In the initial port, the concurrency might not be obvious. You need to carefully infer/reverse-engineer concurrency/simplification opportunities from the current code. For example, you might have coded a sequential list of functions that actually don't need to be executed in the given order, that is, they can be run concurrently. In other cases, you might have code (such as cleanup) that can be done after the function returns. This can be efficiently coded using the defer() keyword.

The tests that you wrote in Phase 3 will help you avoid regressions and thus allow you to experiment with multiple levels of concurrency.

Having seen a strategy for migration, let's look at the final (but actually, the most important) part of migrating applications—building an awesome team to get the job done.