Accelerated life test (ALT) has been widely used to accelerate the product reliability assessment process by testing product at higher than nominal stress conditions. For a system with multiple components, the tests can be performed at component-level or system-level. The data at these two levels require different amount of resources to collect and carry different values of information for system reliability assessment. Even though component-level tests are cheap to perform, they cannot account for the correlations between the failure time distributions of different components. While system-level tests can naturally account for the complicated dependence between component failure time distributions, the required testing efforts are much higher than that of component-level tests. This research proposes a novel resource allocation framework for ALT-based system reliability assessment. A physics-informed load model is first employed to bridge the gap between component-level tests and system-level tests. An optimization framework is then developed to effectively allocate testing resources to different types of tests. The information fusion of component-level and system-level tests allows us to accurately estimate the system reliability with a minimized requirement on the testing resources. Results of one numerical example demonstrate the effectiveness of the proposed framework.