We don't need to re-invent MapReduce or message passing. There are ready made solutions/frameworks like Hadoop and Storm that are fairly easy to implement. There are also a number of databases that are designed for distributed computing. You don't need to be Google to have a fault-tolerant distributed computing setup. It's really a matter of choosing the appropriate technologies, rather than re-inventing the wheel. Because of that, I'd consider it a "solved" problem, even if a particular implementation might need some thought to match up with appropriate technologies. It's not necessarily trivial, but it's probably as "solved" as it needs to be.