On Sun, May 15, 2016 at 01:15:15PM +0200, Sven Eckelmann wrote:
TODO statistics, batadv_tt_roam_node?, batadv_tt_change_node?, batadv_tt_req_node?, ....
Some statistics here:
https://www.open-mesh.org/projects/batman-adv/wiki/Kmalloc-kmem-cache-tests
In a nutshell, on a small TP-Link TL-841NDv8 wireless router (MIPS/AR9341), I was able to achieve 34% more global TT entries (or 25% less RAM usage).
Regarding SLAB_HWCACHE_ALIGN, it did not seem to make much of a difference. There is a slight tendency for less RAM usage without this flag. But it is so small, that it might be a coincident, too. And it is definitely negligable. At least for this architecture.
Regards, Linus
PS: I might make some more, pretty graphs for the other test variables, too. But for now, there is just this one raw table for that.