بهینه سازی منابع در گسترش اینترنت اشیا بر اساس مه Resource Optimization in Fog Enabled IoT Deployments

I. Introduction Mobile Internet of Things (IoT) [1] devices that typically operate on the sense-compute-actuate framework [2], suffer from common restrictions: 1) Low computational capacities on sensing devices. 2) Low energy and battery capacities. 3) Limited on-board memory and storage capacities. 4) Low-latency turnaround times needed between sensing and actuation. 5) Variability in network connectivity due to mobility patterns of sensor/actuating devices. In order to overcome these challenges in IoT devices, cloud based architectures [3] have been proposed. Cloud computing [4] offers scalable hardware resources on demand for storing and computing data. This minimizes the computational load on network devices and increases their battery lifetime. However the communication latency and power consumed for communication with the cloud turn down its advantages for delay constrained applications. Mobile Robotic Sensor/Actuator (Edge Node) Mobile Robotic Sensor/Actuator (Edge Node) Mobile Robotic Peer (Edge Node) Mobile Robotic Peer (Edge Node) Mobile Robotic Peer (Edge Node) Mobile Robotic Peer (Edge Node) Fog VM Fog VM Request Resources Assign Tasks To Compute Nodes Network Variability / High Latency Cloud VMs Cloud VMs Run-time Optimizer (Computation Capacity, Energy Depletion) Fog Compute Node Fig. 1. Mobile IoT Deployments with Runtime Optimization. For low latency requirements, Fog Computing has been proposed, which brings the merits of cloud computing to the Edge of network [5], [6]. In Fog Computing, Fog nodes1 (such as smart gateways) that have high computational capacities, storage power and network access are used to perform computational tasks. As a result, computational capacities of both Edge nodes (low end peripheral nodes) and Fog nodes may be exploited for distributed computations. Typically, latency overheads and limited reliability of centralized cloud access is also mitigated, as a result of this distributed computation. Fig. 1 shows the general model of our approach, with a Fog node and multiple mobile IoT sensor/actuator nodes. Directly offloading computations to cloud based VMs is not viable due to network variations and latency overheads [7]. Furthermore, the Edge nodes are low end nodes that have limited battery and computational resources for carrying out high computational tasks. Hence, distributing a heavy task among the Edge/Fog nodes reduces computational burden on a single node and improves overall battery lifetime.