Service-oriented computation in magnetic fusion research

Abstract

Fusion science seeks a new power source and is advanced by experiments on fusion devices located worldwide. Fundamental to increasing understanding of fusion is the comparison of theory and experiment; measurements from fusion devices are analyzed and compared with the output of simulations to test the validity of fusion models and to uncover new physical properties. Integrating simulations with experimental data (or with other simulations) is in many cases a laborintensive task as different codes use different data storage formats. Moreover, the timely comparison of simulation with observations made during an experiment requires rapid turnaround both of analysis codes and simulation runs. Many simulations require extensive input and output processing, further increasing the amount of work necessary to achieve viable scientific results. Workers with the National Fusion Collaboratory are developing, deploying, and evaluating new technologies that facilitate analysis of experimental data and comparison with the results with those of simulations. Complex physics codes are made available on the National Fusion Grid (FusionGrid) as comprehensive computational services. Using the Globus Toolkit", a servicebased approach was developed and subsequently combined with the TRANSP transport code to the benefit of fusion scientists. Output from both simulation and experimental codes are stored in MDSplus, the de facto standard for secure data storage of fusion data. Access control∀r the resources of FusionGrid is greatly simplifiedfor both users and administrators_through unified authentication and authorization using X.509, a gridwide certificate management system, and a gridwide authorization system. Webbased solutions such as the recently developed Elfresco reflectometry code further simplify the process by making simulations available to scientists and providing an alternative to traditional distribution of code. Future work includes the development of parallelized modules to speed up longrunning codes along with the extension of MDSplus. These improvements will help accommodate the continuous data streams that will be found in future fusion devices such as ITER. This paper will present a discussion on specific solutions, examine deployment areas that present a challenge, and highlight areas where further work is required. © 2007 International Federation for Information Processing.