In this part we look at the round trip (wall clock) time distribution for a request-response style message sequence exchange between two local Akka actors –PingActor and PongActor. In the course of the performance analysis we uncovere an unexpected performance behavior, which should be given due consideration in the application and configuration of the Akka runtime in certain application environments with particular workload patterns and latency requirements. Whether the performance observations are applicable in a real world context is something that can’t be so easily ascertained without further measurements under actual workload conditions. At this stage in our investigation it would appear that the out-of-box configuration of actors and their mailboxes might not always give the optimal results in throughput or CPU consumption. Like most software engineering endeavors there are trade-offs to be made, which need to be understood and reasoned on in light of expected or actual workloads, resource constraints, and performance objectives.

http://www.jinspired.com/site/performance-instrumentation-monitoring-of-an-efficient-runtime-akka-part-2