Research indexConference paper
Optimizing Cold Start Performance in Serverless Computing Platforms
Kang
1 Jan 2026Open access
NobleIDark:/48914/w1/20261105/F2D37F4ADOI10.4108/eai.22-5-2026.2365240
Optimizing Cold Start Performance in Serverless Computing Platforms
Kang · 2026
NobleIDark:/48914/w1/20261105/F2D37F4ADOI10.4108/eai.22-5-2026.2365240
Abstract
Serverless Computing, through its Function as a Service (FPS) model, delegates infrastructure management, resource scaling, and runtime environment maintenance to the cloud platform, reducing application deployment and operational complexity.However, creating execution environments on demand introduces cold start latency and significantly impacts tail latency for interactive services, real-time data processing, and event-driven applications.This paper analyzes the main influencing factors related to the formation mechanism and optimization path of cold starts, focusing on code download, instance startup, runtime initialization, and user code initialization.It further summarizes methods such as container reuse, snapshot recovery, warm-up and elastic scaling, intelligent scheduling, and machine learning-based instance lifecycle management, comparing their differences in startup speed, resource overhead, implementation complexity, and load adaptability.The paper also illustrates the engineering implementation in a real-world platform using Alibaba Cloud Function Compute's instance pre-allocation, minimum instance count, concurrency configuration, and elasticity strategies.Analysis shows that a single method typically only optimizes a localized part of the cold start chain.For low-latency and cost-sensitive scenarios, a more effective solution is to dynamically combine multiple optimization strategies based on function runtime, dependency scale, call cycle, and business service level.