As technologies related to sensor network are currently emerging and the use of GeoSensor is increasing along with the development of Internet of Things (IoT) technology, spatial query processing systems to efficiently process spatial sensor data are being actively studied. However, existing spatial query processing systems do not support a spatial-temporal data type and a spatial-temporal operator for processing spatialtemporal sensor data. Therefore, they are inadequate for processing spatial-temporal sensor data like GeoSensor. Accordingly, this paper developed a spatial-temporal query processing system, for efficient spatial-temporal query processing of spatial-temporal sensor data in a sensor network. Lastly, this paper verified the utility of System through a scenario, and proved that this system's performance is better than existing systems through performance assessment of performance time and memory usage.

For the Ubiquitous Sensor Network (USN) environment, which generally uses spatial as well as aspatial sensor data, a sensor database system to manage these data is essential. For this reason, sensor database systems such as TinyDB and Cougar are being developed by researchers. However, as most of these systems do not support spatial data types and spatial operators for managing spatial sensor data, they are not suitable for the USN environment. Therefore, in this paper, we design and implement Spatial TinyDB which is a spatial sensor database system that extends TinyDB to support spatial data types and spatial operators for the efficient management of spatial sensor data. In particular, Spatial TinyDB provides memory management and filtering functions to reduce system overload caused by sensor data streams. Finally, we prove that Spatial TinyDB is superior by comparing its actual performance, in terms of execution time, accuracy, and memory usage, with that of TinyDB.

Recently, the importance of big data has been emphasized with the development of smartphone, web/SNS. As a result, MapReduce, which can efficiently process big data, is receiving worldwide attention because of its excellent scalability and stability. Since big data has a large amount, fast creation speed, and various properties, it is more efficient to process big data summary information than big data itself. Wavelet histogram, which is a typical data summary information generation technique, can generate optimal data summary information that does not cause loss of information of original data. Therefore, a system applying a wavelet histogram generation technique based on MapReduce has been actively studied. However, existing research has a disadvantage in that the generation speed is slow because the wavelet histogram is generated through one or more MapReduce Jobs. And there is a high possibility that the error of the data restored by the wavelet histogram becomes large. However, since the wavelet histogram generation system based on the MapReduce developed in this paper generates the wavelet histogram through one MapReduce Job, the generation speed can be greatly increased. In addition, since the wavelet histogram is generated by adjusting the error boundary specified by the user, the error of the restored data can be adjusted from the wavelet histogram. Finally, we verified the efficiency of the wavelet histogram generation system developed in this paper through performance evaluation.

Research has been conducted to efficiently transfer blocks and reduce network costs when decoding and recovering data from an erasure coding-based distributed file system. Technologies using software-defined network (SDN) controllers can collect and more efficiently manage network data. However, the bandwidth depends dynamically on the number of data transmitted on the network, and the data transfer time is inefficient owing to the longer latency of existing routing paths when nodes and switches fail. We propose deep Q-network erasure coding (DQN-EC) to solve routing problems by converging erasure coding with DQN to learn dynamically changing network elements. Using the SDN controller, DQN-EC collects the status, number, and block size of nodes possessing stored blocks during erasure coding. The fat-tree network topology used for experimental evaluation collects elements of typical network packets, the bandwidth of the nodes and switches, and other information. The data collected undergo deep reinforcement learning to avoid node and switch failures and provide optimized routing paths by selecting switches that efficiently conduct block transfers. DQN-EC achieves a 2.5-times-faster block transmission time and 0.4-times-higher network throughput than open shortest path first (OSPF) routing algorithms. The bottleneck bandwidth and transmission link cost can be reduced, improving the recovery time approximately twofold.

Recently as efficient processing of aggregate queries for fetching desired data from sensors has been recognized as a crucial part, in-network aggregate query processing techniques are studied intensively in wireless sensor networks. Existing representative innetwork aggregate query processing techniques propose routing algorithms and data structures for processing aggregate queries. However, these aggregate query processing techniques have problems such as high energy consumption in sensor nodes, low accuracy of query processing results, and long query processing time. In order to solve these problems and to enhance the efficiency of aggregate query processing in wireless sensor networks, this paper proposes Bucket-based Parallel Aggregation (BPA). BPA divides a query region into several cells according to the distribution of sensor nodes and builds a quadtree, and then processes aggregate queries in parallel for each cell region according to routing. It sends data in duplicate by removing redundant data, which, in turn, enhances the accuracy of query processing results. Also, BPA uses a bucket-based data structure in aggregate query processing, and divides and conquers the bucket data structure adaptively according to the number of data in the bucket. In addition, BPA compresses data in order to reduce the size of data in the bucket and performs data transmission filtering when each sensor node sends data. Finally, in this paper, we prove its superiority through various experiments using sensor data.

Deep learning-based object detection technology can efficiently infer results by utilizing graphics processing units (GPU). However, when using general deep learning frameworks in embedded systems and mobile devices, processing functionality is limited. This allows deep learning frameworks such as TensorFlow-Lite (TF-Lite) and TensorRT (TRT) to be optimized for different hardware. Therefore, this paper introduces a performance inference method that fuses the Jetson monitoring tool with TensorFlow and TRT source code on the Nvidia Jetson AGX Xavier platform. In addition, central processing unit (CPU) utilization, GPU utilization, object accuracy, latency, and power consumption of the deep learning framework were compared and analyzed. The model is You Look Only Once Version4 (YOLOv4), and the dataset uses Common Objects in Context (COCO) and PASCAL Visual Object Classes (VOC). We confirmed that using TensorFlow results in high latency. We also confirmed that TensorFlow-TensorRT (TF-TRT) and TRT using Tensor Cores provide the most efficiency. However, it was confirmed that TF-Lite showed the lowest performance because it utilizes a GPU limited to mobile devices. Through this paper, we think that when developing deep learning-related object detection technology on the Nvidia Jetson platform or desktop environment, services and research can be efficiently conducted through measurement results.

Various techniques have been used in distributed file systems for data availability and stability. Typically, a method for storing data in a replication technique-based distributed file system is used, but due to the problem of space efficiency, an erasure-coding (EC) technique has been utilized more recently. The EC technique improves the space efficiency problem more than the replication technique does. However, the EC technique has various performance degradation factors, such as encoding and decoding and input and output (I/O) degradation. Thus, this study proposes a buffering and combining technique in which various I/O requests that occurred during encoding in an EC-based distributed file system are combined into one and processed. In addition, it proposes four recovery measures (disk input/output load distribution, random block layout, multi-thread-based parallel recovery, and matrix recycle technique) to distribute the disk input/output loads generated during decoding.

No abstract