User satisfaction determines the success of web-database applications. User satisfaction can be expressed in terms of expected response time or expected delay. Given the bursty and unpredictable behavior of web user populations, we model user requests as transactions with softdeadlines. For such a model of user requests with soft-deadlines, the hit ratio is not the most expressive metric. Instead, the average tardiness is a better measure in such cases. In this paper; we propose and evaluate an adaptive self-managing algorithm called ASETS, which optimizes for the average tardiness. ASETS prioritize resources as needed in order to keep users satisfied under varying workloads. Our performance evaluation shows ASETS to out-perform both EDF and SRPT which are known to be optimal for the under and over utilization system conditions respectively.