Even multiple threads - any sort of concurrent access/modification on the same dataset becomes problematic if you just use a HashMap.

Sites that have any kind of download/upload component, even though does not have too much traffic, will need to serve multiple requests using threads/processes.

That is where we normally use relational database. However Redis presents a way to be faster than conventional databases by storing everything in memory. This is as close to a concurrency-safe HashMap that you can get with the least amount of effort.