boot time is not related to CPU patches specifically but might be a scheduling issue. * Update on use cases: - Dynamic partitioning use case: No update. AR for Martine: ping Silvester for status - Virtualization use case: The document license wording is being changes, we're waiting on OSDL attorney to provide the proper wording. AR for Martine: get/integrate feedback from virtualization team and post an updated version * Others Patch for top: there was a discussion over mail on what best to use for the number of CPUs. After updating his patch David will submit it to the maintainer, at the same time Joel send it to RH and SuSE to be integrated in their next enterprise releases. Joel will make the presentation at OLS, he also plans to have a demo (hopefully live but w/ a back-up simulation on slides...just in case). Next meeting is scheduled for July 5th at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST, since I'll be on vacation that week Mary will be the moderator. Thanks for your participation. Martine J. Silbermann
