Experimental results from a testbed show that TAN models involvingsmall subsets of metrics capture patterns of performance behavior ina way that is accurate and yields insights into the causes ofobserved performance effects. TANs are extremely efficient torepresent and evaluate, and they have interpretability propertiesthat make them excellent candidates for automated diagnosis andcontrol. We explore the use of TAN models for offline forensicdiagnosis, and in a limited online setting for performanceforecasting with stable workloads.