Deep Learning for Failure Recovery and Efficiency Enhancement in Hypercube Networks

Abstract

Hypercube networks are widely used in parallel and distributed computing systems due to their high scalability and inherent fault tolerance. However, node or link failures lead to significant performance degradation, negatively impacting data throughput and network availability. This research presents a novel deep learning-based framework for intelligent fault recovery in hypercube networks, utilizing dimensionally sensitive convolutional neural networks. The proposed model is designed to operate across different dimensions of the hypercube (2D, 4D, and 6D) and is trained using artificially generated fault scenarios involving node and link failures. Experimental results demonstrate a significant performance improvement compared to traditional methods. In the 2-dimensional hypercube, the recovery time decreased to 0.1905 seconds, the throughput increased to 0.5, and the response time was reduced to 1.9608 seconds, compared to 3.8462 seconds in the traditional approach. In the 4-dimensional cube configuration, recovery time improved from 0.2319 seconds to 0.1825 seconds, throughput increased from 0.1484 to 0.25, and response time decreased from 6.3116 seconds to 3.8462 seconds. In the 6-dimensional configuration, cycles per second decreased from 0.2014 to 0.1709, throughput increased from 0.0405 to 0.0938, and response time decreased from 19.7913 seconds to 9.6386 seconds. These results confirm the effectiveness of deep learning in enabling adaptive, resilient, and self-healing networking, thereby enhancing the reliability and performance of large-scale distributed systems in fault-prone environments.