4th EGEE User Forum/OGF 25 and OGF Europe's 2nd International Event

Results and Evaluation of ThIS on the Grid

3 Mar 2009, 16:10

5m

Foyer (Le Ciminiere, Catania, Sicily, Italy)

Poster Experiences from application porting and deployment Poster session

Speaker

Mrs Sorina Camarasu Pop (CNRS - CREATIS LRMN)

Description

ThIS is a Therapeutic Irradiation Simulator for cancer therapy which is now being integrated into the OpenGATE project. It computes the 3D dose distribution resulting from an irradiation with carbon ion beams. The simulator has been ported on the EGEE Grid for computational speed-up. The porting methodology was presented at EGEE’08. We would now like to present current activity as well as an evaluation of the performances obtained due to the grid usage and the chosen porting method.

Impact

The advanced porting method and execution of ThIS on EGEE brings significant improvement. The results can be evaluated from different points of view. First, we look at the global speed-up obtained with the proposed architecture with respect to a conventional grid usage. We then analyse the reasons behind the differences in performance. Results show that job failure is a key element that distinguishes the two approaches. Therefore we analyse the source of failed jobs as well as their impact in the two cases. Another important issue that must be taken into account is the influence of the number of agents per simulation. Up to a certain number of agents, the larger the number, the more important the global speed-up will be. Nevertheless, after a certain threshold the speed-up is no longer significant. Moreover, an extremely large number of submitted agents may damage the grid equilibrium. Therefore, a compromise has to be found for optimum results and fair grid usage.

Keywords

application porting, pilot jobs, experience feed-back

Detailed analysis

The idea behind an optimized execution of ThIS on the EGEE Grid was based on a simple principle: use parallel computing by splitting one simulation into sub-simulations executed concurrently on different CPUs. The simulator is Monte-Carlo based so it is possible to make each job independent by using a specific random number seed for each job. Further optimization was possible by using a master-agent architecture with the help of the DIANE and Ganga RESPECT tools. The master distributes the work among the running agents depending on their performances. This approach presents two major advantages. First, it takes advantage of rapidly executed jobs and of powerful processors which will be able to simulate more than the others, thus increasing computational speed-up. Second, it is robust with respect to failed jobs that may hinder the achievement of a 100% result.

URL for further information

https://www.creatis.insa-lyon.fr/rio/ThIS
http://opengate-redesign.healthgrid.org

Conclusions and Future Work

The results are promising and we hope this advanced porting will help us boost both the number of users and the simulation results. In parallel, a high level interface for users with no grid specific knowledge is essential for opening ThIS to a larger research and medical community. We are currently interfacing the existing ThIS-DIANE architecture with tools like Moteur and VBrowser for a research usage and with the HOPE platform for clinical use.