We find that divide-and-conquer is arguably another form of tuning. This phenomenon is unique to the big data analysis. In other words, it is unnecessary to choose an optimal smoothing parameter. Rather, this can be compensated by a proper choice of sample partitioning.