(In reply to Nilesh from comment #0)
> The reason we couldn’t create the mavinir instance is that we don’t have
> enough free memory (HugePages) on p23.
> Mavenir instance needs 65536MB memory, and we only have total 21504 MB free
> huge memory left.
>
> Then why openstack still assigns compute resource from p23 to the instance?
> why used_ram, which doesn’t tell the truth. And because of this the nova
> scheduler think we will have enough free memory left on p23, and then
> allocate the compute resource from p23 to this instance.
>
Is the NUMATOpologyFilter well configured? is the only piece of code which is aware of the difference kind of memory pages in a system. Nova in general only care about small pages.