Park Joonsuu, Park KeeHyun
Department of Computer Engineering, Keimyung University, Deagu 42601, Korea.
Sensors (Basel). 2020 Mar 2;20(5):1364. doi: 10.3390/s20051364.
Internet of Things (IoT) technologies are undeniably already all around us, as we stand at the cusp of the next generation of IoT technologies. Indeed, the next-generation of IoT technologies are evolving before IoT technologies have been fully adopted, and smart dust IoT technology is one such example. The concept of smart dust IoT technology, which features very small devices with low computing power, is a revolutionary and innovative concept that enables many things that were previously unimaginable, but at the same time creates unresolved problems. One of the biggest problems is the bottlenecks in data transmission that can be caused by this large number of devices. The bottleneck problem was solved with the Dual Plane Development Kit (DPDK) architecture. However, the DPDK solution created an unexpected new problem, which is called the mixed packet problem. The mixed packet problem, which occurs when a large number of data packets and control packets mix and change at a rapid rate, can slow a system significantly. In this paper, we propose a dynamic partitioning algorithm that solves the mixed packet problem by physically separating the planes and using a learning algorithm to determine the ratio of separated planes. In addition, we propose a training data model eXtended Permuted Frame (XPF) that innovatively increases the number of training data to reflect the packet characteristics of the system. By solving the mixed packet problem in this way, it was found that the proposed dynamic partitioning algorithm performed about 72% better than the general DPDK environment, and 88% closer to the ideal environment.
物联网（IoT）技术无疑已经在我们身边，因为我们正处于下一代物联网技术的风口浪尖。事实上，下一代物联网技术在物联网技术尚未被完全采用之前就已经在不断发展，智能尘埃物联网技术就是这样一个例子。智能尘埃物联网技术的概念以具有低计算能力的非常小的设备为特色，是一个革命性的创新概念，它使许多以前无法想象的事情成为可能，但同时也产生了一些尚未解决的问题。最大的问题之一是大量设备可能导致的数据传输瓶颈。双平面开发套件（DPDK）架构解决了瓶颈问题。然而，DPDK解决方案产生了一个意想不到的新问题，即所谓的混合数据包问题。混合数据包问题发生在大量数据包和控制数据包快速混合并变化时，会显著减慢系统速度。在本文中，我们提出了一种动态分区算法，通过物理分离平面并使用学习算法来确定分离平面的比例来解决混合数据包问题。此外，我们提出了一种训练数据模型扩展置换帧（XPF），它创新性地增加了训练数据的数量，以反映系统的数据包特征。通过这种方式解决混合数据包问题，发现所提出的动态分区算法的性能比一般的DPDK环境大约好72%，并且比理想环境接近88%。
