Study and comparison of bisectional and hypercube networks for dynamic task reallocation

A class of interconnection network architectures, known as bisectional networks is studied for the purpose of dynamic task allocation in large-scale distributed systems consisting of hundreds or thousands of nodes. A bisectional network is constructed by using the theory of symmetric balanced incomplete block design. It is also isomorphic to folded hypercube in that a binary hypercube network can be easily extended as a bisection network by adding additional links. These additional links add to the network some rich topological properties such as node symmetry, small diameter, small inter-node distance, and partitionability. The important property of partitioning is exploited to investigate a redundant task allocation and a semi-distributed task redistribution strategy under real-time constraints. The same approach is used to partition hypercube systems as well. We study the performance of both networks and show that the bisectional network performs considerably better than the hypercube. The performance evaluation and comparison has been done through simulation by taking into account system load, task migration overhead, task timing constraints, node failure and repair rates, and by examining important measures such as task deadline missing probability, task response time, and probability of task being lost.<>