Published papers

This paper presents a new framework for constructing flexible event-driven systems for distributed networked devices and appliances, referred to as Dragon. Dragon enables constructing and modifying an event-driven system easily by visualizing distributed services which are handled in an integrated manner. The update of the system is conducted immediately following a modification by Dragon. Dragon is built on to of Jini/Java and can be executed on any Jini/Java-compliant machines. In addition, Dragon incorporates soft real-time event delivery to enhance the predictability of behavior of the constructed system. In this paper, we describe in detail of design and implementation of Dragon.

In this paper we propose a scheme of rate control for continuous UDP flows with moderate fairness. First, per-packet Relative One- Way Trip Time is investigated extensively and it is found that spikes with successive plots often appear on a time-ROTT graph. In addition, congestion-related losses are found to be strongly correlated only to the spikes. Based on the observation, we propose a rate control with moderate fairness using path status.

A key challenge in packet scheduling over a wireless channel is to achieve effective throughput against bursty errors. Although there have been several proposals on packet scheduling for a wireless channel, they are only based on bi-modal channel states: good or bad. In this paper we propose a new packet scheduling scheme for wireless links based on the receiver's signal to noise ratio (SNR). The scheme is referred to as SNR-based packet scheduling (SPS). In SPS, outgoing packet flows are scheduled such that the flows with higher SNR values have larger weights in scheduling. SPS can achieve effectiveness by adopting a flexible mapping between weight and SNR. We present measured results of the relationship between throughput and SNR as well as the design and implementation of SPS. In our experiments, SPS has shown its effectiveness in enhancing total throughput for hosts with good and bad wireless connectivity over a wireless LAN.

本論文では，「間に合わせ」手法に基づいてアプリケーションの動的構成を支援するためのツールキットであるポゼッションウェアの設計・実装について述べる．ポゼッションウェアは，遍在するデバイスやソフトウェア部品を抽象化したBodyコンポーネントと，それらを統合してアプリケーションを構成する実体であるSoulコンポーネント，および基盤システムから成る．Soulコンポーネントが，基盤システムを通じてディレクトリサービスやコンポーネントレポジトリと通信することで，ユーザの近隣に遍在する部品を探索・選択し，アプリケーションを動的に構成する．本論文では，ポゼッションウェアにおけるアプリケーション構成の基本となる「間に合わせ」手法について述べ，本手法に基づくアプリケーション構成例のいくつかを応用事例として示す．

Many audio and video stream applications in the Internet require good network performance but not perfect reliability. Such applications should employ congestion control and should not shut down the TCP flows by restricting buffering too much. However, the Internet is not a small local community, and it is impossible to rely on all developers to incorporate end-to-end congestion control in their Internet applications. The network should take part in controlling its own resource utilization. Most research for fair queuing tend to focus on reducing processing overhead and achieving exact fairness for each flow. In this paper we claim that perfect fair queuing is not effective for end users in the current complicated network environment. We propose Soft Fair Flow Control(SF-FC) which is a light weight router mechanism that employs simple FIFO queuing, and shows high performance than other fair queuing. SF-FC provides fair network resource allocation to non-adaptive and adaptive flows. Each SF-FC router informs the upstream SF-FC router of its link state. Based on the received state, SF-FC dynamically sets the packet discard rate of the non-adaptive flows. '

This paper presents a new architecture for integrating Networked Appliances, called Virtual Network Appliance(VNA). Instead of merely connecting appliances seen in existing approaches, VNA architecture enables users to create a networked appliance by assembling functional components of existing appliances. Users can achieve dynamic adaptation and personal customization, exploiting VNA.

The objective of detecting the change in one-way delay is to capture a congestion signal that should be reflected to a rate control of an RTP/RTCP-type unicast stream communication. Although a packet of the RTP flow may not be lost, the delay of the flow will increase due to an increase in queuing delay at the link. Therefore, only a change in delay instead of the absolute value of delay should be obtained for the rate control. Measuring the change in delay is also significant for a path that includes an error-prone wireless link. In such a wireless link,some packet losses may be irrelevant to congestion. Therefore, congestion control depending on packet loss ratio alone results in an erroneously low rate. In this paper, we describe the design and the implementation of Estimation of Skew with Reduced Samples (ESRS). We have measured changes in delay using the scheme over a path between Keio University (Japan) and Carnegie Mellon University (U.S.), and a path between Keio University and CESAT (Spain). The variation in one-way delay obtained with ESRS is shown to include the path status of the flow.

This paper proposes a new load-sensitive handover scheme called LSH to manage wireless LANs. Future wireless LANs will is difficult to guarantee the network quality in current wireless LANs because of poor network resources including narrow bandwidth, high latency and high error rate. Under these conditions, mobile station should share resources efficiently. LSH utilizes information about the network load. By quantitative measurement and update messages, each base station maintains the cell information including a number of mobile stations, transmit/receive throughput and loss rate. Mobile stations in the cell overlapped area can select the most suitable cell by using cell information. It works to balance the network load of each cell.

