An Adaptive Fault Diagnosis Scheme for the Augmented Cube-Based Data Center Networks

System-level fault diagnosis is the process of identifying faulty nodes within a multiprocessor system. This technique has been used in various multiprocessor-based systems, including mobile networks, wireless sensor networks, and data center networks. Adaptive diagnosis is an effective strategy for system-level fault diagnosis, as it significantly reduces the number of tests required when compared to a one-step strategy. The k-dimensional augmented cube, denoted as AQ_k, is a recursive network with the scalability and fault tolerance needed for data center networks. The term AQDC_k refers to the data center network based on AQ_k, with its logic graph represented as L-AQDC_k. In this article, we propose an efficient, novel diagnosis algorithm, AQDO, that can adaptively identify faulty nodes in a data center network. Theoretical analyses indicate that the number of tests required by the AQDO algorithm is (2k − 1) × (2^k + 2k − 1) for 4 ≤ k ≤ 7, and k2^k+1 for k ≥ 8. These results suggest that the algorithm is approximately optimal in terms of the number of test rounds required. Let F be a set of faulty nodes in L-AQDC_k. Simulation results show that when |F | ≤ 2 − 1, AQDO can accurately identify all faulty nodes. In contrast, if |F | > 2k − 1, although AQDO cannot diagnose all faulty nodes, it still maintains a precision rate of 100% and an accuracy rate above 90%. Furthermore, even when |F | exceeds the diagnosability, AQDO outperforms previous adaptive diagnosis algorithms designed for L-HSDC_k(k) and BCube(n, k) networks. Moreover, the AQDO algorithm achieves over 75% reduction in the average number of tests relative to traditional methods.