SC²S Colloquium - December 16, 2016

From Sccswiki
Jump to navigation Jump to search
Date: December 16, 2016
Room: 02.07.023
Time: 3:00 pm, s.t.

Julian Spahl: Evaluation of Zonal Methods for Small Molecular Systems

Molecular dynamics simulations are an increasingly important tool for research in biological and chemical processes. To decrease the time needed for a simulation parallelization is introduced. One way to introduce parallelization is to cut the domain in zones. To exchange the forces each process needs information about the particles of other processes. This communication can be done in different schemes. In this work three of these zonal parallelization methods were implemented and evaluated. Finally they are compared to the current implementation of MarDyn.

Christian Lichtmannegger: Efficient Parallelization of Trajectory Prediction Algorithms on Multicore Systems

The goal of this thesis is to create a generic system for trajectory calculation. Trajectories define the movement of an object under certain conditions. Many different application fields require trajectory calculations, for example astrophysics or nautical navigation. De- pending on the underlying mathematical model, trajectories can be arbitrary complex. Most calculations are based on differential equations which can only be solved using nu- merical algorithms. There already exist different systems specially tailored for one field of research. This thesis shows a generic system based on an already existing system. This system is decoupling the overall algorithm for the given use case to provide a generic software architecture. Starting from a set of possible initial parameters, trajectories must be calculated and rated on a use case specific basis. The system developed is deisgned for real time usage, such that the calculations must be finished before a certain deadline. Given a large enough input space it is infeasible to calculate all trajectories, so the system must strongly rely on concurrent computation and approximation techniques for the cal- culation. The capabilities and restrictions of the legacy system are analyzed and refactored into a generic new system. The refactoring covers a change of the software architecture, the use of more efficient different data structures and also addresses problems induced by concurrent programming. This thesis covers multiple aspect of software development as well as sequential and parallel optimization strategies. The system designed in this thesis is mainly based on software pattern to achieve a very high flexibility and modulariza- tion which is necessary for future extensions. Therefore tailoring the system for specific needs is easily possible. As a proof of concept the generic system is mapped to a use case. Here trajectories must be calculated on a standard processor with limited computational resources under a given deadline. It is evaluated both the performance of the system and the effort in modifying the system. The resulting system is a generic system which can easily be adapted to various needs. This increases the maintainability and allows easy fu- ture extensions. Using concurrent optimizations the system is scalable and therefore can be used for arbitrary complex scenarios.