Neighbor-aware shared container instance warming framework for serverless edge computing

In serverless edge computing, cold starts can hinder timely task execution. In this paper, we propose a neighbor-aware shared container instance warming (NA-SCIW) framework. NA-SCIW employs two key concepts: (1) shared container instances; and (2) a cooperative keep-warming mechanism. By enabling multiple IoT devices to share a container instance, the framework ensures that as long as at least one request is made, the instance remains warm, preventing cold starts. However, if no requests are generated for a certain period, the shared instance transitions to a cold state. To mitigate this, NA-SCIW introduces a cooperative keep-warming mechanism, where each IoT device strategically decides whether to send a virtual request based on the actions of nearby devices. We formulate the problem as a constrained stochastic game model, aiming to minimize cold start occurrences while keeping the average additional traffic overhead and energy outage probability of IoT devices within acceptable limits. To solve this model, we develop an algorithm based on best response dynamics. Evaluation results demonstrate that NA-SCIW effectively reduces cold start probability while ensuring low traffic overhead and minimal energy depletion for IoT devices.