On a few occasions we have seen bottlenecks in SSL code that were extremely difficult to diagnose. Can we add some instrumentation to make them visible? A simple example might be to track cumulative time spent in calls to ERR_get_state which have been a cause of bottlenecks, but there may be other places higher in the stack that would be better places to watch for unexpected bottlenecks. This would be comparable to the way we monitor other possible bottlenecks like locks, tickets, some key storage bottlenecks, etc.