Reducing communication overhead by scheduling TCP transfers on mobile devices using wireless network performance maps

摘要

The performance of wireless communication networks has been shown to have a strong location dependence. Measuring the performance while having accurate location information available makes it possible to generate performance maps. In this paper we propose a framework for the generation and use of such performance maps. We demonstrate how the framework can be used to reduce the retransmissions and to better utilise network resources when performing TCP-based file downloads in vehicular M2M communication scenarios. The approach works on top of a standard TCP stack hence has to map identified transmission intervals to predicted TCP data volumes. The file download optimisation is evaluated using extensive simulations comparing the TCP scheduling approach to a normal transfer. The performance map is generated by capturing round-trip time measurements and a threshold approach for the mean value for a given area is used to decide if data should be transmitted or not. A simple TCP model and a more advanced model are compared in their ability to accurately forecast how much data to transmit in the selected regions. The results show that using an intelligent TCP data transfer scheduling there is a significant potential for improvement in reducing the communication overhead compared with an ordinary TCP transfer.