Research and implementation of software architecture optimization algorithm for high concurrency scenarios
Zhen Fan, Rumeng Zhang, Qipeng Hu · IET conference proceedings. · 2026
Aiming at the performance challenges faced by software systems in high concurrency scenarios, this paper proposes a dynamic adaptive software architecture optimization algorithm. The algorithm effectively improves the performance and resource utilization of the system through three key modules: dynamic load balancing, intelligent caching strategy and elastic resource allocation. Dynamic load balancing algorithm is based on load prediction, which predicts the load trend of nodes by exponential smoothing method, and allocates requests according to probability by using softmax function to avoid local overload. The intelligent caching strategy adopts LSTM prediction model, which combines access frequency and predicted access probability to optimize caching decision, so as to improve cache hit rate and data freshness. The elastic resource allocation module uses the deep reinforcement learning (RL) algorithm to dynamically adjust the resource allocation according to the system state and balance the throughput, delay and resource cost. The experimental results show that, compared with the traditional methods, the proposed optimization algorithm has significantly improved the key performance indicators such as average delay, cache hit rate, resource utilization rate, etc. The system throughput has increased by 23%, the average response time has decreased by 41%, the resource utilization rate is more stable, and the error rate has been greatly reduced, effectively improving the overall performance and stability of the software system in high concurrency scenarios.