Report host memory bandwidth as a metric in Nova (partial)
This blueprint proposes to introduce host memory bandwidth as a host metric. Memory bandwidth can be a essential piece in determining VM performance bottlenecks and further can be used for better NUMA based placements.
Using Linux platform interfaces like linux perf APIs, nova-compute should be able to expose Host's memory bandwidth utilization on every NUMA node.
This memory bandwidth can be leveraged in Openstack by exposing it as a monitor.
This will follow a similar approach as the already existing monitor for CPU and will be optional.
Blueprint information
- Status:
- Complete
- Approver:
- John Garbutt
- Priority:
- Not
- Drafter:
- Sudipta Biswas
- Direction:
- Needs approval
- Assignee:
- Sudipta Biswas
- Definition:
- Obsolete
- Series goal:
- None
- Implementation:
- Beta Available
- Milestone target:
- None
- Started by
- Sudipta Biswas
- Completed by
- Matt Riedemann
Related branches
Related bugs
Sprints
Whiteboard
Gerrit topic: https:/
Addressed by: https:/
Report host memory b/w as a metric in Nova
Addressed by: https:/
Allow NUMA based reporting for Monitors
Removing the target of liberty-3 as all non-priority blueprints must be merged by liberty-2 --johnthetubaguy 15th July 2015
We should try merge what we have here, so marking this as partial, and that it needscodereview. If more code is uploaded by the end of the day, that can be included too. --johnthetubaguy 16th July 2015
Addressed by: https:/
Base enablement for memory bandwidth monitor
Addressed by: https:/
Introduce utility module for pcp
Turns out not all the code is really up for review, so it should have been delayed in the feature proposal freeze.
Sorry, we have now hit the non-priority feature freeze for Liberty. You will need to resubmit this blueprint for Mitaka or apply for an exception. For more details on why this is happening, and the rest of the process details, please see: https:/
--johnthetubaugy 3rd July 2015
Addressed by: https:/
Report host memory bandwidth as a metric in Nova
====
Merged patches:
Addressed by: https:/
Allow NUMA based reporting for Monitors
=====
Pending patches:
Gerrit topic: https:/
Addressed by: https:/
Base enablement for memory bandwidth monitor
Addressed by: https:/
Introduce utility module for pcp
Addressed by: https:/
Allow monitor plugins to set own metric object
Addressed by: https:/
Remove hardcoding namespaces for monitors
Addressed by: https:/
Base enablement for memory bandwidth monitor
Addressed by: https:/
add memory bandwidth base class
Addressed by: https:/
Allow monitor plugins to set own metric object
Addressed by: https:/
WIP: define NUMA memory bw monitor
Sorry, we have now hit the Non-Priority Feature Freeze for Mitaka. For more details please see: http://
--johnthetubaguy 2016.01.30
Addressed by: https:/
Report host memory bandwidth as a metric in Nova
Addressed by: https:/
Introduce the memory bandwidth monitor
Addressed by: https:/
Update setup.cfg for memory bandwidth monitor
There are concerns over needing the external native package installed for pcp to make this work, and how it's different from how we get CPU stats/metrics from the virt drivers. There isn't an alternative in libvirt today though, so we'd be far out from using something directly from libvirt. I personally don't think we have enough consensus on getting this in for Newton with a FFE but wanted to bring it up for completeness. The core team will discuss in #openstack-nova in freenode IRC on Wednesday 7/6. -- mriedem 20160705
There are too many open questions/concerns around this, and gathering metrics within nova in general, so we're going to hold off on this for Newton and plan on discussing the future of metrics monitors in Nova and it's scheduler at the midcycle meetup. -- mriedem 20160706
For an update on this from the Newton midcycle, see:
http://
http://