Cloud Power Management

Registered by Dustin Kirkland 

I was able to prove in Jaunty that at least some Ubuntu servers can successfully, suspend and hibernate, as well as resume from remote triggers such as wake-on-lan. See:
 * https://blueprints.edge.launchpad.net/ubuntu/+spec/server-suspend-hibernate

As Ubuntu is being promoted as a cloud computing hosting solution, I believe that an integral aspect will be managing the power consumption of the systems providing the virtualization services.

Workloads should be coalesced onto a subset of available hosts, optimizing cpu/memory utilization, via live migration. Under-utilized hosts should be suspended to a low power state, and awaiting a 'wake' signal when demand spikes. As such, these 'suspended' servers will consume a tiny fraction of the power of running systems, yet will be in a hot-standby mode, ready to operate again on demand within seconds.

We should integrate into existing cloud management frameworks (eucalyptus? openebula? ovirt? landscape?):
 a) tools for pro-active sysadmins, who want to manually migrate workloads and manually suspend/resume servers
 b) automatic algorithms that can perform these operations in an unattended manner when certain configurable thresholds or conditions are met

:-Dustin

Blueprint information

Status:
Complete
Approver:
Rick Clark
Priority:
Essential
Drafter:
Dustin Kirkland 
Direction:
Approved
Assignee:
Dustin Kirkland 
Definition:
Approved
Series goal:
Accepted for karmic
Implementation:
Implemented
Milestone target:
milestone icon karmic-alpha-6
Started by
Dustin Kirkland 
Completed by
Steve Langasek

Related branches

Sprints

Whiteboard

Discussion Points:
 * Identify a small, finite set of consumers (1 or 2) of this technology for the karmic development cycle (eucalyptus, landscape, open nebula, ovirt, etc).
 * Identify a working kvm live migration configuration and requirements
 * Identify the mechanisms/protocols by which hosts should be suspended and resumed
  * ssh + pm-suspend ?
  * wakeonlan ?
 * Identify the heuristics and tunables by which a program might use to auto suspend/resume hosts
-- Dustin Kirkland
 * Demo of scheduler prototype for power management on clusters for OpenNebula
-- Tino Vazquez

(?)

Work Items