Efficient Container Retention Strategies for Serverless Edge Computing

Serverless edge computing follows an event-driven function execution model in which the function gets executed only when the event triggers, efficiently utilizing the edge server resources. However, starting the containers upon event trigger involves considerable delay known as cold-start latency, hindering the edge services’ responsiveness. In a typical serverless platform, cold-start latency can take up to a few seconds, which is unacceptable for latency-sensitive applications. Current literature proposes many solutions to alleviate the cold-start latency by leveraging various container states such as warm containers, pre-warm containers, pause containers, and pre-bake containers. All these approaches have some overhead, especially in terms of memory and storage, and may not be suitable for all types of applications. Moreover, finding a single solution that reduces the cold-start latency for all applications is challenging. In this work, we focus on characterizing various container states to understand their resource requirement and their impact on reducing the cold-start latency. This is the fundamental step toward designing a container retention strategy to minimize the cold-start latency experienced by the services while using the edge resources efficiently.