Adaptive Control of Stream Data Processing in Distributed Systems Based on Load Forecasting

Introduction. To propose a reference architecture for closed-loop stream-processing adaptation and preliminarily evaluate predictive scaling policies under variable synthetic workloads. Materials and Methods. A closed-loop control architecture is proposed that integrates multi-level monitoring, short-term workload forecasting, a stabilized decision controller, controlled reconfiguration, and audit logging. The simulation compares a reactive threshold policy with two predictive approaches: EWMA with hysteresis and delayed reconfiguration, and a warm-pool approach with pre-initialized resources. A robustness analysis over 30 random seeds was additionally performed. Results. For the main seed, the number of intervals with SLA-violation risk decreased from 26 to 3 and 0, while average estimated latency decreased from 201.23 ms to 107.62 ms and 92.17 ms, respectively. Conclusions. The obtained results are preliminary and support further implementation and validation of the proposed approach on a real stream-processing platform.