The increase of Internet of Things devices and the rise of more computationally intense applications presents challenges for future Internet of Things architectures. We envision a future in which edge, fog, and cloud devices work together to execute future applications. Because the entire application cannot run on smaller edge or fog devices, we will need to split the application into smaller application components. These application components will send event messages to each other to create a single application from multiple application components. The execution location of the application components can be optimized to minimize the resource consumption. In this paper, we describe the Distributed Uniform Stream (DUST) framework that creates an abstraction between the application components and the middleware which is required to make the execution location transparent to the application component. We describe a real-world application that uses the DUST framework for platform transparency. Next to the DUST framework, we also describe the distributed DUST Coordinator, which will optimize the resource consumption by moving the application components to a different execution location. The coordinators will use an adapted version of the Contract Net Protocol to find local minima in resource consumption.

This paper presents low-cost laboratory which has been designed and developed to enhance learning experience and help students gain skills and knowledge in the field of distributed systems. In order to build a comprehensive distributed file system, we used the laboratory consisted of 40 card-sized Raspberry Pi devices, with the accent on stability, scalability, and its low-cost. Aiming to assess the impact of this new learning environment on the learning process and its outcomes, we surveyed students following the completion of three project stages during the 17 laboratory exercises in one academic year, assuring that we maintained the same subjects of study during the experiments. Supported by interesting answers on various set of questions, we provide a valuable insight into students' experience, obstacles and observations during system's implementation. This particular insight paves the way toward: 1. further laboratory's improvement, 2. adopting this approach in other courses related to ours, 3. encouraging teachers to embrace similar practice regardless of type of education field.

Recent advances in the field of Neural Architecture Search (NAS) have made it possible to develop state-of-the-art deep learning systems without requiring extensive human expertise and hyperparameter tuning. In most previous research, little concern was given to the resources required to run the generated systems. In this paper, we present an improvement on a recent NAS method, Efficient Neural Architecture Search (ENAS). We adapt ENAS to not only take into account the network's performance, but also various constraints that would allow these networks to be ported to embedded devices. Our results show ENAS' ability to comply with these added constraints. In order to show the efficacy of our system, we demonstrate it by designing a Recurrent Neural Network that predicts words as they are spoken, and meets the constraints set out for operation on an embedded device, along with a Convolutional Neural Network, capable of classifying 32x32 RGB images at a rate of 1 FPS on an embedded device.

No abstract

No abstract