Susumu Shibusawa

Scheduling algorithms for efficient gather operations in distributed heterogeneous systems

Advances in networks and the availability of low-cost PCs and workstations has resulted in PC/workstation clusters being used as a computer environment that replaces parallel computers. These clusters recently comprise different kinds of nodes and are connected over different kinds of networks, such as Ethernet and Myrinet. They are called distributed heterogeneous systems, and efficient collective communication operations for them have been studied. We propose heuristic scheduling algorithms for efficient gather operations in distributed heterogeneous systems, and we present the results of simulations evaluating their performance. These results show that gather operations scheduled by these algorithms are more efficient than sequential gather operations in these systems. They also show that, the efficiency of gather operations scheduled by the algorithms depends on the ratio of the cost that depends on message size in point-to-point communication to the cost that does not depend on message size in point-to-point communication.

Scatter and gather operations on an asynchronous communication model

Collective communication such as scatter, gather, broadcast, and total exchange plays an important role on the performance of parallel and distributed processing. So far, several parallel collective communication algorithms have been studied in synchronous comnmnication models, but it is necessary to design and anMyze parallel algorithms using a more realistic communication model. This paper introduces an asynchronous communication model between processors, and evaluates the execution time of operations which include scatter, gather and some computations for data initially located at one processor. The results show that the execution time in sending data in a single direction is less than that of dual directions for asynchronous communication. This paper also introduces the logarithmic correction of the bisection communication tree, and evaluates the communication time of several instances.

Signature amortization using multiple connected chains

Amortization schemes for authenticating streamed data have been introduced as a solution to reduce the high overhead that sign-each schemes suffer from. The hash chains structure of amortization schemes and the number of hash values appended to other packets affect the efficiency of the authentication scheme specially against packet loss. Which packets should have hashes appended to the signature packet and how many hashes to append to it have no solutions yet. This paper introduces a new hash chain construction to achieve longer resistance against packet loss and reduces the overhead. The proposed scheme consists of multiple connected chains, each chain links several packets together. Our scheme specifies clearly how to choose the packets that should have hashes appended to a signature packet, in addition to deriving their loss probability. We study the effect of the number of hashes that are appended to a signature packet on the overhead. We introduce a measure so as to know the number of packets receivers need to buffer until they can authenticate the received packets. The number of chains of our model plays a main role in the efficiency of our scheme in terms of loss resistance and overhead.

A file replication method based on demand forecasting in P2P networks

In peer-to-peer (P2P) networks that support file-sharing services, the level of access demand can vary widely between different files. Since fewer nodes store files for which there is a lower demand, these files are more likely to be lost from the P2P networks if the users leave the network or delete the files. The loss of files can cause users to seek alternatives to P2P services, and leads to the degradation in service quality. In this study we propose and evaluate a replication method considering a service quality that aims to prevent the loss of low-demand files. In this method, the number of file replicas to be placed is determined based on the forecast demand for the file, so that the loss of low-demand files is likely to be prevented by placing replicas at nodes that frequently use P2P services. Based on the result of a simulation, we compared our proposed method with basic replication methods in terms of the amount of storage used, and the number of files. Our experimental results show that the proposed method prevents the loss of files by preserving low-demand files over extended periods of time. We also confirmed that the node storage resources consumed by this method are efficiently used.

Motion Detection and Evaluation of Chair Exercise Support System with Depth Image Sensor

Sustaining lower limb functionality is extremely important in the preventative care of the elderly. Chair exercise, in which the exerciser sits on an ordinary chair, offers a way for seniors with little physical strength to exercise without a great deal of effort. Meanwhile, Microsofts Kinect sensor that is capable of detecting human motion without the subject having to wear any kind of a special marker are becoming widely available. Exploiting this new sensor technology, this paper describes the development and deployment of a prototype lower limb chair exercise support system that takes full advantage of the Kinect depth sensor. The system supports five different chair exercises designed to strengthen the lower limbs, recognizes and evaluates exercises based on 3D position data and joint angles for each joint obtained from the Kinect sensor. The system illustrates how to do the exercises by voice instructions and model images, and superimposes the muscles used onto an image of the exerciser in real time. The system also provides exercise assessment results and advice by voice and text. In a series of trials involving seven elderly subjects in their late 70s and early 80s, an overall average recognition rate of 89% was obtained for the five exercises. Feedback was obtained through a questionnaire given to five male subjects ranging in age from 50 to 65, which highlighted a number of issues that should be addressed to improve the effectiveness of the system.

