On Sun 08-09-19 13:45:13, David Rientjes wrote: > If the reverts to 5.3 are not > applied, then I'm not at all confident that forward progress on this issue > will be made: David, could you stop this finally? I think there is a good consensus that the current (even after reverts) behavior is not going all the way down where we want to get. There have been different ways forward suggested to not fallback to remote nodes too easily, not to mention a specialized memory policy to explicitly request the behavior you presumably need (and as a bonus it wouldn't be THP specific which is even better). You seem to be deadlocked in "we've used to do something for 4 years so we must preserve that behavior". All that based on a single and odd workload which you are hand waving about without anything for the rest of the community to reproduce. Please try to get out of the argumentation loop. We are more likely to make a forward progress. 5.3 managed to fix the worst case behavior, now let's talk about more clever tuning. You cannot expect such a tuning is an overnight work. This area is full of subtle side effects and few liners might have hard to predict consequences. -- Michal Hocko SUSE Labs
