QoS management refers to the allocation and scheduling of computing resources. Static QoS management techniques provide a guarantee that resources will be available when needed. These techniques allocate resources based on worst-case needs. This is especially important for applications with hard QoS requirements. However, this approach can waste resources. In contrast, a dynamic approach allocates and deallocates resources during the lifetime of an application. In the dynamic approach the application is started with an initial resource allocation. If the application does not meet its QoS requirements, a resource manager attempts to allocate more resources to the application until the application's QoS requirement is met. While this approach offers the opportunity to better manage resources and meet application QoS requirements, it also introduces a new set of problems. In particular, a key problem is detecting why a QoS requirement is not being satisfied and determining the cause and, consequently, which resource needs to be adjusted. This paper investigates a policy-based approach for addressing these problems. An architecture is presented and a prototype described. This is followed by a case study in which the prototype is used to diagnose QoS problems for a web application based on Apache.