Le 28/02/2011 21:35, Jeff Squyres a écrit : > On Feb 28, 2011, at 3:31 PM, Brice Goglin wrote: > > >>>>> That would seem to imply that I should always hwloc_set_area_membind() if >>>>> I want it to persist beyond any potential future swapping. >>>>> >> The kernel only looks at the current virtual memory area policy. If you >> didn't call set_area_membind, the default policy is used. It doesn't >> care about the location of the old page before swap-out. >> > I'm not sure exactly what you mean here -- if I call hwloc_set_area_membind() > to bind an area to numa node X, and then area pages get swapped out, then > swapped back in, will be guaranteed to be swapped back in somewhere on numa > node X? (I don't care if they're exactly the same physical page) >
Yes. set_area_membind() will make things work as you wish. "No set_area_membind()" lets pages go on random NUMA nodes (but still *preferably* near the process that first-touches them or swaps them back in). That's what I call "default policy". Brice