We consider a model where the interfering links employ on-off modulation in each transmission slot. In the on (active) state, a link obtains a data rate determined by the interference from other active links in the network. Based on this model, we compare the throughput regions of centralized scheduling and a probabilistic random access scheme, wherein in each slot, a link is active with a fixed probability chosen independent of other interfering links. We observe that for the case of two interfering links, the probabilistic scheme does not suffer any loss in the rate region relative to the centralized scheme if the interference between the links is sufficiently low. For more than two interfering links, the characterization of throughput rate region for the probabilistic scheme becomes intractable and similar observations are not easily forthcoming. However, we give a distributed algorithm where each link independently updates its transmission probability based on its measured throughput to achieve any desired feasible rate vector in the throughput region of the probabilistic scheme and prove its convergence.