Computation Offloading and Resource Allocation for Low-power IoT Edge Devices Farzad Samie 1 , Vasileios Tsoutsouras 2 , Lars Bauer 1 , Sotirios Xydis 2 , Dimitrios Soudris 2 , J¨ org Henkel 1 1 Chair for Embedded Systems (CES), Karlsruhe Institute of Technology (KIT), Germany 2 Microprocessors and Digital Systems Laboratory, ECE, National Technical University of Athens, Greece Abstract—With the proliferation of portable and mobile IoT devices and their increasing processing capability, we witness that the edge of network is moving to the IoT gateways and smart devices. To avoid Big Data issues (e.g. high latency of cloud based IoT), the processing of the captured data is starting from the IoT edge node. However, the available processing capabilities and energy resources are still limited and do not allow to fully process the data on-board. It calls for offloading some portions of computation to the gateway or servers. Due to the limited bandwidth of the IoT gateways, choosing the offloading levels of connected devices and allocating bandwidth to them is a challenging problem. This paper proposes a technique for managing computation offloading in a local IoT network under bandwidth constraints. The existing bandwidth allocation and computation offloading management techniques underutilize the gateway’s resources (e.g. bandwidth) due to the fragmentation issue. This issue stems from the discrete coarse-grained choices (i.e. offloading levels) on the IoT end nodes. Our proposed technique addresses this issue, and utilizes the available resources of the gateway effectively. The experimental results show on average 1 hour (up to 1.5 hour) improvement in battery life of edge devices. The utilization of gateway’s bandwidth increased by 40%. 1 Keywords-Internet of Things, IoT, Edge Computing, Re- source Allocation, Computation Offloading I. I NTRODUCTION Recent advances in technologies of sensors, wireless communication and embedded processors have enabled the design of small-size low-power and low cost devices that can be networked or connected to the Internet. These are the key components of the emerging paradigm of Internet- of-things (IoT) [1, 2]. IoT is covering an ever increasing range of applications, such as healthcare monitoring, smart home, smart building, smart city, etc. One of the challenges in IoT is to process and analyze a huge amount of data from heterogeneous devices. The massive number of IoT devices will lead to a rapid explosion of the scale of collected data. This challenge has two aspects: 1) Big Data [3], and 2) diverse application requirements of IoT [4]. Handling all these collected data with central cloud servers is inefficient, and even sometime is unfeasible, because of: • the limitation of computing, communication, and storage resources, • the overall energy and cost, 1 This research has been partially supported by the E.C. funded program AEGLE under H2020 Grant Agreement No: 644906 Latency (communication + computation) IoT Edge devices gateways Fogs & cloudlets Clouds Processing Capability Predictability Quantity    "               ! Figure 1: Computation layers in IoT systems and their properties [1] • and unreliable latency [5, 6]. To deal with these issues, the task of processing the data is pushed to the network edges introducing concepts of Fog computing, cloudlet, and Mobile Edge Computing (MEC) [6, 7, 8, 9]. According to a report by IDC Futurescape, around 40% of IoT-generated data will be processed, stored, and acted upon close to the edge of network [10]. Edge computing (EC) enables analysis of information processing at the source of the data which sometimes is also referred to as in-network or on-board processing. EC not only reduces the huge workload of central com- puting servers (e.g. clouds), but also decreases the latency of data processing which includes the network latency for sending/receiving the required data plus the response time for performing the task on the cloud server. Figure 1 shows the hierarchical layers of computation in an IoT system [1]. As we move to the higher levels (i.e. from edge devices to the cloud servers), the processing capability increases. However, the latency would increase due to two factors: 1) network delay and 2) more workload on the servers. Therefore, the predictability for the real-time properties would decrease. With the proliferation of portable and mobile devices, and increasing processing capabilities of endpoints in IoT, 978-1-5090-4130-5/16/$31.00 c 2016 IEEE