Deep learning-based port-classification approach incorporating LSTM network for high-throughput data center interconnect

This paper presents a solution to the throughput challenges of interconnecting systems in data centers that process big data every second. The Long-Short-Term-Memory (LSTM)-based neural network is implemented. It discovers the available switching ports and classifies them for short distances and minimum error to receive data with high throughput and low latency. The structure is promising by showing a prediction accuracy of 96.88% and a classification accuracy of 99.7% with a precision of 99.6%. The specificity of the model turns out to be 98.6%. The neural network training is done on the comparative analysis, with training samples varying between 50–2000 epochs, to know the best epochs to achieve a higher accuracy rate. The comparison of various parameters w.r.t. word length presents a solution to minimize cross-interference by controlling the word length between 10–14 bits at the input. The bit error rate of 10^–18 and extinction ratio of 19.5 dB show the effective error-free transmission of bits. It is observed that the minimum log of bit error rate and eye-opening factor show immediate improvement of -3 and 10% respectively when the switching of the signal is done by using the information created by the neural network model. The small switching time of 0.35 ns and throughput of 96 Tbps prove the applicability of the proposed neural network-based structure for low-latent high-throughput configurations.