Abstract:
The reliability of modern distributed systems directly depends on the stability of network connections; however, traditional monitoring methods are unable to adequately assess the stochastic nature of failures at the TCP transport protocol level. This paper proposes an approach based on Stochastic Differential Equations (SDEs) to model packet loss probability as a continuous random process, accounting for mean reversion and random fluctuations. A practical implementation of the model is presented in C++ using AVX-512 vector instructions for the numerical solution of the SDE via the Euler–Maruyama method. Experimental evaluation on an Intel Xeon Silver 4410Y server platform demonstrated that the module's performance reaches 30.1 million estimations per second, which is nearly 9 times faster than reference scalar implementations. The results prove that the proposed stochastic approach is computationally efficient and can serve as a foundation for creating real-time monitoring and adaptive control systems capable of predicting TCP performance.
