TU Chemnitz, Fakultät für Informatik, Professur Rechnerarchitektur: Fakultät für Informatik: Professur Rechnerarchitektur
Analysing Abinit on Cell Clusters (AbiCell)
Project Description (Abstract)
The Cell BE is a new processor architecture. In order to obtain the maximum performance, it is necessary to adapt the code to match the underlying hardware architecture.
This means addressing issues of vectorization, memory alignment and communication between main memory and local stores. Additionally, in case of Cell clusters communication performance issues between interconnected nodes need to be considert too.
The goal of this project ist to study the performance potential of InfiniBand connected Cell clusters by implementing selected parallel compute kernels of Abinit, a widely used open source code for ab initio electronic structure calculations.
Thereby we rely on former studies of parallelization methods implemented by using the MPI library.
We envisage the MPI microtask programming model that can be considered as an SPEcentric computational accelleration model. Particularly we focus on Open MPI.
Nonetheless, the first step will be studies based on an function offload model in which the main application program runs on the PPE and performance-critical functions in the program are offloaded to SPEs.
The minimal hardware test platform will be a Cell-Blade equipped with InfiniBand adapters connected via an external InfiniBand switch. A more general version will consist of at least two Blades.
Studentische Hilfskräfte gesucht!
Bitte bei Prof. W. Rehm rehm@informatik.tu-chemnitz.de melden.
This work is part of a collaboration with the Center of Advanced Study IBM Böblingen, Development and Research, Germany.
