AITALC is a useful tool for automating calculations needed in High Energy Physics. Using this package we chose the interesting process of electron-positron scattering, and considered the final executable delivering numerical results for porting into a grid environment. We have learned under which conditions porting is feasible, and also the gain achievable in the EGEE infrastructure.
The porting of the application was done with the help of the GRIDWAY metascheduler. A scheme of master/worker was developed where the master remain at the user's side, giving instructions through GRIDWAY about how the submission should be partitioned and managed. Because the code created by AITALC is a dynamically linked executable, a few modifications to the original were required. These modifications allowed the worker to run as a static binary accepting input arguments under different configurations at each grid node. Finally, some postprocessing utilities and improvements were implemented.
HEP applications mainly use GRID resources via data collection and analyses from LHC events and simulations. Theoreticians also require heavy computational tasks to match the same level of accuracy achieved by the experimental measurements. The products generated by AITALC are codes behaving as many other automated tools in the community, and serving as a good candidate to understand and exploit the parallelization capabilities of a typical calculation at high energy colliders.
We consider this example specially useful as a test ground for other, more complex and cumbersome, theoretical calculations strongly needing speed-ups without data loss.
grid, application porting, HEP
Conclusions and Future Work
The porting was succesfully implemented and different workload balances was studied with respect to performance. It turned out that the master/worker method requires a reasonably small amount of user intervention in the worker code and allows the user prepare the master according to the output needs. Further work at the master includes fine-graining of data results through automated feedback and the adoption of the DRMAA standard. We propose also some improvements in AITALC.
URL for further information