nova-compute charm should set vm.swappiness to 0 or 1

To be a little more explicit; with the current/default swappiness,
we're trading off guest memory latency for extra page cache on the
host. Whether you're converged or not, that doesn't seem like a good
trade off to me.

I agree, lowering vm.swappiness is beneficial on kvm hosts.

I'd like to see this exposed as a config value, as the wider impact to users of a forced default behavior change of 60 to 1 or 0 is unmeasured.

This is exposed as a config value already via the sysctl charm config option. I think what we're talking about here is setting a best practise default

Fix proposed to branch: master
Review: https://review.openstack.org/431568

Reviewed: https://review.openstack.org/431568
Committed: https://git.openstack.org/cgit/openstack/charm-nova-compute/commit/?id=9b9a712ec4e96e2af2d05476cb7e48c8d1602f16
Submitter: Jenkins
Branch: master

commit 9b9a712ec4e96e2af2d05476cb7e48c8d1602f16
Author: Liam Young <email address hidden>
Date: Thu Feb 9 11:57:45 2017 +0000

    Lower vm.swappiness to reduce guest memory latency

    Lowering the value of vm.swappiness to 1 (minimum amount of
    swapping without disabling it entirely) should reduce the
    guest memory latency.

    The default of 1 in the charm can be changed by the user by
    explicitly setting vm.swappiness in the sysctl charm setting.

    Change-Id: If9d3a9a0d15a84b86cfa7ba3620b66e7ea61d414
    Closes-Bug: 1660248