A study of TCP performance in wireless environment using fixed-point approximation

Loss of IP packets occurring as a result of imperfect local error correction is one of the major reasons for TCP performance degradation in wireless networks. These losses are misinterpreted by TCP senders as congestion indications and subsequently lead to the sharp decrease of the sending rate. In addition to wireless losses, packets can be lost as a result of buffer overflows at the IP layer. To study performance experienced by TCP sessions sharing a wireless channel, we adopt the fixed-point approximation. We distinguish between two modes of operation: (i) packet losses are mainly caused by imperfect error correction, (ii) the packet loss process is dominated by the buffer overflow. Applying different TCP models for these two regimes we approximate TCP throughput as a function of underlying layers’ parameters. Using this approach we study the effect of various protocol parameters on TCP throughput. We also investigate the effect of the queuing system involved in our analysis and demonstrate that usage of complex models with correlated arrival processes and generally distributed service time does not qualitatively affect the estimated TCP throughput. This observation allows to use simple queuing models having closed form solutions for performance metrics of interest, e.g. M/M/1/K or Geo/Geo/1/K. Finally, we address the question of choosing appropriate queuing model for quantitative TCP analysis.