Priority Based Approach For Pre-Warming and Pre-Baking Functions to Optimize Cold Start In Serverless Computing

A recent cloud computing webinar spoof called “Serverless” or “Function as a service” (FAAS) claims that developers should primarily concentrate on building code that controls an application's business logic, with the cloud service provider handling the back-end code execution. This is different from more traditional cloud models like Infrastructure-as-a-Service (IaaS) and containers, which required a lot of management to orchestrate virtual machines and apps at the level that serverless technology, especially Function-as-a-Service (FaaS) offerings, largely avoids by allowing service providers to control the underlying infrastructure. The ability of serverless computing to scale to zero and not charge users when not in use is a major advantage. Another drawback of cold start serverless computing is the modest latency associated with initializing function execution settings. Real-time applications are impacted by the overhead associated with configuring containers, deploying functions, and then unpacking function images for every new request. Even though serverless billing is based on complete execution time, resource management and cost inefficiencies will cause problems during an unplanned spike in traffic. To address these cold start issues, we provide predictive “Pre-warming” and “Pre-baking” techniques in this study. “Pre-baking” and “Pre-warming” optimize cold starts by pre-preparing containers and dependencies in advance for the function executions. Through the use of machine learning models trained to forecast function priorities in real time, predictive pre-warming and pre-baking minimize cold starts and optimize costs. In this paper we have investigated the combination of “Pre-baking” and “Pre-warming” techniques with a predictive modal to optimize cold starts in serverless computing systems.