ISSN 0278-6419 (*printed)
ISSN 1934-8428 (electronic version)
ISSN 1934-8428 (electronic version)
Recieved: 01/28/2025
Accepted: 02/10/2025
Keywords: prediction of time characteristics, training of random forest models, principal component ananlysis, multilayer perceptron, convolutional neural networks
Piskovsky V.O., Lycheva E., Mogilinets V. Forecasting the time characteristics of network services. // Moscow University Journal. Series 15. Computational Mathematics and Cybernetics. 2025. N 3, p.62-73 https://doi.org/10.55959/MSU/0137–0782–15–2025–49–3–62–73.
The efficiency of network services depends on how applications are distributed in the network. At the same time, these services are to be provided with a certain quality of services and experience for users, i.e.:
1) access to information “at any time anywhere”;
2) seemless integration of informational systems;
3) timely monitoring and analysis of different data sources;
4) migration from a centralized scheme “data download with cleaning followed by analysis and distribution” to the scheme “distributed placement and preprocessing of data and, if necessary, subsequent loading, analysis and distribution”.
One of the key preconditions to provide an acceptable level of services is forecasting the execution time of network services on a variety of virtualization platforms and physical equipment.
The work analyzes approaches to predicting temporal characteristics of network services. The methods are based on operational log data of services and physical equipment, take into account the current and predictable state of hardware and software resources. Among the solutions considered are machine learning models, including a random forest, multilayer perceptrons, and convolutional neural networks.
