Detection and classification of network congestions in multi-machine computing systems

The purpose of the research is to develop criteria for detecting and classifying overloads occurring in multi-machine computing systems based on network traffic data based on an analysis of its regular trend and subsequent probabilistic interpretation of the overload state.

Methods. The wavelet regression processing of the time series is used as a method for detecting the regular trend of network traffic, which ensures the suppression of high-frequency disturbances and the preservation of the regular component. To identify overload intervals, a Bayesian pulse signal detector is used, which generates a smoothed pulse sequence of a posteriori probabilities of exceeding the normal mode. To quantify each overload episode, a parametric pulse model based on Gaussian approximation is introduced, which makes it possible to associate overload with three measurable characteristics: amplitude, duration, and integral area. Additionally, normalized parameter indicators relative to normal mode statistics and an integral deviation criterion are proposed, which provides an assessment of the severity of the overload condition.

Results. Criteria and rules for classifying overloads into five types are obtained: impulse, background, progressive, periodic and attacking. The results of simulation modeling on data with different load modes showed the efficiency of the proposed approach: overload episodes are consistently highlighted, and the typing corresponds to the expected structure of scenarios. The features of the probabilistic detector’s response to a smoothly increasing load, which can manifest itself as a sequence of local pulses, are revealed.

Conclusion. The developed technique provides a transition from visual interpretation of the Bayesian detector output signal to quantification and typification of congestion and can be used in tasks of automated monitoring of network traffic and support for quality-of-service management solutions in multi-machine computing systems.

1. Petushkov G.V. Computational system performance evaluation. Izvestiya Yugo- Zapadnogo gosudarstvennogo universiteta = Proceedings of the Southwest State University. 2025;29(2):201-220. (In Russ.)

2. Hassan S.O. AD-RED: A new variant of random early detection AQM algorithm. Journal of High Speed Networks. 2024;30(1):53-67.

3. Toopchinezhad M.P., Ahmadi M. Machine learning approaches for active queue management: A survey, taxonomy, and future directions. Available at: https://arxiv.org/abs/2410.02563 (accessed 15.03.2026).

4. Gadasin D.V., Kobelkova A.D., Rodina A.A., Surova M.A. Network traffic optimization technologies. Sistemy sinkhronizatsii, formirovaniya i obrabotki signalov = Synchronization Sys- tems, Signal Generation and Processing. 2025;16(3):9–16. (In Russ.)

5. Muhammad S., Chaudhery T.J., Noh Y. Study on performance of AQM schemes over TCP variants in different network environments. IET Communications. 2021;(15):93-111. https://doi.org/10.1049/cmu2.12061.

6. Khandozhko V.A., Fedonin O.N., Matlakhov V.P., Khandozhko A.V. Mathematical modeling of automatic control system by numerical integration. Izvestiya Yugo-Zapadnogo gosudarstvennogo universiteta = Proceedings of the Southwest State University. 2025;29(2):55-70. (In Russ.)

7. Gong T., Lee J., Cheng X., Xie Y. Neural network-based CUSUM for online changepoint detection. Available at: https://arxiv.org/abs/2210.17312 (accessed 15.03.2026).

8. Shimono M., et al. Performance measurement of Bayesian online changepoint detection and its application. In: Second International Conference on Advanced Robotics, Automa- tion Engineering, and Machine Learning (ARAEML 2025). Vol. 13815. SPIE; 2025. P. 83– 103.

9. Tsaknaki I.-Y., Lillo F., Mazzarisi P. Bayesian Autoregressive Online Change-Point Detection with Time-Varying Parameters. Available at: https://arxiv.org/abs/2407.16376 (accessed 15.03.2026).

10. Corneck J., et al. Online Bayesian changepoint detection for network Poisson processes with community structure. Statistics and Computing. 2025;35(3):1-29.

11. Bekhtin Y.S., Balanev K.S. Simulation Modeling Network Traffic Behavior Using Regression Analysis in Wavelet Domain. In: 2024 6th International Youth Conference on Radio Electronics, Electrical and Power Engineering (REEPE). Moscow: IEEE; 2024. P. 1–6.

12. Bekhtin Y.S., Balanev K.S., Dubrovskaya E.A., Pavlovich A.V. Stability Appraisal of the Wavelet-and-Regression Method for Unsteady Network Traffic Trend Highlighting. In: 2025 7th International Youth Conference on Radio Electronics, Electrical and Power Engineering (REEPE). Moscow: IEEE; 2025. P. 1–5.

13. Klimenko A.B. A Technique of the Distributed Information Systems Control Method Choice under the High Network Dynamics Conditions. Izvestiya Yugo-Zapadnogo gosu- darstvennogo universiteta = Proceedings of the Southwest State University. 2022;26(1):57- 72. (In Russ.)

14. Ferriol-Galmés M., et al. RouteNet-Fermi: Network Modeling With Graph Neural Networks. IEEE/ACM Transactions on Networking. 2023;31(6):3080-3095.

15. Bicski B., Pekar A. Early Detection of Network Service Degradation: An Intra-Flow Approach. In: 2024 20th International Conference on Network and Service Management (CNSM). Prague: IEEE; 2024. P. 1–5.

16. Graf F., Watteyne T., Villnow M. Monitoring Performance Metrics in Low-Power Wireless Systems. ICT Express. 2024;10(5):989-1018.

17. Espinal A., Sanchez Padilla V. Queuing Delay Reduction based on Network Traffic Patterns: A Predictive QoS Framework For Point-To-Point Communications. Transport and Telecommunication Journal. 2025;26(3):237-249.

18. Ali I., Hong S., Cheung T. Congestion or No Congestion: Packet Loss Identification and Prediction Using Machine Learning. In: 2024 International Conference on Platform Technology and Service (PlatCon). Jeju; 2024. P. 72–76.

19. Jin Q. Optimized transmission of multi-path low-latency routing for electricity internet of things based on SDN task distribution. PLoS One. 2025. P. 1–23. https://doi.org/10.1371/journal.pone.0314253.

20. Alauthman A., Al-Hyari A. Analysis of LTE/5G network performance parameters in smartphone use cases: a study of packet loss, delay and slice types. International Journal of Computer Networks & Communications (IJCNC). 2025;17(4):75–93.

21. Gaoyang G., Zhiyi C., Qamar F., Hafizah Mohd Aman A. A Comparative Study of TCP and UDP Performance Using NS-3 Simulation. Asia-Pacific Journal of Information Technology and Multimedia. 2025;(14):1-19.