those properties allow you to divide up work on data in a way that guarantees that no two processes will work on the same data at the same time. You can then ta

those properties allow you to divide up work on data in a way that guarantees that no two processes will work on the same data at the same time. You can then take data, divide it up, use 'const correctness' in the language to guarantee original data doesn't change and then you don't need a mutex because you know only one thread is working on that piece of data. A typical approach is if you can divide your algo into map/reduce, you can do this. Going back to the properties of you computation allows you to reason how to apply something like map/reduce.