Load balancing and cloud node resilience enhancement algorithm based on predicted integral index

The paper addresses the problem of load balancing in cloud environments designed for processing stochastic IoT traffic. It is established that traditional reactive methods are insufficiently effective under conditions of sharp fluctuations in request intensity. An adaptive routing algorithm based on the predicted integral node load index is proposed. This index aggregates CPU, RAM, disk I/O, and network activity metrics into a single criterion. An LSTM recurrent neural network model is used to forecast node states. The problem of minimizing expected service time and overload is formalized. The implementation of the proposed approach enables preventive resource distribution, thereby enhancing system stability. Furthermore, the proposed approach significantly contributes to the cybersecurity and resilience of cloud infrastructure. Within IoT ecosystems, node saturation frequently leads to compromised service availability and heightened susceptibility to Denial-of-Service (DoS) attacks. Leveraging the predicted integral load index facilitates not only the optimized allocation of computational resources but also the early detection of load anomalies attributed to aberrant IoT device behavior or hostile actions. This fosters the integration of load balancing with security monitoring frameworks, bolstering cloud fault tolerance and minimizing service degradation risks arising from both systemic failures and malicious intent.