The convergence of computer and physical world calls for next generation Wide Area Network (WAN) infras- tructures for hard real-time and embedded applications. Such networks need virtual topologies to achieve scalabil- ity, configurability, and flexibility. Virtual topologies are made of virtual links, for which, the state-of-the-art build- ing tool is Guaranteed Rate server based aggregates (GR- aggregates). However, common-practice weight assign- ment scheme couples GR-aggregate End-to-End (E2E) de- lay bound with aggregate's data throughput inverse propor- tionally. This is undesirable for many hard real-time embed- ded sensing/actuating applications, whose traffic has small data throughput but requires short E2E delay. We propose Guaranteed Delay server based aggregates (GD-aggregates), which allow assigning weights according to priorities instead of data throughput. This decouples E2E delay guarantee from data throughput, hence meets the needs of hard real- time embedded applications. In addition, GD-aggregates can be analyzed with simple closed form formulae, and can be easily planned with optimization tools.