by Christian Hergert
Traditional problems in Mutli-Processor programming

As we reached the speed-limit of serialized CPU instructions we started to add more cores. Obviously, adding more cores means we have to change how we process information. Instead of optimizing for the fastest serial processing of data, we look for ways to split the work into multiple tasks. That way, we can put tasks on different cores and efficiently use the resources available.

So now we have lots of individual tasks. It would be most excellent if they could communicate. Often times we end up using a Mutex or other synchronization primitive to lock access to some shared state. Unfortunately, this doesn’t scale well as you add more concurrent workers. This ends up creating more lock-contention and reduces the overall usefulness of splitting the problem into tasks. What if we could communicate to manage shared state without introducing locks? That is precisely what some very smart folks have been designing for years. Now the rest of us are catching up.