Investigate performance overhead of ingress/egress actors

[dplyukhin]

An Akka cluster consists of ActorSystems. To support fault recovery, each ActorSystem records statistics about outgoing messages via the "egress stage" and statistics about incoming messages via the "ingress stage". I'd like to know how much overhead these stages impose on message-passing throughput, and whether these stages can be implemented more efficiently.

This project involves:

1. Writing a benchmark to measure distributed message-passing throughput.
2. Profiling.
3. Optimizing data-intensive (but sequential) code paths.

First steps: Implement an Akka Cluster application in which an actor on System 1 sends messages to System 2 at a fixed rate.