First, thanks to those who wrote me, both online and offline, about my last post on the distribution of cluster node counts. As a follow-up, here’s a look at cluster asymmetry – that is, how often a cluster’s nodes are not identical. This is a non-academic question, since we’re in the process of adding new servers to SORT’s cluster. It’s been a few years since our last hardware refresh, so our server model is no longer available, plus all newer models come with more CPU and memory than our current boxes.
So, I looked at how often we see asymmetric clusters among configurations customers have uploaded to SORT:
As shown above, a third of the clusters showed any kind of asymmetry. Memory was the most common. A careful review of the graph shows ‘Any’ is less than the sum of the other categories. In other words, of the asymmetric clusters, most had more than one type of asymmetry.
For a variety of reasons, this sample is quite a bit smaller than the previous post, so put your mental uncertainty bars a bit wider.
This is all somewhat reassuring for SORT’s cluster. However, I’d really like to split apart cluster file system clusters (i.e., those using Storage Foundation Cluster File System, Storage Foundation for Oracle RAC, and Storage Foundation for Sybase CE) from failover clusters (i.e., those using VCS, Storage Foundation HA, etc.).
After all, you can easily imagine a smaller passive node and a large active node in a failover cluster. It’s not optimal, but it is better than an outage. CFS does have some clever logic to adjust its behavior based on asymmetric memory on different cluster nodes, but it would still be nice to see that other users are running successfully in that configuration.
I’d also like to quantify the asymmetry beyond yes/no. Hopefully in the New Year --