||In recent years, data traffic has grown exponentially. In order to bypass the bottleneck of electronic processing speed, all-optical networks are proposed as an ideal solution for high-speed transmission. At the same time, more and more applications require some kind of Quality of Service (QoS) guarantee, which makes the current best-effort service model inadequate. It is a consensus that Internet architecture should be extended with some type of service diferentiation, so that the applications with QoS requirements receive better service than others while the network resources are efficiently used. Proportional Differentiated Service(PDS) architecture is proposed as a scalable and simple QoS service model. In addition, it is controllable and predictable, meaning the level of differentiation between service classes is adjustable by network operator and consistent independent of the load conditions. We first introduce the proportional QoS model into wavelength routing networks on call blocking probability differentiation. We design three wavelength assignment algorithms: intentional blocking, intentional termination and a hybrid one as a combination of these two. Simulation results show that our algorithm is effective even when the traffic is bursty. We then consider proportional QoS provisioning in a prosperous optical network architecture-optical burst switching (OBS) network. A previous differentiation provisioning scheme is studied and several problems associated with it are pointed out. Unlike the previous work, packet delay differentiation can not be configured by network operator. A new burst assembling scheme is proposed to provide favorable delay differentiation. We also provide an algorithm similar to intentional blocking for burst loss rate differentiation. Since the proportional differentiation model was originally developed in traditional packet networks, we also consider how to make a practical implementation of this QoS model. A uniform scheduler and droppers with resetting give satisfctory results on delay and packet loss rate differentiation. The proportional differentiation model is a relative QoS model, which makes absolute QoS guarantee impossible by itself. We applied QoS constraints releasing instead of a dynamic class selection approach. Extensive simulation results show that our scheme successfully fulfills the absolute loss rate/delay requirement even in a proportional QoS provisioning.