The subject of this study is the impact of event batching strategies on performance and resource utilization in stream processing systems under dynamic load. The object of the study is event batching strategies, including fixed-size batching, time-window batching, and an adaptive batching algorithm. The paper examines the relationship between throughput and processing latency and the parameters of batch formation. It also analyzes the impact of batching strategies on CPU utilization and queue behavior under varying input rates. Particular attention is given to the behavior of algorithms under bursty load conditions and during transient system states. The aim of the study is to evaluate the effectiveness of batching strategies and to identify the most efficient approach. The research follows an empirical methodology. A Java 21 application was developed. A producer–consumer architecture was implemented. A Poisson event stream with variable intensity was simulated. A comparative analysis was conducted using the following metrics: throughput, average latency, P95 latency, CPU usage, and queue size.The main finding of the study is the confirmation of the effectiveness of the adaptive batching strategy. It provides the best balance between performance and resource utilization and reduces the risk of queue buildup under changing load conditions. It is shown that the absence of batching results in low throughput, high latency, CPU overload, and significant queue growth. Fixed-size and time-window batching improve performance but introduce trade-offs between latency and resource efficiency. The best results are achieved with adaptive batching. The novelty of the study lies in the analysis of batching strategies under bursty load and transient conditions, as well as in the development of an adaptive algorithm based on an exponential moving average. The results can be applied to high-load stream processing systems, including fintech platforms, monitoring systems, and analytical services.
Daria Zolotukhina (Sun,) studied this question.