On Wed, May 30, 2012 at 01:34:21PM -0500, Christoph Lameter wrote: > On Wed, 30 May 2012, Linus Torvalds wrote: > > > On Wed, May 30, 2012 at 2:02 AM, <kosaki.motohiro@xxxxxxxxx> wrote: > > > > > > So, I think we should reconsider about shared mempolicy completely. > > > > Quite frankly, I'd prefer that approach. The code is subtle and > > horribly bug-fraught, and I absolutely detest the way it looks too. > > Reading your patches was actually somewhat painful. > > It is so bad mostly because the integration of shared memory policies with > cpusets is not really working. Using either in isolation is ok especially > shared mempolicies do not play well with cpusets. Yes the cpusets did some horrible things. I always regretted that cpusets were no done with custom node lists. That would have been much cleaner and also likely faster than what we have. > > If we could just remove the support for it entirely, that would be > > *much* preferable to continue working with this code. > > Well shm support needs memory policies to spread data across nodes etc. > AFAICT support was put in due to requirements to support large database > vendors (oracle). Andi? Yes we need shared policy for the big databases. Maybe we could stop supporting cpusets with that though. Not sure they really use that. > Its not going to be easy to remove. Shared policies? I don't think you can remove them. cpusets+shared policy? maybe, but still will be hard. -Andi > -- ak@xxxxxxxxxxxxxxx -- Speaking for myself only. -- 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/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>