1. Blocking computations block the thread they are using in case they need to wait for some event to happen. Non-blocking computations release the thread if they don't need it. Released threads can be reused by other tasks.
2. You need asynchronous programming so that the threads are utilized doing useful work and not waiting upon an event to happen.
3. You can separate the business logic from the strategy of concurrent execution.
4. IO is an effect type which encapsulates the side effect of delayed computations. It uses the computation-as-a-value approach, and thus is a specification of the computation to be performed.
5. start to start IO asynchronously from a monadic flow. flatMap to compose IO with other IOs sequentially. Please see API documentation of IO for the full list.