[Linux-HA] Should We Upgrade from 2.0.4?

[Andrew Beekhof]

On 1/3/07, Robinson, Eric <eric.robinson at pmcipa.com> wrote:
> >worth mentioning however, is the performance improvements in .8 (when
> it comes out).
>
> Right now, my active database server (dual-core, dual processor Xeon)
> runs at <2% CPU utilization on average. I don't expect the workload to
> really tax it for another 1-2 years at our current rate of growth. By
> then we'll be ready to purchase a 16-core monster. :-)
>
> The only thing that does not work properly is STONITH, but I can live
> with that because we use DRBD with a shared-nothing architecture, so
> STONITH failure is not a real show stopper. (Works fine from a shell
> prompt, but when I use it from heartbeat, I get errors. No luck figuring
> out why.)
>
> My active LVS load-balancer (single-core Celeron) runs at about 2-5% CPU
> on average.
>
> Bottom line is that I probably would not benefit from the extra
> heartbeat performance right now. Thanks for the feedback, though.

its more an issue for larger clusters where one node failing could
_potentially_ create enough work on the remaining nodes that resources
appear to start failing (due to CPU starvation) there also.

no-one's reported seeing anything like this in the wild, its just
something that I've observed when pounding on a 8 node cluster (with
50-ish resources) for hours at a time.

we like to be pre-emptive about this sort of thing :-)