Analyzing the Performance of Graph Neural Networks with Pipe Parallelism