Summary list of ARs from this meeting: AR for Bryce: Send mail to Dave and/or community to inquire about desired format to report test results. AR for Mary: Put a list of statistics tool (such as TOP and SAR) on the wiki. AR for David: Look at the SAR issue and provide a patch if needed. AR for Martine: Update use case on virtualization and send out ################################# Attendees: George Mann, Individual contributor David Lively, Virtual Iron Mary Edie Meredith, OSDL Mark Wong, OSDL Bryce Harrington, OSDL Martine Silbermann, HP * Report from OLS. Martine gave a quick report on the hotplug memory BOF. There was a lot of discussion on memory pressure and how one can deal with it to keep the memory zones between "easy to remove" and "hard to remove" memory separate. Too strong memory restrictions are not acceptable (customers and distros wouldn't go for it). A compromise would be to give the user control over how much memory can be allocated to non-reclaimable kernel memory. However making the boundaries between those 2 zones dynamic has its own set of problems. Slab caches are also an issue since they aren't currently very hotplug friendly. Several ideas were formulated: amongst them using Mel's patches which allow you to find the actual physical page from the slab cache, using the slab shrinker function to determine if memory can be removed,... Also there are lots of problems w/ X (since you cannot remove memory that you don't have a reference to). Anyways hotplug remove presents many challenges that provided good material for a very animated and interesting BOF. The discussion that we had during the SIG con call led to the conclusion that it might require major kernel reworks to make memory removal work. In the meantime some hardware provides support that makes removal easier. Virtualization was the subject of many papers and a few BOFs. We also noted that there was more attention given to testing in a more structured way. David mentioned that Virtual Iron does testing over NFS and use netperf for network testing. During the virtualization BOF Rik van Riel gave a brief recap of what happened at LKS in particular that Xen will be merged as an i386 subarchitecture. VMware presented a proposal for a common VMI (Virtual Machine Interface) * Memory hotplug latest patches Bryce noted that patch 2.6.13-rc3-git7-mhp1 failed to apply, Dave Hansen is aware of the issue. * Status of CPU test cases (new scripts) Mark Delcambre worked on the CPU hotplug test scripts, currently 4 out of 6 cases are implemented. To address memory leaks the plan is to run the tests under random workloads. Mary already ran OLTP type workload while taking CPUs off/online. These tests showed that SAR was not behaving properly when CPUs are added. AR for David: Look at the SAR issue and provide a patch if needed. It would be good to keep track of all the statistics tools and make sure they work w/ hotplug patches. AR for Mary: Put a list of statistics tool (such as TOP and SAR) on the wiki. Since the goal is to do regression testing on the mm tree and all rc candidates, Bryce is planning to automate both running the tests and analyzing the results. What format would be most efficient to communicate test results to the community? AR for Bryce: Send mail to Dave and/or community to inquire about desired format to report test results. As soon as the final memory add patches are available we can start testing them as well. * Status of use cases. Status quo. Martine is working on updating the virtualization one and will publish it shortly. Next meeting is scheduled for August 16th at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST. Thanks for your participation. Martine J. Silbermann
