JP1.3
Building the GSFC Land Information System with distributed and parallel computing technologies
Yudong Tian, UMBC/GEST, Greenbelt, MD; and C. Peters-Lidard, S. V. Kumar, J. V. Geiger, S. Olden, L. Lighty, J. L. Eastman, P. R. Houser, J. Sheffield, E. F. Wood, K. E. Mitchell, J. Meng, H. Wei, P. Dirmeyer, B. Doty, and J. M. Adams
We have developed a land surface simulation and data assimilation system, the Land Information System (LIS), at NASA, Goddard Space Flight Center in the past 3 years. LIS is a seamless fusion of traditional land surface modeling and data assimilation with state-of-the-art distributed and parallel computing technologies, to achieve an unprecedented level of performance, scalability and interactivity of global land surface processing. Specifically, LIS can perform global land surface simulation at a spatial resolution as high as 1 km in real-time, thanks to our development in the array of technologies including advanced software engineering, cluster computing, distributed processing, parallel I/O, remote interactive visualization and on-demand data serving.
We will demonstrate the overall LIS hardware and software architecture, and present the new enabling technologies we have developed and adopted which made the high-performance, high-throughput and highly interactive land surface processing possible. In particular, we will demonstrate the job control system to manage parallel computing on our distributed platform, a 200-node Linux cluster, with strong fault-tolerance and scalability up to 190 processors, In addition, we will illustrate our innovation in the parallel I/O and distributed storage design to eliminate throughput bottlenecks, with an extra benefit of 10-times increase in performance for on-demand data serving to users. Furthermore, we will report on the application of cluster computing with the integration of a parallelized on-demand data-serving system, the GrADS-DODS (or -OPeNDAP) server, to enhance remote interactive data retrieval and to increase data throughput, and show the potential application to other Earth modeling systems. Finally, LIS' superior portability resulted from our disciplined software engineering, and interoperability based on the Earth System Modeling Framework (ESMF), will also be illustrated, as manifested by the rapid adoption of LIS code by multiple research communities, by the porting to multiple platforms including Linux, SGI, IBM SP and Windows, and by the coupling between LIS and other models such as GCE and WRF. With LIS’ unique capability, we have obtained substantial scientific results, which will be presented in other AMS conference sessions. The data products are also readily available to the public from our web-based, high-performance data portal.
Joint Poster Session 1, Poster Session: Distributed Earth Science Information Systems (Joint with the 16th Conference on Climate Variability and Change and the 21st International Conference on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology)
Tuesday, 11 January 2005, 9:45 AM-11:00 AM
Previous paper Next paper