Simulation and sampling are both important and
widely used techniques in the research of networks. Scholars
have already developed a variety of packet sampling techniques.
However, few of these are focused on network reliability,
especially in the simulation situation. In this paper, the necessity
of using packet sampling techniques when evaluating network
reliability in the simulation environment is proved. Several
experiments are designed to discuss the applicability of different
statistic sampling techniques, including simple random sampling,
periodic sampling and equidistant sampling. Orthogonal
experimental design is used to reduce the experiment time. The
experimental result proves that the simple random sampling is
the best of the three. The larger the sampling probability is, the
better the applicability of the simple random sampling is. When
the sampling probability increases to a threshold, the
applicability becomes stable