Design of highly reliable tree architectures

Abstract

It is well established that tree interconnections provide the most natural interconnect architecture for hierarchically organized distributed computing systems. One advantage of the tree architecture is the 0(log N) speed of information exchange between any two nodes of an N node system. Another advantage is that the tree architecture can naturally map several important classes of problems that can be described as divide-and-conquer algorithms. In this paper we address the issue of fault-tolerance in binary tree architectures and present two new reliable tree architectures: the L-tree and the LN-tree. The L-tree is formed by augmenting the simplex binary tree with redundant links; the LN-tree is formed by augmenting the simplex binary tree with redundant nodes as well as links. Comparing the fault-tolerance performance of these two tree structures with other augmented tree structures previously proposed, we find that the L-tree is more reliable than existing fault-tolerant tree structures and has the further advantage of permitting simple algorithmic routing. The L-tree tolerates faults with a degradation in performance, however, whereas the LN-tree is not only highly reliable, but also maintains a rigid tree structure in the event of faults.