When under high load (many queries in parallel) our Oracle RAC gets in a state where most of the activity is reported as "cluster" by Enterprise Manager. It stays like that for 30 minutes to a couple of hours, and then just recovers by itself and happily proceeds.
Sometimes instead of "cluster" the activity is reported as "configuration" or "concurrency".
I'm looking for any advice on how to go about diagnosing what is going on (and how to make it stop doing it). The graph shows only one node in question. Queries are run on a service which is limited to that node.