A desktop computer can not provide enough computing power to satisfy traditional sequential geospatial information ... to support near real-time geospatial applications with near real-time routing as an example. ... 2) Each section with a MITSIMLab program is submitted to the grid as a job; ... Institute of Technology,1997.
UTILIZING GRID COMPUTING TO SUPPORT NEAR REAL-TIME GEOSPATIAL APPLICATIONS Jibo Xie, Chaowei Yang, Ying Cao, Menas Kafatos Joint Center for Intelligent Spatial Computing, College of Science George Mason University, 4400 Univ. Dr., Fairfax, VA, 22030-4444 1. INTRODUCTION The increasing utilization of geospatial information to facilitate emergency decisions requires a Geographic Information System (GIS) to process large amounts of geospatial data in a timely fashion. Real-time GIS applications require sub-second response times to analytical operations on large and highly dynamic datasets for bringing good interactivity to the user interface [1]. For example, near real-time routing requires predictions based on integrating massive transportation patterns in several minutes. A desktop computer can not provide enough computing power to satisfy traditional sequential geospatial information processing. Exploiting idle workstations [2, 3, 4] becomes a feasible way to meet the needs of these computing challenges. Grid computing [5, 6], a new computing infrastructure, provides a potential solution to harness idle workstations and other resources in a flexible manner. Therefore, recent research has been conducted on leveraging grid computing to enhance the computing capability for GIS applications [7, 8]. This paper describes our research on utilizing grid computing to support near real-time geospatial applications with near real-time routing as an example. MITSIMLab [9], a popular routing simulation platform, is used as the simulator for integrating and predicting information from traffic networks. 2. METHODOLOGIES Utilizing grid computing to support near real-time geospatial applications, we need to construct a grid and grid-enable the applications: 1) To construct a grid, a middleware is needed and extended to schedule geospatial applications across different computers. Condor [10] is used and we developed relevant wrappers for scheduling parallel processing to reduce the overall processing time. 2) There are two methods to grid-enable a geospatial application: first, we can redesign the algorithm to make sequential processing executable in parallel. Secondly, we can decompose the study area into different parallel executable sections. The latter approach is chosen to grid-enable the near real-time traffic simulation. Grid-enabling near real-time traffic simulations is conducted in six steps: 1) Data decomposition: the entire traffic dataset is decomposed into sections. Data redundancy is needed because there is geospatial correlation between neighboring sections; 2) Each section with a MITSIMLab program is submitted to the grid as a job; 3) The grid job scheduler matches the jobs to computing nodes, and uses matched computing nodes to execute jobs; 4) Computing nodes return results to the central manager separately and concurrently; 5) Merge separates results received into a complete travel time table; 6) Conduct near real-time routing based on the complete travel time table. 3. EXPERIMENTATION A Condor computing pool is constructed based on a Linux cluster connected by a local area network. One server is used as the central manager and the others as computing nodes. The central manager serves as a broker between resources and jobs and schedules the best available resource for a given job. As an example of the experiment, part of the traffic network in Washington DC, a region with over 50,000 roads, is used as test data. The study area is divided into 25 regular sections with an overlap of 1000 meters for each section. The overlap between sections is used to reduce the boundary errors induced by data decomposition. Each section is sent to a computing node matched by the job scheduler in the condor pool and computed
separately. To test the performance, 2, 4, 8, and 22 CPU cores in the condor pool are used. The experimental results show that the response time is reduced with the increase of the CPU cores. 4. CONCLUSION & DISCUSSION Using real time traffic simulation as an example, this paper studied the method of utilizing grid computing to support near real-time geospatial applications. The performance tests demonstrate that we can improve the performance of grid-enabled traffic simulations by reducing the simulating time by adding more computing nodes. The method can also benefit other relevant near real-time geospatial applications, such as emergency management. 5. REFERENCES [1] R. G. Healey, M. J. Minetar, and S. Dowers, Eds, “Parallel Processing Algorithms for GIS”, Taylor & Francis, Inc,England,1997. [2] Achary, A., G. Edjlali, and J. Saltz,“The utility of exploiting idle workstations computation”, ACM SIGMETRICS Performance Evaluation Review,25(1),1997. [3] Basney, J. and M. Livny, “Managing network resources in Condor”, High-Performance Distributed Computing Proceedings. The Ninth International Symposium on 1-4 Aug, 2000, pp.298-299,2000. [4] Thain, D., T. Tannenbaum, and M. Livny, “Distributed computing in practice: the Condor experience: Research Articles”, Concurrency and Computation: Practice & Experience(17), pp. 2-4,2005. [5] Foster, I. and Kesselman C., “The Grid: Blueprint for a New Computing Infrastructure”, Morgan Kaufmann, Los Altos, CA,1999. [6] Foster, I., C. Kesselman, and S. Tuecke, “The Anatomy of the Grid: Enabling Scalable Virtual Organizations”, International Journal of High Performance Computing Applications, pp. 200-222, 2001. [7] Yang, C., Kafatos, M., Wong, D., Yang, R., Cao, Y., “GridGIS: A next generation GIS”, CITSA 2004, Jul. 21-25, Orlando, FL, pp.22-27, 2004. [8] Cao, Y., “Transportation Routing with Real-Time Events Supported by Grid Computing”, Ph.D. Dissertation, George Mason University, 2007. [9] Yang,Q., “A simulation laboratory for evaluation of dynamic traffic management System”ˈPhD thesis, Massachusetts Institute of Technology,1997. [10] Condor Project Homepageˈhttp://www.cs.wisc.edu/condor/, 2007