Synchrophasors are sensors that sample power grids and publish these measurements over a network to a number of grid applications such as voltage monitoring, state estimation, visualization, etc. The sampled data is QoS sensitive and must be delivered reliably with minimal delays to the target applications. However, during network overloads or grid emergencies when the volume of data transmitted is high, it is important to gracefully degrade performance and data stream delivery in an applicationspecific manner. We propose CPS-Net, a flexible 3-layered network architecture that allows application-specified in-network aggregation of synchrophasor data streams during overload. The lowest layer provides basic path-specific QoS while the middle layer provides real-time wide-area publish-subscribe capabilities integrated with traffic engineering of data streams across multiple lower level paths and trees. The top layer provides a distributed stream processing infrastructure for application-specified aggregation functions. During network overload, the lower layer triggers the co-optimization of higher layers and application-specific aggregation of data is performed. The user is presented with a simple stream processing programming model and the details of the network, placement and composition of operators are abstracted away. Initial simulation results, using a voltage stability monitoring smart grid application, show that CPS-Net architecture can gracefully degrade data streams for synchrophasor applications. © 2011 IEEE.