Minimizing Cold Start Time on Serverless Platforms Based on Time Series Prediction Methods

The research in this paper focuses on minimizing cold start times in serverless platforms using time series forecasting methods. The study begins by analyzing the mechanism of cold start issues in serverless computing and proposes a method based on time series prediction to forecast future function call volumes. This method is designed to reduce cold start occurrences by pre-starting functions based on predicted demand. Experimental validation confirms the effectiveness of this approach. The methodology includes collecting historical request data, preprocessing, parameter determination, and model training. The SARIMA model is employed for predictions, allowing for dynamic resource allocation in response to predicted demand, significantly reducing cold start times and resource wastage compared to traditional static threshold-based methods. The study demonstrates that using SARIMA models for future function call predictions effectively minimizes cold start times on cloud-native platforms, enhancing system responsiveness and resource utilization.