Parallel Computation of the Singular Value Decomposition on Tree Architectures

We describe three new Jacobi orderings for parallel computation of SVD problems on tree architectures. The rst ordering uses the high bandwidth of a perfect binary fat-tree to minimise global interprocessor communication costs. The second is a new ring ordering which may be implemented e ciently on an ordinary binary tree. By combining these two orderings, an e cient new ordering, well suited for implementation on the Connection Machine CM5, is obtained.

• Type: Conference
• Year: 1993
• Where: ICPP
• Authors: Bing Bing Zhou, Richard P. Brent
• Keywords: Distributed And Parallel Computing, ICPP 1993, Interprocessor Communication Costs, Ordinary Binary Tree, Perfect Binary Fat-tree