Summary list of ARs from this meeting: AR for Russ(re-post from last meeting): Contact Christoph Lameter for input on fragmentation testing. AR for Mary: 1. Add a new dimension (real arch or virtual machine) to the test plan. 2. Update the memory test plan with the memory migration data. AR for Martine: Find out what the statistics tools output when run on a VM. ################################# Attendees: Joel Schopp, IBM Mary Edie Meredith, OSDL Bryce Harrington, OSDL Bruce Vessey, Unisys Natalie Protasevich, Unisys Martine Silbermann, HP * Status report on CPU test cases The list of statistics tools used for test case 6 needs to be updated, for example getcomp from the syscomp package provides the number of online CPUs as well as the number of physical CPUs in the system. Also we need to find out what those tools would/should output when they're run on a VM. For example should the tools consider the max number of CPUs to be the max number assigned to the VM or the actual max physical number of CPUs. Also in cases like Xen when you can have more virtual than physical CPUs what should be considered. AR for Martine: Find out what the statistics tools output when run on a VM. Mary proposed to add a new dimension to that test case scenario: real architecture or virtual machines. The archs considered currently are ppc64, ia64, x86_64, ia32 and the virtualization implementations are Xen, Virtual Iron VFe and VMWare GSX. AR for Mary: add this new dimension to the test plan *Discussion to identify memory hotplug common regressions - memory migration (cont. from last meeting) Memory migration will be part of memory hot-remove and should therefore be included in the test plan. There are currently no scripts or formal tests for memory migration. Joel suggested to develop a kernel module that would allocate memory and fill it with a known pattern, then identify the section of memory that it was allocated from, migrate that section and finally verify that the migration happened properly by checking the content of the memory for the known pattern. We discussed the most common issues with memory migration which are that the migration doesn't happen properly (not all the memory is moved or the content of the memory after the migration is erroneous) or that the application using that chunk of memory is trying to access it at the old location after the migration is completed. The problem of process migration on a NUMA box was also discussed. When the task manager migrates a process to a different node the memory used by that process should be migrated to that node as well to optimize locality. This is a NUMA specific issue and will not be included into the test plan at this time. Stress testing using OSDL's DBT2 test would exercise most of the memory and will be added to the test plan. AR for Mary: Update the memory test plan with the memory migration data. * Status of use cases. Martine has posted the new version of the virtualization use case to both the hotplug SIG and the virtualization mailing lists....waiting for feedback. Silvester was sidetracked over the summer with the software patent issue raging in the European Community but is working again on the dynamic partitioning use case. Next meeting is scheduled for September 13th at 11:00am -12:00pm PST, 2:00pm - 3:00pm EST. Thanks for your participation. Martine J. Silbermann
