Abstract: Cascading failures always occur in computer networks in which network traffic is severely impaired or halted to or between larger sections of the network, caused by failing or disconnected hardware or software. “Load-capacity” models are usually used for solving network traffic problems and exploring the mechanisms of cascading failures. Centering on cascading failures in complex networks, the following work is done. Firstly, the effect of network traffic load on cascading failures is analyzed. It indicates that the communication activity among network nodes presents a self-organized criticality phenomenon, and the influence on the network robustness brought by the traffic change of those original inactive nodes is much greater than that brought by those active ones. Secondly, under the constraints of economy and technology, a cost factor is introduced to model the relationship of network capacity and load, so as to reveal their constraints relationship. Finally, Centering on the issue of “how to allocate the limited redundant resources to a network with specific structure in order to improve its robustness”, an evolutionary algorithm to search an optimized capacity-allocation strategy is proposed, which makes the network achieve optimal robustness with the same resources. Experiments show that our algorithm can find a better result of capacity-allocation than the ones with linear or preferences-attached strategies.