Data that travels

A separate kind, or for the lack of a better word, family of distributed systems includes data that travels from a node to node instead of being stored in one place. The range here covers anything from event processing systems (including workflow systems) to map-reduce clusters. Current state of data travels from step to step, being enriched or otherwise transformed by every processing step it goes through. It's really difficult to run queries or even know precisely how much data you have in your system at any given time. It's also difficult to predict how much time a given piece of data spends inside. Usually you collect statistics on the system behavior and that allows you to make fairly good guesses. There also some special tricks that help you run queries and get better estimates. People use these systems when there is more interest in processing results than data storage.
Consistency
Any kind of data storage and processing system can be eventually or immediately consistent. I use "Immediately consistent" to describe read-after-write consistency by any thread (the thread that wrote the data and any other). "Eventually consistent" means that the thread, which wrote the data can get a previous version of the data on the subsequent read. There is a really useful behavior where only the thread (client etc.) which wrote the data is guaranteed to read a new version immediately afterward, while any other thread (client) can read old or new version - however its luck works out.