This work presents a resource allocation algorithm in K-user, M-subcarrier and NT-antenna systems for on-line scheduling. To exploit temporal diversity and to reduce complexity, the ergodic sum rate is maximized instead of the instantaneous one. Dual optimization is applied to further diminish complexity together with a stochastic approximation, which is more suitable for online algorithms. Weighted sum rate is considered so that users can be either prioritized by higher layers or differentiated by proportional rate constraints. The performance and complexity of this algorithm is compared with well-known benchmarks and also evaluated under real system conditions for the MIMO Broadcast channel.