2014-03-03 23:44 GMT-03:00 Steve Gordon <[email protected]>: > ----- Original Message ----- > > Hi everyone, > > > > I'm currently taking part of a project at the Federal University of > Campina > > Grande called fogbow, which aims on providing an energy efficient > scheduler > > and an opportunistic compute node (which deactivates the service if the > > host isn't idle) on top of openstack. > > Hi Marcos, >
Hi Steve, thank you very much for the reply, > > This sounds like an interesting idea - have you had a look at the way this > functionality is implemented in the oVirt (http://ovirt.org) scheduler > [1] in case there are any ideas there that could be leveraged? I believe > the determination there is also made by examining pCPU load (albeit with > some additional computations applied) though it does not use powernap. > Regarding oVirt, thanks for the pointer. We'll take a look on how it implements this feature. > > > Regarding the opportunistic aspect of the project, we're planning on > using > > powernap (http://manpages.ubuntu.com/manpages/lucid/man8/powernap.8.html > ) > > to tell if the host is idle or not. > > > > Finally, my question is: what do you guys think would be the best way > for a > > powernap action script to interact with nova-compute in order to stop all > > running instances and disable the host? > > If I understand correctly your suggestion here is that the powernap action > script would run when powernapd on the host has determined it to be at low > CPU usage at which point it would stop the instances and disable/poweroff > the host, rather than having one of the other nova components monitor the > load and initiate the required action? > Actually we want nova-compute to run opportunistically, in the sense that it should *only* run when the host is idle. Btw, by reading my previous message again, I realized it doesn't seem very clear. > > > Would killing and respawning the nova-compute be an elegant way of doing > > that in your opinion? Or do RPC calls sound better? > > Is the desire necessarily to *stop* all instances, or evacuate/migrate > them to another host so that you can consolidate instances on a subset of > your compute hosts and either power the remainder down and/or disable the > compute service on them? Another consideration here might be how much > available pCPU you want to keep as a buffer in case of a burst in instance > activity (probably something to make configurable?)? > Instances' migration is definitely a desirable feature, but we are not putting much effort on this for now. When we notice that the host is being actively used by a local user, we want to simply kill all running instances. > > Thanks, > > Steve > > [1] http://www.ovirt.org/Features/oVirtSchedulerAPI#Balance >
_______________________________________________ Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack Post to : [email protected] Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
