oVirt / RHEV Power Management?

[TeeJayHoward]

I just finished swapping over from vSphere to oVirt. One of the features I loved about vSphere is that vCenter will migrate VMs around and automatically power off hosts until the minimum number of hosts are powered on for the VMs you've got running. If more resources are needed than the cluster has available, it'll power a host back up. The whole thing is automated, and worked wonderfully. Is there a way to do this in oVirt / RHEV?

 

[EffrafaxOfWug]

RHEV's lack of this (as well as just general crapitude of its live migration in general), at least at the time, was our primary reason for moving away from it to VMware, (although we did it for load distribution reasons rather than the "power off hosts" thing).

Details on RHEVs equivalent can be seen here (you probably want the power saving policy) but IME it never worked well
Chapter 6. Load Balancing, Scheduling, and Migration

oVirt-specific settings on the load balancing policy on the same thing should be here although I've never used it myself;
Global Configuration — oVirt

 

[TeeJayHoward]

Details on RHEVs equivalent can be seen here (you probably want the power saving policy) but IME it never worked well
Chapter 6. Load Balancing, Scheduling, and Migration

When an under-utilized host is cleared of its remaining virtual machines, it can be shut down by an administrator to preserve power.

Damn, so it's just something RHEV doesn't offer

 

[EffrafaxOfWug]

I daresay you could bodge something up with some scripting, but TTBOMK it doesn't offer any automated way to do it like VMware does. KVM works fine for my home lab experimentation but I'd never recommend it over VMware for a serious production environment because of the immaturity of stuff like this.

 

[TeeJayHoward]

I daresay you could bodge something up with some scripting, but TTBOMK it doesn't offer any automated way to do it like VMware does. KVM works fine for my home lab experimentation but I'd never recommend it over VMware for a serious production environment because of the immaturity of stuff like this.

One of the reasons I'm switching my home lab is because my workplace is spinning up on virtualization, and they've chosen OpenStack/KVM/etc as their platform of choice. I was very vocal in my desire for VMware, but I was a small voice in a very big crowd and got drowned out. Needless to say, I need to get good at Red Hat's solution. Work is hardcore on "zero impact to customers", so any effort I can put in now would pay dividends.

For work, things like power management aren't that important. We keep everything spun up all the time "just in case we need it." For a home lab, though... I'd much rather pull down a continuous 8A instead of 10A. Electricity might be relatively cheap, but that's hardware upgrade money I'm dumping into the bill every month.

 

[EffrafaxOfWug]

Depends a great deal on your overall capacity, type of work, predictability of workload and the overall makeup of your IT department I guess - VMware's expensive, but for us at least it meant less maintenance overhead, better overall performance, greater flexibility and (probably most important for the beancounters) cheaper staff to keep it ticking over. But if you've gone openstack I'd guess you're running a very different environment to the sort that I have been used to - at a guess I'd say you're a hosting/services provider of some sort running mostly linux VMs?

I'd assume for an architectural shift of this magnitude you must have had a good few PoCs running in parallel to evaluate all of your regular use-case scenarios anyway...?

For power bills and a home lab, I'd only ever run a home cluster for exactly as long as it took my how to learn the cluster management and the foibles thereof, and then consolidate as much as possible on to singular non-clustered hosts - KISS. As much as you'll learn a lot by breaking it and learning how to fix it, IME the stuff at work always breaks in exactly the same way as the home stuff doesn't