I tried to capture what I heard today at the meeting on memory migration. http://www.developer.osdl.org/maryedie/HOTPLUG/planning/testcase3_memory.txt I also updated the plan matrix to point to it: http://www.developer.osdl.org/maryedie/HOTPLUG/planning/hotplug_memory_test_plan_status.html Bryce, I'd appreciate your review to see if this is the kind of description you requested in your "lessoned learned" email. Before I go on to memory add/remove, I had a few questions as I wrote this out (and please excuse my ignorance). Assume we trigger a memory migration for a segment in user space, and we have a (test) kernel module looking at the behavior. Is it possible that the memory segment does not migrate to physical memory while the originating segment goes away, i.e. the segment contents (or parts) are written to the swap space? Or is it possible that some pages get swapped out quickly so by the time the kernel module looks at the contents, they are not in physical memory? And related to the above, could swapping behavior be an indication of a problem, assuming that we take care to have a small memory footprint capable of avoiding swap altogether? Finally, assuming we write a test kernel module, how does it know what is target physical memory? Is the target memory a contiguous memory location, or can it be a bunch of randomly placed pages (possibly in swap, see above) adding up to a memory segment? -- Mary Edie Meredith Initiative Manager Open Source Development Labs maryedie@xxxxxxxxxxx 503-906-1942
