Vol.2 No.1 2009

44/88

Research paper : How Grid enables E-Science? (Y. Tanaka)−41−Synthesiology - English edition Vol.2 No.1 (2009) procedures (process flows) for a large quantity of data. To obtain required data quickly when conducting multiple simulations, such as in seismic vibration analysis and liquefaction projection during the occurrence of earthquake, or in flood projection as water levels rise, it is desirable to construct a workflow for the procedures. The preceding workflow studies have already been done for the grid, and the introduction and construction of workflows is necessary for the GEO Grid.(4) High performance processingLarge-scale image data processing requires large-scale computation resource, and existing software may consume several minutes to several tens of minutes for image processing. Considering interactive data transmission, it is desirable that image production takes only a couple of minutes, at most. Recently, multi-core architecture suitable for image processing, like CELL/B.E.TM is beginning to be used, and high speed capacity for image processing and simulation, utilizing the latest architecture, can be expected in the GEO Grid.(5) Development of a metaschedulerIn the system constructed in this research, the computation services for simulation and image processing are provided from a single site. In the future, when the same services are provided from multiple sites, it is necessary to develop a registry that manages where data is located and what kind of services are available, i.e., a monitoring system that checks the usage status of the computation server that provides the system, and a metascheduler that selects “seemingly best” services based on that information.Research is being done on all five issues, and issues (1) to (3) can be addressed in about one or two years. For issue (4), it is necessary to increase the speed capability of the software, and in many cases, the source program is not provided due to licensing conventions for commercial software. However, preliminary assessments have been conducted for increasing the speed of image processing software on CELL/B.E.TM, and there is a possibility for significant improvement. Issue (5) is the biggest issue for the grid. To realize the concept of the grid – a world “where service can be available without thinking about which resource is used when the computer is connected to the network,” a metascheduler function is mandatory. But it is extremely difficult to select the “optimal” resource in a complex environment where the decision of what is “optimal” is based on the characteristics of the computation (such as the ratio of quantities of communication to computation) and where the resources (network and compute server) that compose the grid changes dynamically in structure and efficacy. Research will continue to try to solve this issue by setting a limit on the range of criteria that will not interfere with the user in the scenario of the GEO Grid.7 How to proceed with subsequent R&DIn this study, the implementation of the system was considered by studying the demands and ideas from the Institute of Geology and Geoinformation, the Geological Survey of Japan, the Research Institute for Environmental Management Technology, and the Grid Technology Research Center (currently, a part of the Information Technology Research Institute), as a transdisciplinary research project of AIST. The Grid Technology Research Center strengthened the organization to help advance the R&D effort for the GEO Grid by shifting personnel from the GSJ to the Center; by employing researchers in application fields of the National Research Institute for Earth Science and Disaster Prevention, the Japan Aerospace Exploration Agency, and the National Institute for Environmental Studies; and by close discussions between the researchers in the information and application fields.GEO Grid management meetings were held regularly to discuss the policy of the GEO Grid, relationships with external organizations, the clarification of pressing issues and measures, as well as for progress management. To ensure the smooth operation of a large-scale project with total of about 20 people from the application and information fields from the Grid Center alone, the application and information fields regularly held meetings for progress management, to pinpoint problems and to discuss the solutions, and great effort was made to share information on the issues among the research team members. It was important for the researchers of both fields to work at the same location, and to be in an environment where frequent discussions and brain-stormings were possible, other than just at the regular meetings.For this research, several middleware suites, developed not just by AIST but also by overseas institutions, were used. Basically each middleware suite is implemented through the employment of a standard protocol and interface, but there were some problems due to defective implementation and a lack of function in the middleware when actual tests were conducted. Since there was close collaboration among the developers of each middleware suite through the Open Grid Forum[27], a grid standardization organization, and through international conferences, it was possible to ask the developers to take immediate measures when issues arose. Also, it was possible to have them incorporate new functions in the next version by communicating the required specifications, and they, in turn, provided advice for specific implementation methods. In many cases, a grid system is realized by linking an extremely large number of middleware suites, and it is not practical in terms of development cost to implement everything on our own, and therefore it is important to engage in daily communication with other developers to create a system where R&D can be conducted with cooperation from overseas organizations, while

※このページを正しく表示するにはFlashPlayer9以上が必要です