With the recent development of mobile computing technologies, mobile terminals such as smartphones and tablet PCs are rapidly and explosively spread over the world. In wireless network technology, long-term evolution (LTE) services become popular, which enables wireless data communication faster than ever before. In the future, it is expected that information and contents services for mobile terminals will be diversifying, and that the resulting amount of data communications from and to mobile terminals will massively increase. In computing service, scale-out oriented cloud computing becomes popular, in which big data with the order of Tera/Peta-byte size is simultaneously processed in parallel by a large number of commodity servers.
In the future, it is expected that large-scale data processing and virtulization services over cloud computing environments are provided to a large number of mobile termininals. Expected application services are high-speed massive-data storage service, fast content-delivery service, streaming services of high-quality audio and high-definition TV, and interactive 3D communicatoin service. In such mobile-cloud information services environments, more sophisticated design of large-scale systems and more effective resource management are indispensable.
Large-Scale Systems Management Lab. research is aimed at developing mathematical modeling and simulation techniques for design, control and architecture of large-scale systems such as large-scale data centers and computer/communication networks, with which the resulting systems achieve high performance, low vulnerability and highly efficient energy saving. Our research focus is on network-science oriented design framework, fundamental technologies and highly-qualified services, in particularly for large-scale computer and/or communication network systems.
In cloud computing services, large-scale parallel data processing is realized with so-called scale-out computing environment with a huge number of commodity servers. In large-scale parallel data processing framework, a large-sized job task is divided into a number of small-sized subtasks, and each subtask is processed by its own worker machine, resulting in a high task throughput. When a large number of subtasks are processed with a large number of worker machines in a distributed computing manner, however, hardware failures and software malfunctioning are likely to occur among a not-small number of worker machines, making some subtask processing times extremely large. As a result, the overall task processing time is large and the task-level throughput is degraded significantly. This problem is widely known as the issue of stragglers. In the future, data centers with a huge number of server machines will be connected by each other with ultra high-speed data communication technology. In such a huge-scale data-center computing environment, more sophisticated computing resource management and more elaborate task scheduling are indispensable in order for both high task throughput and efficient energy saving. Here, we focus on scale-out cloud and mobile cloud environments. In order to realize ultimate scale-out cloud computing environment, we study efficient computing framework and energy-efficient task scheduling for processing big data with a huge number of worker machines. For mobile cloud computing, we study dynamic and elastic management schemes for computing and networking resources.
Autonomous Distributed Cognitive-Radio Networking
Recently, the explosive growth of smartphones and tablet devices causes the shortage of available frequency channels. One of the solutions for this spectrum shortage is the cognitive radio technology. In cognitive radio networks, radio devices recognize the surrounding radio spectrum environment and effectively use frequency channels without interference with other systems. This technology enables secondary users to use sufficient radio resources without any change of spectrum allocation for various wireless systems. Important and fundamental technologies for high-performance cognitive radio networks are dynamic spectrum access control, routing for multihop networking, and packet scheduling for guaranteeing quality of services (QoSs). We study and develop these key technologies in order to realize cognitive radio multihop networks in which spectrum resources are efficiently utilized in temporal and spatial senses, and the resulting overall throughput is maximized. With the technologies, fast content delivery service, high quality audio, and high definition TV can be supported for mobile cloud computing environments. The autonomous distributed cognitive-radio technologies realize base-station-free management of wireless channels, providing rapid recovery from communication infrastructure destruction due to crucial disasters such as earthquakes and floods.
System Analytics Based on Network Science
Network Science (also called Internet Science or Network Science of Complex Systems) is one of emerging interdisciplinary fields for characterizing the nature of information networks. Network Science is based on not only the conventioanl networking theories such as queueing theory and network optimization, but also mathematical and physical engineering, social economics, and cognitive sicence. In LSM laboratory, focusing on network modeling, network performance analytics, and designing methods for high-performance computer/communication networks, we study theoretical approaches based on stochastic analysis and probability models such as Markov chains and extreme value theory, scheduling algorithms for information packet flows, and high-speed simulation techniques for large-scale network systems.
With the recent development of informatin and communication technology (ICT), ``service'' provided by companies to customers is increasingly diversified. In general, the way of service is developed within a company, and the quality of service depends on empirical rules of thumbs of the company. Service science is an interdisciplinary approach to the design, implementation and improvement of service systems, whose ultimate goal is to create service innovation. Our approach to service science is based on not only conventioanl Operations Research and Management Science, but also the analysis of economical aspects of service, social-scientific analysis of the quality of service, and characterization of management quality. Currently, we are studying the cost-effective design and operator-management of call centers.
- System Analytics
- Mathematical modeling and analytics of large-scale complex systems such as cloud computing environments and network systems
- System Design
- Design, control and architecture for high-performance complex systems, queueing theory, Markov analysis
- Service Science
- Analytics, evaluation and implementation of high-quality service systems, operations research, management science
- Discrete-event simulation
- Modeling and high-speed simulation techniques for large-scale complex systems, Monte-Carlo simulation
- Online Algorithm
- Optimization problems and its algorithms for situations without future knowledge
- Mechanism Design
- Designing mechanism, auction theory, selfish routing