Network Latency Prediction on a 2D Distributed Shared Memory Multiprocessor

Zayid E. I. M., AKAY M. F., Alfaki K. B. O., Parwekar P.

10th International Conference on Microelectronics, Electromagnetics and Telecommunication, ICMEET 2025, London, İngiltere, 30 - 31 Ekim 2025, cilt.1600 LNEE, ss.249-260, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 1600 LNEE
  • Doi Numarası: 10.1007/978-3-032-20533-9_21
  • Basıldığı Şehir: London
  • Basıldığı Ülke: İngiltere
  • Sayfa Sayıları: ss.249-260
  • Anahtar Kelimeler: Machine Learning, Model and Simulation, Multilayer Feedforward, Network Latency, Network Measures, Prediction, Shared Memory
  • Çukurova Üniversitesi Adresli: Evet

Özet

The durable and dependable two-dimensional simultaneous optical multiprocessor exchange bus (2D SOME-Bus) is a petaflop-class computer architecture. The purpose of this paper is to forecast the average network latency (NL) on a distributed shared memory (DSM) by employing multilayer feed-forward neural network models. OPNET Modeler is used to generate train and test datasets across the multiprocessor topology architecture. By controlling the input numbers in the hidden layer, several neural network models have been developed. Performance measures such as correlation coefficients (R), root mean square errors (RMSEs), are calculated to assess the prediction models’ effectiveness using 10-fold cross validation. Prediction performance improves when the hidden layer of neurons increases. The prediction value of the mean NL was extremely accepted based on the RMSEs. Our study findings concluded that an effective method for forecasting the mean NL of a 2D DSM architecture is a multilayer feed-forward neural network.