> > But we use a static logging API anyway, and this works adequately well. I > have debugged stuff largely through logs in multi-node simulators, I > generally rely on the thread name to identify the node (e.g. the UDP receive > and send thread include the port number in the thread name). >
It could become an issue when trying to reproduce rare concurrency bugs that require running tests in a loop until they fail. The smaller the synchronization overhead of the logging subsystem, the less noise you introduce in the simulation as you increase parallelism. It also runs faster and you reproduce the bug sooner.
