Predictive intelligent orchestration of cluster computing resources in high-load financial systems based on the state evolution vector

Problem Statement. In modern, highly loaded distributed financial infrastructures, traditional monitoring and orchestration systems rely on reactive heuristics and static thresholds for technical load (CPU, RAM). This makes it impossible to proactively detect hidden microservice degradation and targeted application-layer cyberattacks, which entails unacceptable risks of cascading system failures.

Goal is to develop a mathematical model for dynamic cluster profiling and create a hybrid neural network core for iterative, proactive forecasting of infrastructure health, followed by integration of this data into software orchestrator controllers.

Results: The concept of a five-dimensional normalized infrastructure evolution vector, which continuously aggregates metrics of performance, queue load, quality of service, resource consumption, and computed behavioral anomalies, is introduced and mathematically substantiated. A hybrid analytics framework combining the capabilities of LSTM recurrent networks for finding hidden patterns with the mathematical stability of multi-step linear extrapolation using a discrete Kalman filter was designed and successfully tested on a full-scale testbed.

Practical Relevance. The implementation of the developed multidimensional state predictive kernel for container management systems such as Kubernetes enables fintech enterprises to transition from a reactive disaster recovery paradigm to proactive self-regulation. The technology enables proactive autoscaling of resources and intelligent isolation of compromised network nodes before the impact of an incident leads to a violation of strict SLA requirements for processing customer transactions.