Deep learning architectures give better results than other artificial intelligence methods in solving many problems. For this reason, they have been used to solve many real-world problems. However, these models consume more hardware resources than classical machine learning methods. Especially when these models are used in web-based applications, it is vital to adjust the server traffic according to the increasing requests. One of the most important ways to adjust the server traffic is to estimate the data length of the incoming packets and thus dynamically adjust the analysis processes of the deep learning models published on the server. For this reason, in this study, a deep learning model consisting of LSTM and GRU layers is proposed to estimate the web traffic packet data length for dynamic use of deep learning models published on different ports on a server. By running deep learning models that analyze medical images on a real server, a dataset consisting of the number of packets and packet length at different time units was created by monitoring web traffic. Training and testing of the designed deep learning models were performed with this data set. The root mean square error value of the best model was obtained as 1.43. In the results obtained, the time periods with low web traffic on the server were predicted with low error value and dynamic analysis control was performed by running deep learning models in these time periods. From these results, the feasibility of deep learning based software for traffic control is estimated. ORCID NO: 0000-0003-4728-8438
Anahtar Kelimeler: Deep Learning, LSTM, GRU, Web traffic estimation
|
