Fog Orchestration meets Proactive Caching

Running fog computing applications on edge servers requires to match activation of applications containers to time varying demands. In this paper we study the dynamic orchestration of a batch of applications over a network infrastructure including fog servers and a cloud. Cloud application deployment faces higher cost and high latency, but unlimited computational capacity. Fog servers, conversely, have limited computational resources, but ensure low latency at low cost. In this context we propose a new scheme for joint caching and placement in fog: the aim is to minimize the deployment cost while satisfying the applications' constraints. In fact, image caching appears mandatory to reduce the containers' activation time. On the other hand, proactively caching images on target servers is effective to match the expected activation pattern while optimizing load balancing via container replication. Using two-stage stochastic programming we derive a one-step-ahead policy to minimize the total running cost and satisfy applications' requirements. Extensive numerical results demonstrate the potential for this novel approach over traditional caching and placement algorithms.