Please use this identifier to cite or link to this item: http://repository.aaup.edu/jspui/handle/123456789/1448
Title: Time Series Prediction of Server Workload Using Hybrid Model of Recurrent Neural Network and Genetic Algorithms
Authors: Talahmeh, Amin $AAUP$Palestinian
Awad, Mohammed $AAUP$Palestinian
Eleyat, Mujahed $AAUP$Palestinian
Keywords: Server workload
Time Series Prediction
RNNs, LSTM, NARX,
Hybrid Model, GAs
Issue Date: 17-Dec-2021
Publisher: The International Journal of Engineering & Science
Citation: Amin Talahmeh, Mohammed Awad*,Mujahed Eleyat Time Series Prediction of Server Workload Using Hybrid Model of Recurrent Neural Network and Genetic Algorithms, The International Journal of Engineering and Science (IJES), Volume 10 Issue 12 Series I |Pages PP 01-10 2021
Series/Report no.: Volume 10 Issue 12 Series I;PP 01-10
Abstract: Internet users demand service from online websites by sending requests to website servers which process the requests and return responses to the clients. Many requests could be sent to a server at the same moment; therefore, predicting server workload becomes one of the factors that affect its efficiency. In this paper, we applied three optimized hybrid models to predict the workload of the Portal server (Server 1) and web server (Server 2) of the Arab American University Palestine (AAUP). The models are called Recurrent Neural Networks combined (RNNs), Long-short term memory network (LSTM), and Nonlinear Auto-Regressive Exogenous Neural Networks (NARX) hybrid with Genetics Algorithms (GAs). The experimental results showed that the hybrid model (NARX-GAs) has a better performance than (RNNs-GAs and LSTM-GAs), while the LSTM-GAs model produces better accuracy than RNNs-GAs when used to predict the workload of Server 1, and the RNNs-GAs model produces better accuracy than LSTM-GAs in predicting Server 2 workload. These findings, which are expressed by the RMSE factor, were obtained after the proposed models were applied to the used datasets (Servers 1&2 Processor and Memory usage). Accordingly, when the proposed models (RNNs-GAs, LSTM-GAs, and NARX-GAs) were applied to Server 1 Processor dataset, the RMSE test values were 0.1003, 0.1031, and 0.0998 respectively while they were 0.1687, 0.1676, and 0.1668 when applied to server 1 memory dataset. In addition, they were 0.0547, 0.0609, and 0.0417when applied to Server 2 Processor dataset and 0.1052, 0.1115, and 0.1125 when applied to Server 2 Memory dataset. This showed that the RMSE test value for NARX-GAs is slightly greater than other models which makes the NARX-GAs the best hybrid model compared to the other models that we tested.
URI: http://repository.aaup.edu/jspui/handle/123456789/1448
ISSN: ISSN (e): 2319-1813 ISSN (p): 20-24-1805
Appears in Collections:Faculty & Staff Scientific Research publications

Files in This Item:
File Description SizeFormat 
A1012010110.pdf593.19 kBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Admin Tools