Typically, management of networked computational and sensing nodes is based upon a quality of service metric (QoS) that is based on some generic principles, like "be fair in allocating resources" or "utilize the CPU capacity to the maximum". The consequences of accepting such a starting point is that (1) task-specific resource requirements are not taken into consideration, and (2) computational and communication resources are saturated without paying attention to whether such a high load is necessary or not. In this paper we describe some of our efforts on how to improve the situation described above. In particular, we discuss one of the approaches that we are currently investigating that can be summarized by the following three points. (1) We use a task-specific QoS (TS-QoS) as a variable that is controlled by our system. (2) Requests for resources are generated based upon the feedback provided by the TS-QoS, where the request generator's parameters are adjusted using a simple PID controller. (3) A Dynamic Programming based algorithm is used for scheduling resource. Simulations using sensor resources show some of the advantages of the proposed approach.