Overview
Topic status: We're looking for students to study this topic.
High Performance Computing is a powerful tool in many areas of scientific research. In a recent parallelisation of a mathematical model simulating chemical kinetics and diffusion processes taking place on the plasma membrane of a biological cell, portions of the membrane (represented as a 2D grid) were allocated to different computational processes. As the simulation progresses, there is movement of proteins from one computational process to another. How should this membrane data be transferred? One option is to transfer an entire column of the grid of membrane data, while another option is to transfer just the occupied columns (but this requires a preliminary message giving the count of the data items about to be transferred).
This is an important issue, as the parallel simulation allows an entire cell membrane to be modelled, yet the data requirements are then large - the membrane may be represented by an array of 5000 x 5000 data elements, for example. Choosing an inappropriate strategy for message-passing would have significant impact on the performance of the simulation.
Hypothesis/Aims
The main aim is to analyse the performance of different message-passing strategies, taking into account message size, message frequency, and latency.
Approaches
The existing simulation is written in C/C++ and OpenMPI, so a knowledge of these environments is required.
References
Burrage, P.M., Burrage, K., Kurowski, K., Lorenc, M., Nicolau Jr, D.V., Swain, M. and Ragan, M., A parallel plasma membrane simulation, Proceedings of 1st International Workshop on High Performance Computational Systems Biology (HiBi2009), Trento, Italy, 14-16 October 2009, ed. Juan Guerrero, Conference Publishing Services, IEEE Computer Society, pp. 105-112, ISBN: 978-0-7695-3809-9, 2009.
- Study level
- Honours
- Supervisors
- QUT
- Organisational unit
Science and Engineering Faculty
- Research area
- Contact
- Please contact the supervisor.
Dr Pamela Burrage
