We consider a packet switched wireless network in which a base station serves the mobiles. The packets for the mobiles arrive at the base station and have to be scheduled for transmission to the mobiles. The capacity of the channel from the base station to the mobiles is varying with time due to fading. We assume the mobiles can obtain different types of service based on the prices they are willing to pay. The objective of the base station is to schedule packets for transmission to mobiles to maximize the revenue earned. Our main result is that a simple greedy algorithm does at least 1/2 as good as the optimal offline algorithm that knows the complete future request pattern and channel conditions. We also show that no other online algorithm can do better.