Since Redis is memory backed the idea was to transfer rarely accessed data on disk, to reload swapped data when needed (that is when a client will try to access it). The actual implementation of Redis Virtual Memory is completely done in user space: we try to approximate an LRU algorithm, encode data that should be swapped, write it on disk, and reload if needed, decode, managing pages in the swap file, and so forth. It's a non trivial piece of code but it is working well. - Tracy