Thus it seems that those two things have fixed the problems. *t
Since upgrading to the 4.5.0-0.bpo.2-amd64 kernel and setting
"vm.min_free_kbytes = 65536" we have not seen any more "page allocation
failure" on our servers.