This paper presents a new framework for constructing applications for wearable computers. We do not consider wearable computing merely as a single self-confined system. Rather, wearable computing can be treated as a cooperative application with information surrounding a person and various devices attached to him/her. We have developed a frameworkof such wearable computing called ``Wapplet.'' Wapplet provides adaptation to the availability of devices as well as a systematic scheme of constructing applications for wearable computing. In this paper, we describe the design and implementation of Wapplet, and show its prototype system.

Mobile agent software architecture has attracted a great deal of attention as a basic technology for building such a ubiquitous computing environment. In our framework, a mobile agent is divided into two parts: a core component and a set of add-on functional modules. Each runtime system dynamically collects add-on modules necessary for a new environment, and links them with the core component. This feature allows us to describe an application as seamless against changes of computing environment. For example, we can realize the following application behavior. When the user changed his location, his applications also move to a device that he personalized as though they followed him. Then, they adapt their contexts to the new location and continue to carry out their tasks. On a device that has a small LCD, an application can indicate its progress to a user progress. On another device that has an input interface such as a touch-panel, the application may wait for instructions from the user. We implemented mobile agent systems based on this framework on actual PCs and small, networked devices. Especially, one of the m-P@gent runtime environments, @TINI, designed for the TINI devices provides X-10 protocol libraries. An ‘agent @TINI’ can control appliances connected over X-10 network by using X-10 commands.

It is important for application programmers and users to easily control networked appliances and devices without the need to be aware of their locations. In particular, network transparent control is essential for the ubiquitous computing environment. This is because end-users accomplish their tasks by coordinating devices that they are wearing, such as wearable computers and cellular phones, with various networked appliances, such as display devices and loud speakers. This paper presents the Wapplet framework, a new application framework that addresses this issue for the ubiquitous computing environment. To show the efficiency of the proposed Wapplet framework, we have evaluated it in terms of processing time. Requirements of the Wapplet framework are as follows: (a) support for object-migration, (b) location transparent method of invocation, and c abstraction of device specifics. Object migration is necessary for the ubiquitous computing environment because it is impractical to pre-install all applications on networked appliances. Thus, applications should be composed of multiple mobile modules. Location transparent control mechanisms like RPC (Remote Procedure Call) help application programmers in coordinating networked appliances. The framework needs to provide device independent Application Program Interfaces (APIs) that define interfaces to networked appliances, which eliminates the need to change device control programs for every appliance. In our implementation of the Wapplet framework we satisfied the above requirements using Java 1.2.2 on FreeBSD-4.2RELEASE as follows. Our object migration system (a) is realized by Java's Serialization mechanism. The system supports multi hop migration of mobile objects, which adapts to changes in available devices caused by users' movement. As for location transparent control (b), we developed a location transparent method invocation (LTMI) protocol that works like Java RMI or Sun RPC. Most conventional protocol, such as Java RMI, forces application programmers to manage the locations of mobile objects. We developed LTMI so application programmers only need to invoke the method name of the object but do not need to know the object's location (e.g. IP addresses). In addition, we provide device independent APIs to easily control and manage various kinds of networked appliances. The device control APIs classify networked appliances by media data types. There are four media types: for text, sound, image, and video. To evaluate the efficiency of LTMI, we measured the processing time of LTMI in comparison with Java RMI. As the result, we found that LTMI is more efficient than Java RMI in both local and remote method invocations.

In this paper, we propose a proactive operational model for home appliances. Recently, users of home appliances often have difficulty in operating appliances due to their high functionality, complex interface, which have many buttons, and users’ lack of knowledge. The current operational model of home appliances is a reactive model, which it is necessary for users to notify appliances of operational information. This causes an increased burden on the user to understanding correct operational information. To cope with these issues we propose a proactive operational model. In this model, appliances acquire the operational information by themselves, and provide users with their services based on that information. Thus, our model eliminates the users’ burden on explicitly expressing requests to the appliances to access their services. It is necessary for programmers to satisfy the autonomy of obtaining information, interaction between users and system, and operation simplification. Based on this model, we have developed a middleware, called Proactive Support System for Networked Appliances (PRONA). PRONA is realized by acquiring information necessary for appliance operation and learning the users’ operational pattern. PRONA consists of three modules, which are the information acquisition module, the suggestion composition module, and the information manager module. The information acquisition module normalizes information gathered from sensors and appliances, and transmits it to the information manager module. The suggestion composition module learns the importance of related sensors when users operate the appliances. We represent the importance of related sensors as a weight, and transmit the weighted data to the information manager module. The suggestion composition module also provides the user suggestions from this weighted data. The information manager module stores this data in its database. We evaluated the use of PRONA in a proactive television application. PRONA’s quantitative evaluation criteria are its processing time, the accuracy of its suggestions for user’ action, and its learning time for the user’ operational pattern while varying the number of proactive components.

