Recent advances in Internet-of-Things (IoT) technologies have sparked significant interest towards developing learning-based sensing applications on embedded edge devices. These efforts, however, are being challenged by the complexities of adapting to unforeseen conditions in an open-world environment, mainly due to the intensive computational and energy demands exceeding the capabilities of edge devices. In this paper, we propose OpenSense, an open-world time-series sensing framework for making inferences from time-series sensor data and achieving incremental learning on an embedded edge device with limited resources. The proposed framework is able to achieve two essential tasks, inference and incremental learning, eliminating the necessity for powerful cloud servers. In addition, to secure enough time for incremental learning and reduce energy consumption, we need to schedule sensing activities without missing any events in the environment. Therefore, we propose two dynamic sensor scheduling techniques: (i) a class-level period assignment scheduler that finds an appropriate sensing period for each inferred class, and (ii) a Q-learning-based scheduler that dynamically determines the sensing interval for each classification moment by learning the patterns of event classes. With this framework, we discuss the design choices made to ensure satisfactory learning performance and efficient resource usage. Experimental results demonstrate the ability of the system to incrementally adapt to unforeseen conditions and to efficiently schedule to run on a resource-constrained device.

The rapid advancement of IoT technologies has generated much interest in the development of learning-based sensing applications on embedded edge devices. However, these efforts are being challenged by the need to adapt to unforeseen conditions in an open-world environment. Updating a learning model suffers from the lack of training data as well as the high computational demand beyond that available on edge devices. In this paper, we propose an open-world time-series sensing framework for making inferences from time-series sensor data and achieving incremental learning on an embedded edge device with limited resources. The proposed framework is able to achieve two essential tasks, inference and learning, without requiring access to a powerful cloud server. We discuss the design choices made to ensure satisfactory learning performance and efficient resource usage. Experimental results demonstrate the ability of the system to incrementally adapt to unforeseen conditions and to effectively run on a resource-constrained device.

Thermal awareness is increasingly important for real-time systems deployed in harsh environments. As high chip temperature can cause frequency throttling or shutdown of processor cores at unexpected times, many real-time scheduling techniques have been developed to ensure continuous, fail-safe operation of safety-critical tasks with stringent timing constraints. However, their practical use remains largely limited due to the fact that it is extremely difficult to obtain a precise thermal model of commercial processors without using special measurement instruments or access to proprietary information, such as the power traces of micro-architectural units and detailed floorplans.In this paper, we propose a data-driven structured thermal modeling scheme that is directly applicable to commercial off-the-shelf multi-core processors used in real-time embedded systems. By using a small number of thermal profiles obtained from on-chip temperature sensors, our scheme can accurately predict the processor operating temperature under dynamic real-time workloads at various CPU frequencies and ambient conditions. The thermal model derived from our scheme is fast to converge and robust against different sources of errors. Our scheme is non-intrusive, meaning that it does not require changes to the software code or the hardware packaging of the target system. Furthermore, our scheme can estimate the relative power consumption of the processor for a given workload and clock frequency level. Experimental results from a multi-core ARM platform indicate that our scheme estimates the operating temperature with a maximum error of 2.5% while the latest prior work results in 23% error. This highly accurate modeling enables us to obtain the maximum achievable processor utilization that does not cause a thermal safety violation.

The rising demand for powerful embedded systems to support modern complex real-time applications signifies the on-chip temperature challenges. Heat conduction between CPU cores interferes in the execution time of tasks running on other cores. The violation of thermal constraints causes timing unpredictability to real-time tasks due to transient performance degradation or permanent system failure. Moreover, dynamic ambient temperature affects the operating temperature on multicore systems significantly.In this paper, we propose a thermal-aware server framework to safely upper-bound the maximum operating temperature of multi-core mixed-criticality systems. With the proposed analysis on the impact of ambient temperature, our framework manages mixed-criticality tasks to satisfy both thermal and timing requirements. We present techniques to find the maximum ambient temperature for each criticality level to guarantee the safe operating temperature bound. We also analyze the minimum time required for a criticality mode change from one level to another. The thermal properties of our framework have been evaluated on a commercial embedded platform. A case study with real-world mixed-critical applications demonstrates the effectiveness of our framework in bounding operating temperature under dynamic ambient temperature changes.

The recent trend in real-time applications raises the demand for powerful embedded systems with GPU-CPU integrated systems-on-chips (SoCs). This increased performance, however, comes at the cost of power consumption and resulting heat dissipation. Heat conduction interferes the execution time of tasks running on adjacent CPU and GPU cores. The violation of thermal constraints causes timing unpredictability to real-time tasks due transient performance degradation or permanent system failure. In this paper, we propose a thermal-aware server framework to safely upper bound the maximum temperature of GPU-CPU integrated systems running real-time sporadic tasks. Our framework supports variants of real-time server policies for CPU and GPU cores to satisfy both thermal and timing requirements. In addition, the framework incorporates two mechanisms, miscellaneous-operation-time reservation and pre-ordered scheduling of GPU requests, which significantly reduce task response time. We present analysis to design thermal-server budget and to check the schedulability of CPU-only and GPU-using sporadic tasks. The thermal properties of our framework have been evaluated on a commercial embedded platform. Experimental results with randomly-generated tasksets demonstrate the performance characteristics of our framework with different configurations.

Customers in a cloud would like to receive the results of their task as soon as possible while paying less. On the other hand, cloud providers aim to mitigate the operational cost of cloud environments. In other words, a limited budget makes providers create efficient cloud systems that utilize the computational powers of the clouds while minimizing their energy consumptions and environmental footprints. One of the prevalent techniques in mitigating the total energy consumptions of data-centers is through using consolidation of virtual machines (VMs). However, it incurs significant overheads on both computing resources and network infrastructure of a cloud. Furthermore, it causes tasks to be accomplished later or even it might lead to System Level Agreement (SLA) violations. To address the aforementioned challenges, we propose a cooperative two-tier task scheduling approach to benefit both cloud providers and their customers. It regulates the execution speeds of real-time tasks in a way that a host reaches the optimum level of utilization instead of migrating its tasks to other hosts. We also propose several predictive global task scheduling policies to map arrived tasks to feasible VMs. The simulation results show that the proposed task scheduling approach not only reduces the total energy consumption of a cloud by 41%, but also has profound impacts on turnaround times of real-time tasks by 85%.