STEROCEN: Simulation and Training Environment for Resource Orchestration in Cloud-Edge Networks

Abstract

Large scale deployment of Internet-of-Things (IoT) devices is projected to grow in the coming years. These devices are expected to be low-cost while supporting applications with growing computational demands. To enable the necessary computations, offloading of computational tasks to Edge and Cloud nodes is a fundamental technology. However, orchestration for such networks is a complex problem which affects both the network design and the decision system. To aid in solving this problem, simulation tools are essential for predicting the performance of networks in different conditions and under different orchestration policies. In this paper, we propose STEROCEN, a Cloud-Edge network resource orchestration simulation and training tool which allows for different configurations of up-to a four-layer network composed of: (i) end-device, (ii) Close Edge, (iii) Far Edge, and (iv) Cloud layers. Our tool collects delay metrics for flexibly defined applications, especially in regard to computation in the network nodes and including uncertainty in processing times. Additionally, the tool only needs the initial configuration and an independently defined orchestrator, allowing for testing of many strategies. As an example, we provide results of testing some Deep Reinforcement Learning (DRL) algorithms using the same training and simulation environment.