Pre-warming: Alleviating Cold Start Occurrences on Cloud-based Serverless Platforms

Serverless computing has gained increasing attention recently due to its benefits, such as streamlined deployment processes and enhanced efficiency. Despite its growing popularity, the well-known cold start problem in serverless platforms limits the wider adoption of serverless computing across applications that need real-time latency and response time. Understanding and mitigating the impact of cold-start latency is crucial for maximizing the efficiency and effectiveness of serverless computing platforms. Cold start delay remains unresolved, particularly in public cloud platforms as research conducted on private serverless platforms cannot be seamlessly applied to public cloud platforms. In this paper, we propose pre-warming strategy which warms the functions ahead of actual function invocation requests by predicting the arrival time of invocations. The results demonstrate a significant reduction in the cold start occurrences and an enhancement in overall performance. Based on our experiments, we found that compared to the baseline method, our proposed model reduced the cold start occurrences up by more than $80 \%$.