Profiling the hybrid Molecular Dynamics code, DL_POLY, on heterogeneous clusters
Location: Irish Centre for High End Computing (ICHEC), Dublin, Ireland
Project Abstract:
DL POLY is a well-known molecular dynamics simulation code developed by the Science and Technology Facilities Council in Daresbury Laboratory, UK.
The DL POLY application has been parallelised using a domain decomposition approach, employing the Message Passing Interface (MPI) library for inter-process communication. More recently, the code has been“hybridised” to exploit multi-core shared memory architectures as well as Graphics Processing Units. The existing parallel implementation of DL POLY is highly efficient and scales well to many thousands of CPU cores.
Although DL_POLY is considered to be a well parallelized hybrid code, there is potential to improve the performance of the code on heterogeneous clusters by, for example, extending the GPU capability of the code.
This project will see the student profile DL_POLY using advanced profiling tools available for hybrid codes as well NVIDIA’s most up-to-date Visual Profiler. Graphical output from profilers such as TAU and the NVIDIA Visual Profiler will be analysed and reported on by the student.
The performance analysis of DL_POLY will be used as a means of introducing the student to many of the key concepts of parallel computing and performance engineering on advanced heterogeneous systems.
An Example of DL-POLY simulation. Colin L. Freeman, Science & Technology Facilities Council, UK, http://www.stfc.ac.uk/cse/25526.aspx).
Project Mentor: Michael Lysaght
Site Co-ordinator: Simon Wong
Learning Outcomes:
The student will learn how to:
- Compile an application targeted at large scale heterogeneous clusters.
- Run a parallel application on a heterogeneous cluster.
- Profile a parallel MPI code over several compute nodes.
- Profile a hybrid CPU/GPU-enable code.
Student Prerequisites (Compulsory):
The student must have basic programming skills, preferably in C or Fortran, but also in other language using for scientific computing (e.g. Python, Java). After the first week’s training the student should have picked up some of the key concepts of parallel computing, including the difference between shared memory and distributed memory architectures.
The student must also be familiar with using unix-like operating systems and associated editors such as vim or emacs.
Student Prerequisites (Desirable):
Interest in molecular dynamics simulation
Training Materials:
DL_POLY website: http://www.stfc.ac.uk/cse/25526.aspx
Project Application Reference: Ireland – ICHEC – Hybrid Molecular Dynamics
Applications are now closed
