The mobile ecosystem is currently facing tremendous changes; in particular, the actual deployment of 4G will probably not be sufficient to totally fit the upcoming mobile data traffic explosion. This mismatch may be explained by the low marginal revenues expectable from mobile data compared to huge investments it requires. Consequently, congestion may appear soon in radio access networks, thus degrading the customer experience. QoS mechanisms are then required to preserve the most sensitive and/or valuable flows. In this respect, the 3GPP proposes a model capable of supporting several levels of QoS. However, this model was inherited from connection-oriented legacy networks and raises issues in terms of scalability, efficiency, performances and flexibility when used in the context of the foreseen mobile Internet. In this paper, we present an IP-centric approach which aims at adapting usual internet QoS management schemes (e.g. DiffServ) to wireless networks. This approach leverages most of the 3GPP multi-bearer model limitations while offering low-cost open mechanisms in line with usual web-oriented services. This IP centric architecture is described together with its strengths, issues and possible implementations in the radio scheduler of base stations. Simulations results are then presented in order to evaluate the efficiency of such IP aware scheduling.