This paper addresses the problem of joint power and bandwidth allocation for a general multiuser satellite downlink system. In contrast to existing studies, this work focuses on the heterogeneous services, including both real-time and non-real-time ones. Subject to the delay constraints of real-time services, the system throughput is maximized over multiple time slots, while a Weibull-based channel model is exploited for the rain attenuation prediction. Based on the dual decomposition method, the power and bandwidth allocation is derived with the relaxation of delay constraints. Then by introducing a delay-violation parameter, we propose an algorithm to approach the optimal resource allocation solution, while satisfying the delay constraints before relaxation. Simulation results show our algorithm effectively achieves a well-performed trade-off between throughput and delay performance.