Abstract—Proliferation of QoS-sensitive client-server Internet applications such as high-quality audio, video-on-demand, e-commerce, and commercial web hosting has generated an impetus to provide performance guarantees. These applications require a guaranteed minimum amount of resources to operate acceptably to the users, thus calling for QoS-provisioning mechanisms. One good place to locate such mechanisms is in server communication subsystems. Server-side communication subsystems manage an increasing number of connection end-points, thus readily controlling important bottleneck resources. We propose, implement, and evaluate a novel communication server architecture that maximizes the aggregate utility of QoS-sensitive connections for a community of clients even in the case of overload. A contribution of this architecture is that it manages QoS from the user space and is transparent to the application. It does not require modifications to the OS kernel, which improves portability and reduces development cost. Results from an experimental evaluation on a microkernel indicate that it achieves end-system overload protection and traffic prioritization, improves insulation between independent clients, adapts to offered load, and enhances aggregate service utility.