HAPPY NEW YEAR and best wishes to all of you and your families for a prosperous year 2006. AR for Joel: Ask Mel if he'd be interested in writing a paper on the differences between fragmentation approaches w/ some performance analysis in each case. Attendees: Mary Edie Meredith, OSDL Bryce Harrington, OSDL Bruce Vessey, Unisys Joel Schopp, IBM Natalie Protasevich, Unisys Martine Silbermann, HP * Status on memory remove patches Mel Gorman just submitted a set of patches based on the earlier patches from Yasunori on the zone based approach to implement fragmentation and stay compatible w/ memory remove. The performance of those patches is currently bad but it's unclear if this is inherent to the approach or could be rectified w/ tuning. The light fragmentation patches are still an option that may be reconsidered. The memory migration patches have been submitted and should be accepted shortly. This will help somewhat memory remove but being able to free up large chunks of memory is not going to be easy because currently migration only works on certain types of memory. Their current use is mostly for process migration on NUMA systems. AR for Joel: Ask Mel if he'd be interested in writing a paper on the differences between fragmentation approaches w/ some performance analysis in each case. * Status on CPU testing Mary got some positive feedback from Nathan Lynch on the test case for irq migration. She'll post his reply after getting his permission to make it public. * Status on memory testing Memtoy is a good tool to create different types of shared memory, it has a command line interface and supports different operations to populate the memory and keep track of the migration process. It was written for NUMA specifically but could be modified to general memory migration. It is mostly used as a debugging tool. * Plans for 2006 We started the following list last time: - OSDL just received a new x86 box which will be used to run the CPU tests in a fully automated fashion including the report of the results in a visual format. - For memory testing we need to take the existing test plan and turn it into a set of test cases with adequate implementation details. - Proceed w/ regression testing and some performance testing for memory hot-add. - Modify memtoy for non NUMA system and develop other new tests for memory hotplug. - Write a paper comparing ballooning to hotplug memory approaches for virtualization. - Other arguments in favor of memory hot-remove could be * support for node hotplug, * the fact that hot-remove would work both on the iron and in a VM environment, * the fact that the ballooning driver would not be able to handle Windows as a guest on Xen. To that list we've added: - Develop test and scripts for automated regression testing for memory hot-remove. - Run Xen on a test system and try to add/remove CPUs in the virtual machines. - Provide the Data Center Technical working group with a forecast of what features are expected to go into mainline in what timeframe. The next meeting will be scheduled at the regular calendar date of January 31st at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST. Martine J. Silbermann
