High level collective operations, i.e. function/method calls executed in parallel on many cores, are easily applicable mechanisms to coordinate parallel activities. On top of architectures with ultra low latency networks like the SCC it is possible to coordinate groups of cores within a few microseconds only. Efficient consistency control protocols can be implemented in software (cache coherence, DSM etc.) on a high level of abstraction with modest programming effort. TACO (Topologies and Collections) is a distributed object platform that provides various template based RMI mechanisms and efficient collective operations for distributed object groups.

This talks discusses the first implementation of TACO on the SCC and shows how collective operations can be applied effectively to implement efficient consistency control protocols.

Joerg Nolte is Vice Dean and Professor of Computer Science of the BTU Cottbus, Germany. He previously held research positions at the Fraunhofer Gesellschaft FIRST (Berlin, Germany) from 1988-2003 and the Parallel and Distributed Performance Lab (RWCP, Tsukuba, Japan) from 1994-1996 and 1999-2001. He received his M.S. and Ph.D. from TU Berlin, Germany, in Computer Science in 1988 and 1994, respectively. He has published papers in the areas of embedded systems and parallel/distributed systems.