A new type of a decentralized clustering problem for networks is studied in this paper. The so called Decentralized Packet Clustering (DPC) problem is to find for a set of packets that are send around in a network a clustering where the clustering has to be done by the routers without using neither much computational power nor a large amount of memory. Further, no direct information transfer between the routers is allowed. We investigate the behavior of a type of decentralized k-means algorithm — called DPClust — for the DPC problem. DPClust has also some similarities with ant based clustering algorithms. We investigate the clustering behavior DPClust for different cluster problems and for networks that consist of several subnetworks so that there is only a limited amount of packet exchange between the subnetworks. A dynamic situation where the packet exchange rates varies over time is also considered. The proposed DPC problem leads to further interesting research problems for network clustering.