Abstract Weighted queue minimization is considered in Dynamic TDD system where the available resources per cell and per TDD frame can be freely allocated to both uplink (UL) and downlink (DL) transmission depending on traffic demand and user distribution within the network. Given the traffic state, joint UL/DL cell mode selection, beamformer design and resource allocation among available time/frequency/space dimensions is carried out. A centralized solution is found by dividing the combinatorial and non-convex problem into subproblems in which relaxed versions of the original intractable optimization problem are formulated and solved in an efficient manner. The numerical examples demonstrate significant gains from dynamic UL/DL mode selection as compared to the conventional TDD with fixed mode assignment.