TACO (Topologies and Collections) is a template library that introduces the flavour of distributed data parallel processing by means of reusable topology classes and C++ s. This paper introduces TACO's basic abstractions and provides a performance analysis for basic collective operations on various cluster architectures with several different networks.