Le mercredi 24 octobre 2012 à 19:06 +0400, Michael Tokarev a écrit : > Note that -numa option has nothing to do with HOST numa > configuration. It merely adds some topology information > to guest, it does not change how things run on host. > So it is only about testing various numa-related features > in guest. > > Which is of very limited usage, useful only for very > limited configurations -- this is why I don't understand > why you consider this an "important" issue.
I see your point and I'll understand if you don't fix this problem. The issue here is more a performance issue than a missing feature. Imagine that the host has 2 numa cells and you wanna create a virtual machine with all the ressources of the host (all the cpus and all the memory available). Without a correct numa feature on the guest, the guest will think it handles numa cells correctly but on the host, it will not be the case. So memory pages will have to switch from one numa cell to another. If numa parameter works fine, you can have the same numa configuration on the host and the guest so memory is handled at its best. It also works if you don't use all the ressources. If I choose to use cpu 0,2,4 (on numa cell 0) and cpu 1,3,5 (on numa cell 1) on the guest, it will be better if the guest use the matching numa cells too. Sincerely -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
