Prediction of Execution Time for PCIe Performance with Regression Models

Abstract

In manufacturing company that build product based on their customer preferences and requirements due to their strategic build to order manufacturing model. Based on this approach, they have a high combination of products which causes PCIe (peripheral connector interphase) having inconsistency performance in compliance testing. The product that has a bad performance was unable to determine and prevented thus affecting the overall performance of the graphic functionality. Furthermore, data storage using solid-state drive (SSD) currently has limitations includes time consuming, manually process and data leakage. The PCIe data only saved with SSD that possible misplaced and easily occupied. All the chipset machines having a lot of raw data which easily occupied the storage. Due to the limitation of data storage for PCIe in local host and importance of PCIe data to the company, thus, this project aims to enable AV Cloud with Heidi SQL to upload and retrieve from CDS (cloud domain storage) on PCIe data for each machine connected via internal network. Moreover, speeding up execution time on PCIe is needed to improve the performance in compliance testing that predict execution time for PCie using machine learning approaches including five different regression models ((K-Nearest Neighbour, AdaBoost, Bagging, Linear Regression, Random Forest). The evaluation experiments show that the overall predictive models can predict the execution time for PCIe performance with accuracy more than 70%. The project will be improved in future using the predicted execution times to optimize the PCIe testing and recommended to work on different data science techniques for development process.