Nowadays, home gateways such as Wireless Access Points(APs), Cable or DSL modems are widely deployed for residential and Small Office/Home Office (SOHO) customers to access Internet services. The home gateways typically act as middleboxes performing various higher-layer functions, such as network address translation (NAT), traffic filtering or advanced application layer operations. The exact behavior of these functions are not standardized, and they are often undocumented by the home gateway vendors. However, such middleboxes are known to have undesired interactions with normal protocol operation, and they substantially hinder and complicate new protocol development and prevent new protocol extensions from working. For this reason, it is important to learn the deployed middlebox characteristics to allow network engineers design protocols that can be deployed in realistic environments that typically include middleboxes.
In this thesis, the main purpose is to explore the characteristics and unknown behavior of various home gateways which are already widely deployed everywhere. In order to achieve this goal, numerous home gateways from different vendors are deployed into our home gateway test-bed and a set of well-designed software is developed to reveal the realistic behavior of these devices under various circumstances and conditions. Also, in order to simplify the experimental procedures, the home gateway test system automation is utilized to intelligently and systematically organize multiple measurements on all the devices in our testbed. Our experiments emphasize on the latest protocols such as Lightweight User Datagram Protocol (UDP-Lite), Data Congestion Control Protocol (DCCP), Stream Control Transmission and Datagram Protocol (UDP). Moreover, the maximum number of UDP bindings, UDP throughput performance and UDP broadcast experiments are included to comprehensively learn how the home gateway handles UDP traffic in practise.
The experimental results indicate that NAT functionality indeed prevents new protocols such as SCTP, DCCP and UDP-Lite from operating over current home gateway devices. In addition, home gateways significantly influence the UDP throughput performance for the residential users. Moreover, the experimental results indicates that the widely deployed home gateway devices perform a variety of NAT mapping, filtering and hairpinning behavior. Hence, it is extremely crucial to understand and benchmark the NAT behavior, while the existence of middleboxes should also not be ignored by protocol designers.
