Event Details

The nova team delivered major changes to how scheduling works in the Pike release by doing cpu/ram/disk resource "claims" in the scheduler rather than in the compute service.

This not only affects scheduling when creating instances, but how resource consumption is tracked during move operations.

There were also changes in how ironic node resources are tracked for scheduling using custom atomic resource classes rather than the exact core/ram/disk filters, which are now deprecated.

There is a lot to take in when upgrading to Pike, and a lot being worked on in Queens in these areas. Therefore, this session will be about:

1. What happened in Pike and what do operators need to be aware of when they upgrade and run Nova in the Pike release.
2. What is the Nova development team working on in the area of scheduling and placement in the Queens release.
3. Try to get feedback from early adopters on issues they are encountering in production, including any performance/scale testing data we can get back, if even a simple, "yes things are scaling better now" or "you've regressed scheduling performance in my deployment".

This session should be attended by operators and groups interested in large scale nova deployments, so LCOO and public cloud SIGs.