Performance Evaluation of TCP in an Integrated WPAN and WLAN Environment

Short-range low power radio frequency systems such as Bluetooth and UWB enable the deployment of wireless personal area networks (WPAN). A WPAN can interface to larger networks to provide broader network access and Internet connectivity. We evaluate the performance of TCP over an integrated WLAN and WPAN system using a real network testbed. The end-to-end throughput is found to increase, when the window size is increased. However, for large window sizes, measurement results revealed that a wide range of round trip times (RTTs) is experienced. With small window sizes, the variability in the RTT is smaller. The measurements showed that in an integrated network long RTT delays and frequent duplicate acknowledgements lead to an increase in the number of packets transmitted unnecessarily. We study the effect of the number of active short-range devices (Bluetooth in this case) in a piconet. The results show that as the number of active slaves present in the WPAN increases, the bandwidth received by slaves exchanging data decreases. This behaviour is attributed to the Bluetooth scheduling mechanism which uses the round robin polling method. The main finding of our experiments is that for Bluetooth to be successful in enabling WPAN, the time slot allocation scheme should be efficient and fair. Finally, the issue of fair bandwidth allocation among multiple TCP streams is also investigated. The results show that for small number of TCP connections, the bandwidth is fairly distributed. However, as the number of simultaneous TCP connections increases, bandwidth distribution seems to become somewhat less fair.