On 07/11/2012 02:03 AM, Minchan Kim wrote: > Today, I tested zsmapbench in my embedded board(ARM). > tlb-flush is 30% faster than copy-based so it's always not win. > I think it depends on CPU speed/cache size. After you pointed this out, I decided to test this on my Raspberry Pi, the only ARM system I have that is open enough for me to work with. I pulled some of the cycle counting stuff out of arch/arm/kernel/perf_event_v6.c. I've pushed that code to the github repo. git://github.com/spartacus06/zsmapbench.git My results were in agreement with your findings. I got 2040 cycles/map for the copy method and 947 cycles/map for the page-table method. I think memory speed is playing a big roll in the difference. I agree that the page-table method should be restored since the performance difference is so significant on ARM, a platform that benefits a lot from memory compression IMHO. Still, the question remains how to implement the selection logic, since not all archs that support the page-table method will necessarily perform better with it. Thanks, Seth -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>