An exploration of mechanisms and policies for gateways in real-time embedded systems

It is becoming increasingly common to connect traditional embedded system networks to the Internet for remote monitoring, high-level control, and integration. One architectural approach to building internetworked systems is to add a gateway between the embedded system and the external network. These gateways must transfer data between two heterogeneous networks without inducing application failures due to variations in timing or bandwidth between the two networks. Despite the importance of gateways, there is no clear recipe for designing them. To study gateway design, we examine gateway mechanisms that can be used to handle data passing through a gateway and describe policies for configuring these mechanisms. In this work, we examine the differences between generic mechanisms (i.e. queues) and application-aware mechanisms that use knowledge of the data being transferred. Using simulation with abstract network models, we compare the performance of these mechanisms and show that application-aware mechanisms can be useful in improving gateway performance in some situations. We also use a case study of a traffic control application to evaluate the performance of gateway mechanisms with simulations that model different network and environmental scenarios. We find that selection of the proper gateway mechanism can improve performance of the traffic control application, and we provide selection guidance based on the mean inter-arrival time of the network. These results show that in most scenarios, application-aware filter mechanisms outperform generic queue mechanisms