Fig. 8

Comparison of delay and data throughput between OEGA partition method and other partition methods. The upper subgraph is the running time of the neural network under the three partition methods, and the subgraph below is the corresponding data throughput. By comparing the left and right subgraphs, we can see that the runtime is basically the same as the change of data throughput. The larger the throughput, the longer the runtime. In addition, by comparing the three partitioning methods, we can see that the proposed partitioning method can obtain a better partition solution. Evenly partition presents a randomness, i.e., the inability to control data throughput. Binary partition method can get better partition solution than evenly partition method, but overall is not as good as OEGA partition method because it is easy to fall into local optimal solution, while OEGA partition method is easier to jump out of local optimal solution