Speedup of branch and bound method for hardware/software partitioning

Abstract

This paper presents sensitivity analysis of branch and bound (B&B) method used for hardware/software partitioning task. The impact of all B&B parameters on computation time is theoretically analyzed and results of experiments are presented. Results show that most sensitive parameters are a lower bound function, a selection rule, a branching rule and an initial solution. To shorten B&B computation time these parameters have to be set properly and additional pre-optimization step should be applied. This pre-optimization step uses simulated annealing to set parameters in limited time. Results of experiments show that the computation time speedup x130 is achieved on average. This hybrid optimization is the most efficient presented so far.