On 1/16/2016 9:49 AM, Andrea Arcangeli wrote:
In short I don't see the KSM sharing limit ever going to be obsolete unless the whole pagetable format changes and we don't deal with pagetables anymore.
just to put some weight behind Andrea's arguments: this is not theoretical. We're running 3500 - 7000 virtual machines on a single server quite easily nowadays and there's quite a bit of memory that KSM will share between them (often even multiple times).. so your N in O(N) is 7000 to many multiples there of in real environments. And the long hang do happen... once you start getting a bit of memory pressure (say you go from 7000 to 7200 VMs and you only have memory for 7150) then you are hitting the long delays *for every page* the VM inspects, and it will inspect many... since initially they all (all 200Gb of them) are active. My machine was just completely "out" in this for 24 hours before I decided to just reboot it instead. Now, you can make it 2x faster (reboot in 12 hours? ;-) ) but there's really a much higher order reduction of the "long chain" problem needed... I'm with Andrea that prevention of super long chains is the way to go, we can argue about 250 or 500 or 1000. Numbers will speak there... but from a KSM user perspective, at some point you reduced the cost of a page by 250x or 500x or 1000x... it's hitting diminishing returns. -- 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/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>