UTokyo Repository 東京大学
 

UTokyo Repository >
124 情報理工学系研究科 >
40 電子情報学専攻 >
1244020 博士論文(電子情報学専攻) >

このページ(論文)をリンクする場合は次のURLを使用してください: http://hdl.handle.net/2261/43998

タイトル: Sensor Data Management and Transportation over Unreliable Networks
その他のタイトル: センサシステム構成法と非信頼性ネットワークにおけるデータ配送法
著者: Ochiai, Hideya
著者(別言語): 落合, 秀也
発行日: 2011年3月24日
抄録: Communication networks have enabled environmental monitoring and remote control with sensors and actuators. It has been applied to building automation and facility management for the last one decade and shown that they are promising technologies for Green by ICT. The differences of building automation and facility management are their targets and scopes. The target of building automation network is to autonomously control actuators based on observed events with sensors. The target of facility management network is to analyze the usage of facilities in buildings so as to improve the efficiency of energy usage. The recent rise of green-awareness in global society has attracted attentions to these technologies. However implementing them into actual buildings is not an easy mission but requires huge efforts. Only a few people or companies afford to pay their huge time and cost for this. The workload of system maintenance or update is also heavy. The system owners have to operate it for at least several decades: i.e., for the lifetime of the building. Programs of the system should be changed according to the changes of floors or other configurations though it is not always simple. Some system components become out-of-order in several years and need to be replaced. Those system components are not always available in the next one decade. The current development of a facility management system involves proprietary design and implementation. Engineers, with identifying the use cases, design the schema of databases and user interfaces, select suitable communication protocols, and implement software based on the designs and selections. The system components developed in the project cannot be directly used at other projects. The maintenance of the system involves the engineering of the software (e.g., internal design, programming and debugging) for the proprietarily specified data schema and communication protocols. There is another issue at the installation of building automation systems. We must install communication cables in a building to connect sensors and actuators. Recently, radio devices have been also considered for this purpose. However, wireless networks are still unreliable; they frequently become disconnected and lose packets, and to provide network connectivity for sensors and actuators over multiple network relays is challenging. Thus, currently, we only let sensors and actuators be wireless and few relays are used in the practical deployment. Wireless access range is limited, and cables must be installed in buildings to provide network connectivity. Based on those issues, we have setup the following two research challenges, in this thesis. ・Development of a common protocol stack for composing data-centric sensor actuator networks: i.e., facility management system. Data-centric sensor actuator networks focus on sensor data management rather than networking of sensors and actuators. They archive the historical records of sensors. The component that organizes data-centric sensor actuator networks must have the same protocol stack so that we can integrate them by component-and-flow programming style. This structure reduces engineering cost at the system integration phase. ・Development of data transportation scheme for unreliable networks. We must enable data transportation over intermittently-connected networks. Even if nodes are fixed, wireless links frequently become disconnected. Delay or disruption tolerant networking (DTN) has theoretically enabled message transportation over such links, but widely-studied message routing algorithms are still targeted at specific mobility models. The challenge is to design a routing algorithm that can be applied to both stable networks and totally random networks. In this thesis, we present our contributions in two parts. In the first part, we propose component and flow programming model for sensor data management, and design facility information access protocol (FIAP) and central controller-based device management (CCDM) architecture. In the latter part, we propose potential-based entropy adaptive routing (PEAR) and delay tolerant IP networking (DTIPN) as data transportation schemes over unreliable networks. In the research of FIAP, we identify (1) design pitfalls in data-centric sensor actuator networks, (2) challenges in the design of common protocol stack for implementing essential functional components, and (3) policies and algorithms for FIAP specification. We have developed FIAP protocol stack, implemented such components, and integrated into a data-centric sensor actuator network in Engineering Bldg.2 in the University of Tokyo. This experiment has shown that FIAP allows lightweight integration for wide-varieties of applications of facility management systems. In the research of CCDM, we propose a centralized architecture for managing hundreds or thousands of sensor and actuator devices, and their dataflows. CCDM allows the integration and configuration of large number of components in a centralized terminal (or window). This also helps the integration of FIAP components into a facility management system. In our experiment, CCDM has shown its capability of computing moderately complex placement algorithms in the traffic optimization case. We consider that this capability makes other optimizations feasible, such as delivery latency, load-balancing, and fault-tolerance. In the research of PEAR, we propose a reliable communication framework for wide varieties of mobility models from stable networks to totally random networks. PEAR changes message delivery patterns according to the complexity of the topology changes. If the contact patterns among node are somehow related, PEAR chooses almost the best path. If the contact patterns become complex and estimation of contacts become meaningless, PEAR replicates messages in the network and maintains delivery probability. We carried out both simulation-based and prototype-based experiments to validate the behavior of PEAR. We evaluated delivery probability, latency, total transmissions and buffer usage. Though the delay increased according to the hop count, PEAR could achieve 100% message delivery in sensor data transportation. This result indicates that PEAR is quite adaptive to various mobility models and provides reliable communication framework compared to previously proposed routing schemes. We also demonstrate in this thesis that wireless links are totally intermittent even if nodes are physically stable. We carried out experiments with 50 wireless nodes in Engineering Bldg. 2 and Hongo campus, in the University of Tokyo. It has also shown that hop-by-hop transfer and parallel message propagation, which PEAR takes, enables scalable message propagation in hop count and increases message propagation speed. In the research of DTIPN, we propose an alternative architecture of DTN that seamlessly connects intermittently-connected networks to the existing Internet. IP packets can be delayed not only for seconds but also for hours and days. By making use of this fact, we demonstrate that sensor nodes in isolated networks can still send message to the Internet host with the Internet protocol.
内容記述: 報告番号: 甲27284 ; 学位授与年月日: 2011-03-24 ; 学位の種別: 課程博士 ; 学位の種類: 博士(情報理工学) ; 学位記番号: 博情第322号 ; 研究科・専攻: 情報理工学系研究科電子情報学専攻
URI: http://hdl.handle.net/2261/43998
出現カテゴリ:021 博士論文
1244020 博士論文(電子情報学専攻)

この論文のファイル:

ファイル 記述 サイズフォーマット
48087401.pdf8.04 MBAdobe PDF見る/開く

本リポジトリに保管されているアイテムはすべて著作権により保護されています。

 

Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - ご意見をお寄せください