This paper presents a new middleware architecture for integrating heterogeneous information appliances on heterogeneous network for various users, called Virtual Network Appliance (VNA) Architecture. Instead of merely connecting appliances seen in existing approaches, VNA architecture accommodates creating a function alliance called a VNA. VNA Markup Language is our solution to define VNAs independent of specific home networks. Our middleware enables users not only to define VNAs by themselves (bottom-up integration), but also to virtually plug and play pre-defined VNAs (top-down integration).

This paper proposes a new transport protocol for home networks called Light-weight Command Transport Protocol (LCTP).

In this paper we describe an experimental method of measuring per-packet Relative One-way Trip Time (ROTT) of a UDP stream at a receiver on the Internet as well as its experimental results.

This paper presents 'OR2'', an adaptive path tuning scheme for mobile ad hoc networks. In OR2, the active path adapts dynamically to node mobility without any link failures based on the local link quality. Most conventional routing protocols accommodate the change of network topology only when the link fails. Unless the movement of intermediate nodes leads to any link failures, they cannot adapt to the network topology even if other routes with less hop count become available. In contrast to these protocols, OR2 reduces the hop count of an active route while data packets are sent without link disconnection. Specifically, by introducing the notion of proximity that indicates the nearness of two communicating nodes, OR2 skips the upstream node in a proximity area and continues to shorten an active route as possible. For delay sensitive TCP flows, reducing the number of hops and path delay significantly improves the performance of TCP. We have implemented OR2 as an extension to DSR on FreeBSD. The experimental results have shown that OR2 is effective in enhancing TCP throughput and reducing end-to-end delay for all relevant flows.

This paper presents a new framework for constructing flexible event-driven systems for distributed networked devices and appliances, referred to as Dragon. Dragon enables constructing and modifying an event-driven system easily by visualizing distributed services which are handled in an integrated manner. The update of the system is conducted immediately following a modification by Dragon. Dragon is built on Jini/Java and can be executed on any Jini/Java-compliant machines. In addition, Dragon incorporates soft real-time event delivery to enhance the predictability of behavior of the constructed system and composition of distributed services for event management.In this paper, we describe the details of design and implementation of Dragon.

This paper proposes a new framework m-P@gent for environment-aware mobile agents that can run even on resource-limited devices and appliances. In the framework, a mobile agent is divided into a two parts: a core and addon functional modules. Each runtime system dynamically collects add-on modules necessary for a new environment, and links them with the core component. This feature allows us to describe an application seamless against changes of computing environment. We implements mobile agent systems based on the framework on actual PC environments and small PDA devices, and evaluates their basic performance. Our system makes two contributions. One is that it shows a possibility for expanding the field of mobile agent to resource-limited devices. Another is it reduces the cost of an agent migration. The framework will be a core software infrastructure for building ubiquitous computing environment.

In forthcoming home network environment, computation capability will be embedded invisibly in home appliances, sensors, walls, ceilings, and floors. People will conduct various tasks using multiple devices simultaneously without consciousness of using computers. In this paper, first, we propose an application model named Virtual Network Appliance (VNA) model which simplify and expand device utilization. In the model, each device has VNA runtime system and function objects, called VNA components, running on it. A user task is defined in an application called VNA which is a logical appliance consisting of abstract function requirements and a message graph among them. Second, we propose Virtual Plug&Play mechanism which is a dynamic service integration mechanism in VNA model implementation. When a user conducts a task, he/she makes a VNA runtime system on a user-side terminal load a VNA definition appropriate for the task. Virtual Plug&Play dynamically discovers required VNA components and establishes the message graph as defined. Since XML documents are used to describe a VNA, users can share and customize it easily. We call the device integration done by Virtual Plug&Play top-down integration, which existing middleware do not aim at. Finally, we show that Virtual Plug&Play affords practical performance for top-down integration by performance evaluation.

This paper proposes a new practical packet scheduling scheme for wireless LANs called SoCPS and also explores extensively the effectiveness of wireless packet scheduling under a variety of conditions over an actual wireless LAN. One well-known aspect in packet scheduling at the base station is consideration of channel state dependency. In conventional studies, the channel state is expressed in microscopic two states in the Markovian sense. To eliminate a large overhead associated with acquiring such a fine-grain channel state and to establish a software-based enhancement, we introduce the notion of Strength of Connection (SoC) which can be expressed as the long-term strength of the connectivity between the base station and an end host. We examine schemes for identifying the channel state and examine the applicability to packet scheduling. We also design a packet scheduling scheme based on SoC called SoCPS. Finally, we investigate the performance under various conditions of traffic and wireless channel states. We have designed and implemented SoCPS on FreeBSD computers. Our evaluation results obtained with up to eight wireless-LAN nodes have determined the applicability of SoCPS and limitation in wireless packet scheduling for wireless LANs. In particular, while FIFO is sufficient in many cases where TCP traffic is dominant, scheduling with SoC consideration is effective for real-time UDP traffic.