TACO-Exploiting Cluster Networks for High-Level Collective Operations

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.

• Type: Conference
• Year: 2001
• Where: CCGRID
• Authors: Jörg Nolte, Mitsuhisa Sato, Yutaka Ishikawa
• Keywords: Basic Collective Operations, CCGRID 2001, Distributed And Parallel Computing, Reusable Topology Classes, TACO's Basic Abstractions