What I ended up doing was having a set number of machines shut themselves down in response to certain power events. It's not the slickest of solutions, but it works.
> -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of > Andrew Beekhof > Sent: Tuesday, October 28, 2008 8:19 AM > To: General Linux-HA mailing list > Subject: Re: [Linux-HA] Energy Efficiency > > On Tue, Sep 16, 2008 at 00:23, Todd, Conor <[EMAIL PROTECTED]> wrote: > > Hi list! > > > > I haven't seen anyone talk about the energy-efficiency of > high-availability clusters, so I let me bring up my thoughts > on it and see if we find this to be a subject of interest. > > > > Linux has decent power management via APM and ACPI. > Therefore, each Heartbeat node can be expected to manage its > power usage efficiently, given the service load the the > administrator (including the CRM) has decided to run on the > node. What is lacking here is a power management scheme that > looks at load on the cluster overall, and, in the event of a > power loss, consolidates resources onto as few nodes as > possible so that nodes can be throttled back or even > powered-down to conserve energy (in this case, generator fuel > and battery life). > > > > I happen to work from Houston, TX, where we just got hit > with a fairly large hurricane. Our labs were shut down > before the event happened because we expected a complete loss > of mains power. I was playing with the idea of leaving my > team's services cluster running (it's running HB 2.1.4), but > decided against it because I know of no way to have the CRM > consolidate services and shut down un-used nodes. > > You could manually set nodes into standby mode - that would > at least let the machine be mostly idle. > But right now, we're simply not smart enough to handle this > and have no way to power the nodes up again when we need them. > > > Does anyone else think that this would be useful? > > Yep. > > > > > What I envision is a reduction in cluster size after having > been in a power-threatened state for a certain amount of > time. This would be different from fencing, in that the > quorum size would change once the nodes to be shut down have > been verified as offline. This would allow the cluster to > continue operating with the redundancy and quorum logic as > expected, but it would simply be a smaller cluster. > > Its a neat idea - not sure when we'd be able to implement it though. > It would be a reasonable invasive change I think. > > > > > When power is restored and after a certain amount of time > (or perhaps batter condition) has passed, the nodes which > were brought down in order to conserve power would be brought > back online. > > > > Yes/no/that's silly? > > > > - Conor > > _______________________________________________ > > Linux-HA mailing list > > [email protected] > > http://lists.linux-ha.org/mailman/listinfo/linux-ha > > See also: http://linux-ha.org/ReportingProblems > > > _______________________________________________ > Linux-HA mailing list > [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha > See also: http://linux-ha.org/ReportingProblems > _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
