In order to reduce system outages and the associated downtime cost caused by the "software aging" phenomenon, we propose to use software rejuvenation as a proactive system maintenance technique deployed in a CMTS (Cable Modem Termination System) cluster system. Different rejuvenation policies are studied from the perspective of cost and availability. To evaluate these policies, stochastic reward net models are developed and solved by SPNP (Stochastic Petri Net Package). Numerical results show that significant improvement in capacity-oriented availability and decrease in downtime cost can be achieved. The optimization of the rejuvenation interval in the time-based approach and the effect of the prediction coverage in the measurement-based approach are also studied in this paper.