Serverless Cold Start Performance Optimization Based on Multi-Request Processing and Adaptive Hierarchical Scaling

Abstract

Serverless has gained significant attention due to its advantages of high elasticity, low cost, and reduced operational overhead. However, the cold start problem severely constrains its performance and user experience. This paper proposes an innovative hierarchical startup optimization scheme that introduces a critical parameter called scaling threshold based on multi-concurrency execution within a single instance. By employing fine-grained instance concurrency control and adaptive scaling, this approach reduces the probability of cold starts while maintaining resource utilization efficiency. We systematically present the optimization principles, controller design, and implementation details of the hierarchical startup scheme. Extensive experiments are conducted to evaluate the proposed scheme for various types of Serverless applications. The results demonstrate that the hierarchical startup scheme significantly improves the cold start performance of Serverless applications, further reducing the cold start rate by 18.4% to 25.6% and the average response time by 8.9% to 27.5%. Further theoretical analysis and experimental data confirm the rationality and effectiveness of the scheme’s design. The research findings enrich the theory and methods of cold start optimization in Serverless and hold significant value for enhancing Serverless application performance and promoting the widespread adoption of Serverless.

Keywords

• Serverless computing
• cold start
• performance optimization
• adaptive scaling
• multi-concurrency