Performance Optimization Strategy for Large Scale Event Driven Software Architectures

Abstract

Large-scale event-driven software architectures support responsive and scalable applications, but their performance can decline because of uneven event distribution, queue growth, delayed acknowledgements, and inefficient consumer scaling. This study proposes an Adaptive Event Flow Optimization Strategy called AEFOS to improve throughput and latency under heavy workloads. The framework profiles event paths, measures arrival pressure, partition imbalance, retry concentration, acknowledgement lag, and dependency pressure, and then applies targeted actions such as consumer scaling, partition rebalancing, adaptive batching, event prioritization, and delayed retry control. The strategy was evaluated across five workload levels and compared with static processing, CPU-threshold scaling, and queuelength- based optimization. At 100,000 events per second, AEFOS achieved 94,600 successfully processed events per second and reduced 95th-percentile latency to 286 ms. These results show that coordinated event-flow optimization can improve system stability, resource use, and responsiveness in large enterprise event-driven platforms while limiting unnecessary capacity expansion during rapidly changing traffic conditions.