Abstract NETplace is an efficient algorithm to assign virtual nodes to physical nodes on the network emulation testbed, while reducing the experiment runtime for network emulation up to 64%. As an assumption of this algorithm, a detailed defined cost model for communication cost has been provided. This cost model needs expected data rates of the links between each pair of virtual nodes as well as CPU load (in CPU cycles) on the virtual nodes, which are the experimental data produced by SoftwareunderTest (SuT). Therefore, the goal of the thesis is to define a generic load model to efficiently provide placement algorithm with a realistic estimation of experimental load data. In order to reach this goal, several problems should be solved. First, there are thousands of network links and virtual nodes in the model, so it is impossible to manually inquire and input all experimental CPU load and data rates into the model, because it takes too much time. A possible approach to resolve this problem is to divide the nodes into several groups, in which all the nodes have nearly similar characteristics. That is to say, a node classification is made. Thus, we only need specify one node for each group, and the other nodes can be automatically assigned according to the node classification. Second, nowadays the network is already very large, so it still costs much time, if we classify the nodes in the whole network together. A better solution is to analyze the nodes in part networks with the help of network clustering. So generally speaking, through combining the network graph clustering with the nodes classification we can provide the load information, which NETplace needs.