Percolator

Big things happening. Google described how they implemented an ACID-Compliant distributed transaction system on top of BigTable. Some people out there claiming Map/Reduce is dead just because Google is not using batch processing anymore for indexing. The fact is that Percolator is not a general-purpose system, i.e. it can't do joins.

It kind of recovers the old idea of triggers - just that now triggers can also scale horizontally. Plus these triggers are written in a 21th century language, not in PL/SQL - BigTable and Percolator communicate through RPCs.

I don't think Map/Reduce is dead. It will still be an elegant programming model suitable for most data processes.

However speaking about real-time, huge data sets: things are changing.