UEC dynamic power management

Registered by Dustin Kirkland  on 2009-11-09

This blueprint is a follow-up to Karmic's Cloud Power Management:
 * https://blueprints.launchpad.net/ubuntu/+spec/server-karmic-cloud-power-management

In Karmic, we implemented a new scheduling algorithm in Eucalyptus, SCHEDPOLICY=POWERSAVE, which "prefers" systems which are powered on already when allocating new instances. Additionally, we added a "reaper" program in Eucalyptus that would detect idle nodes and bring them to a lower power state, by sending a signal to the PowerNap daemon running on each node.

I performed a statistical analysis of this approach here:
 * http://blog.dustinkirkland.com/2009/08/statistical-analysis-of-powernap-energy.html

While not perfect, it certainly provides some powersavings in most cases.

To totally optimize UEC power efficiency, however, we need a mechanism to "compress" the cloud. This basically entails "evacuating" all VMs off of some systems, live-migrating them to other running systems, and then powering down the idle nodes.

Blueprint information

Status:
Not started
Approver:
Jos Boumans
Priority:
Undefined
Drafter:
Dustin Kirkland 
Direction:
Needs approval
Assignee:
Dustin Kirkland 
Definition:
Discussion
Series goal:
None
Implementation:
Deferred
Milestone target:
None

Related branches

Sprints

Whiteboard

jib review 2009-12-18: As we rely on Eucalyptus for this, and they have deferred this feature for Lucid+1, so will we.

This blueprint depends heavily on upstream development by the Eucalyptus Team.

Dependencies include:
 1) a persistent, remote, networked data storage for backing disk images
 2) working, tested, stable, sec ure live-migration by kvm + libvirt as infrastructure technologies
 3) network configuration that's capable of dealing with a VM that has changed hardware
 4) a suitable compression algorithm for selecting which VMs to move, which hosts to migrate to and from
 5) command line and/or web interface tools for triggering the compression event

(?)

Work Items