PCDA: Prediction-opinion based container Caching Decision Algorithm in Serverless Edge Environment

As the serverless computing model is integrated into edge environments, it enhances the flexibility and scalability of IoT systems, significantly improving the Quality of Service (QoS) for Function as a Service (FaaS) users. However, serverless edge computing faces several challenges, with one primary issue being the latency caused by container cold starts. While container caching mitigates, determining optimal caching duration in resource-constrained edge environments requires balancing conflicting resource and latency costs. We formulate this fundamental trade-off through a ski-rental problem lens and propose the Prediction-Opinion based Caching Decision Algorithm (PCDA). Our solution introduces a Dual-phase caching with deterministic keep-alive windows and prediction-driven pre-warm windows, and a confidence metric λ that dynamically calibrates decision randomness against prediction errors. Extensive simulation experiments demonstrated that, compared to existing algorithms, PCDA performed best under various caching conditions, with an overall system cost reduction of 12.04% to 73.97%.