We study a wireless network under the 802.11 random access protocol, supporting multiple physical layer rates. Based on models for the effective packet rates achieved at the MAC layer, in both the downlink and uplink situations, we study the connection level dynamics. In particular, we study the Markov chain model that results from Poisson traffic being served by these MAC layer rates. For the downlink case, we show the equivalence of the model with a discriminatory processor sharing (DPS) queue, from which we derive stability conditions and study performance through the resulting connection level throughput. We also propose a variation of the flow scheduling policy at the access point that can result in a fairer allocation. For the uplink case, we derive stability conditions by identifying the asymptotic behavior of the chain with another DPS queue, and study its performance numerically. We validate the theory and cover other aspects neglected from the models, through ns2 packet si...