Next: Hardware test environment Up: 802.11 Wireless TCP and Previous: SCTP Contents
The intention of the experiment was to produce a repeatable, real-world test of the impact of wireless noise on TCP and SCTP throughput. Initially, the goal was to test out regular TCP and wireless-TCP extensions. However, most of the previous work was done in a simulator, and no patches to existing ``production'' TCP stacks are easily available. This has a significant drawback in that simulators do not encapsulate the complexity and interactions between an operating system TCP stack and lower level drivers. These interactions can and do dramatically influence performance and throughput, so much so that any theoretical benefit can be swamped by poor implementation and tuning issues . This has been particularly prevalent in the High Performance Computing field and the early adoption of 100Mbit and Gigabit Ethernet in those fields [BHB04]. In both cases TCP tuning and host CPU speeds limited full utilization of the available network bandwidth.
Fortunately, SCTP has received enough attention and has at least two open source implementations available, so it was comparison between TCP and SCTP was conducted. The tests consisted of measuring throughput on a 802.11g link while varying the amount of UDP traffic present on an adjacent conflicting channel.