Design of Lower Limb Chair Exercise Support System with Depth Sensor

Sustaining lower limb functionality is extremely important in the preventative care of the elderly. Chair exercise, in which the exerciser sits on an ordinary chair, offers a way for senioars with little physical strength to exercise without a great deal of effort. Meanwhile, Microsofts Kinect sensor that is capable of detecting human motion without the subject having to wear any kind of a special marker are becoming widely available. Exploiting this new sensor technology, this paper describes the development and deployment of a prototype lower limb chair exercise support system that takes full advantage of the Kinect depth sensor. The system supports five different chair exercises designed to strengthen the lower limbs, recognizes and evaluates exercises based on 3D position data and joint angles for each joint obtained from the Kinect sensor. The system illustrates how to do the exercises by voice instructions and model images, and superimposes the muscles used onto an image of the exerciser in real time. The system also provides exercise assessment results and advice by voice and text. In a series of trials involving seven elderly subjects in their late 70s and early 80s, an overall average recognition rate of 89% was obtained for the five exercises.

A Tabletop System Using Infrared Image Recognition for Multi-user Identification

Many tabletop systems have been developed to facilitate face-to-face collaboration and work at small meetings. These systems often require users to attach sensors to their bodies to identify their positions, but attaching a sensor to one’s body can be bothersome and annoying, and user position and posture may be restricted depending on where the sensor is attached. We have proposed a technique for estimating user position in a tabletop system by image recognition and implemented a tabletop system having a user position identification function incorporating the proposed technique. This technique first obtains touch points and hand-area information from touch operations performed by the user, and establishes an association between the touch points and hand from those positional relationships. Since the direction in which a hand is extended can be derived from that hand’s touch information, the position of the user of the touch points belonging to that hand can be estimated. As part of this study, we also implemented a photo-object manipulation application, which has a function for orienting a photo object to face the user based on the results of the above user-position estimation technique. We performed an experiment to evaluate the position identification rate, and found that the proposed technique could identify user position with high accuracy.

A Walkthrough System with Improved Map Projection Panoramas from Omni Directional Images

Virtual walkthrough system can be used for monitoring some areas, introducing view spots or learning history. Most virtual walkthrough systems use prepared images or 3D virtual objects to create the virtual environment and it is hard for the walkthrough system to be operated in real time. In some situations, real time walkthrough is important, and it is possible to construct a real time walkthrough system using an omni-directional camera which takes a 360-degree image at a time. In this study, we create an interactive walkthrough system with improved map projection panoramas from omni-directional images and evaluate the system. With improved map projection method, we acquire better-quality and more friendly panoramas than with original map projection method. The walkthrough system consists of an omni-directional camera, a server PC and client PCs. The server PC sends panoramas to clients, and the clients construct the virtual environments for walkthrough with panoramas and other prepared image and audio files. We evaluate this system by the effect of image projection methods, functions of walkthrough system and users’ answers to a questionnaire.

A Handy Real-Time Location System for Community Bus Service

We present a handy information system for community bus services. Its main feature is that we have only togo around a new route as a test drive with a terminal device in order to collect and publish its geographic data such asthe positions of the stops in the Googles general transit feedspecification (GTFS) format. A template time table is recorded during the test drive which can be edited to complete a timetable for publication in the GTFS format. From the bus onduty, its real-time position is uploaded in the GTFS realtime format. Those data can be accessible by the passengers through Google. It has been planned and yet to be implemented thatvoice announcements can be recorded during the test drive withthe location data so that the announcements can be played back automatically.

Performance evaluation of distributed algorithms for mining association rules on workstation cluster

The mining of association rules is one of the database mining techniques used to extract useful information from large quantities of data. Finding association rules, however, requires that the transaction database be scanned repeatedly, and we need to handle very large amounts of transaction data. This requires an incredibly large amount of computation time. There have therefore been many attempts to speed-up database mining by using parallel computers. Recent improvements in the performance of PCs and workstations (WSs), and the recent dissemination of network technology have made parallel processing distributed within computer clusters an attractive alternative to parallel computers. In this paper we describe two new algorithms effective for parallel processing distributed in a WS cluster environment. One is an algorithm in which the size of data transmitted between WSs is smaller than that of the former algorithm. The other is an algorithm that reduces the amount of scan processing at each node by dividing data and that uses shift operations for data communication. We have implemented these algorithms on a WS cluster and have evaluated their performance.