Lookahead algorithm for node placement optimization in ShuffleNets

Node placement optimization in ShuffleNets is a combinatorial optimization problem. In this paper, a new heuristic node placement algorithm, called Lo okahead Algorithm, is proposed. Its performance is compared with the lower bounds derived in [I], as well as some existing algorithms in the literatur e. Significant reduction in weighted mean h(d) distance h(d) is obtained, e specially when the traffic distribution in ShuffleNets is highly skewed. Co nsider a ShuffleNet with 8 nodes, the hd obtained using Lookahead Algorithm is only 1.90% above the lower bound under the skewed traffic distribution (with traffic skew factor gamma = 100), and 16.04% under uniform random tra ffic distribution.