| In this paper the network architecture under consideration is an IPSEC networking, in which an encryption device is located at the boundary between the in-secure LAN and the secure mobile ad hoc WAN. This architecture has been used in many military networks and commercial networks. Since there is virtually no information allowed to be sent across the encryption device, a lot of the vital information on the WAN side, like topology and link bandwidth data, can not be made known to the LAN side, where end-users are located. This makes providing end-to-end QoS support in this type of network very challenging: even though it is assumed that Differentiated Services (DiffServ, as described in RFC 2475) is implemented in the routers, due to the dynamic nature of the WAN, DiffServ by itself is not enough to provide satisfactory end-to-end QoS support.
To address this, a study was launched under the ARMY CERDEC Multi-functional On-the-move Secure Adaptive Integrated Communications (MOSAIC) program to investigate the effectiveness of using only real-time observations to provide end-to-end QoS support in this type of secure mobile ad hoc network, without violating any potential security requirement. In this paper, the findings of our investigation are reported.
The core of our solution is a light-weight call admission control and preemption control module located at the LAN side to provide end-to-end QoS assurance. The salient feature of our solution is that call admission control and preemption control decision is based on the real-time measurements on the traffic loading. The traffic loading data enables the admission and preemption algorithm to quickly react to the status change of the network caused by radio dynamics and traffic congestion in the WAN. As such, together with DiffServ, our traffic loading based admission and preemption algorithm ensures the high-priority mission-critical applications are protected.
We simulated the QoS architecture using OPNET to evaluate the performance characteristics. We studied the performance over five different priority classes (as suggested by the DoD GIG QoS/CoS Working Group) for six different applications. The performance metrics under consideration include average delay and jitter per priority class for UDP applications, throughput and file transfer completion time for TCP applications, and per class preemption/blocking probability. Our study showed that this integrated solution exhibited superb end |