An SP2 Supercomputer User Experience : Computational Fluid Dynamics and Parallel Programming

Computer Centre has recently interviewed Dr. LIU Chun Ho on his experience in using the IBM SP2 supercomputer in doing his research work at HKU.

Dr. Liu did his undergraduate studies at the University of Hong Kong and was awarded a BEng (Mech) degree in 1995.  After graduation, he continued to pursue his post-graduate studies at the Department of Mechancal Engineering and completed his PhD degree in 1999.  He then worked as a Research Associate in the University until the end of 1999 when he took up work as a postdoctoral research fellow in National Center For Atmospheric Research, Colorado, USA, with the support of the Croucher Foundation, Hong Kong.  Dr Liu's PhD research were conducted using the IBM SP2 Supercomputer supported by Computer Centre .  We were very pleased to get Dr Liu to share some of his experience in using the IBM SP2 supercomputer and parallel programming.

The interview was conducted by our Senior Computer Officer Mr Kwan Wing Keung and Assistant Computer Office Mr Woo Chat Ming, who are responsible for the SP2 support:

KWK:  "Dr Liu, congratulations on your appointment at the NCAR.  Thank you for agreeing to share some of your experience on using the SP2 supercompter with our readers.  Why did you use the parallel computer in Computer Centre for your research work?"

Dr Liu:  "My research topic concerns the use of computational fluid dynamics (CFD) for simulating the atmospheric boundary layer and air pollutant dispersion. Such kind of computer simulation is similar to most CFD research work  that always takes huge amount of time. Even using the fastest workstation in the world, each numerical simulation experiment may take several days to give me only one set of result data. It is clear that a serial computer cannot solve my problem.  A parallel computer like the IBM SP2 has many CPUs, which can solve the numerical equations concurrently with results obtained much faster than from a serial computer. It produces results for me in hours rather than in days or weeks. I may say that it is the ONLY viable solution for my research."

KWK:  "How long have you been using the parallel computer?"

Dr Liu:  "I started using the SP2 as soon as I started my PhD degree 4 years ago. I didn't even make any attempt on other serial computers. I knew I must use the parallel computer. "

"Is parallel programming difficult to learn?"

"No. I would say it is rather easy. The most common computer software for parallelization is the Message Passing Interface (MPI). MPI is a software library and can be called from Fortran or C. Even though MPI has more than 100 subroutines, most of my parallel programs use less than 20 MPI subroutines. I became very proficient in MPI programming within 2 weeks. Also, Computer Centre staff gave me a lot of advice in the development process. This made it easier than I had imagined."

"Is parallelization of your code difficult?"

"Learning MPI is simple, but using it to build my own project consumes certain amount of time. However, it is not as difficult as I had imagined. The whole process in parallelizing my 20,000 lines of code took less than 2 months. I used this program throughout my PhD study. That effort had proved to be very worthwhile."

"Do you think the parallel computer is suitable for CFD problems alone?"

"There is no doubt that parallel computers play an important role in CFD research nowadays. CFD programs are easy to parallelize and efficient to run.  They are also very scalable, i.e., the execution speed increases proportionally with the number of processors used without huge amounts of overhead. However, I am sure the parallel computer can solve many other scientific problems also. I have heard that researchers in the Faculty of Science have also obtained some signficant results using the SP2."

Kwan Wing Keung
Tel : 28678631
E-mail :

Woo Chat Ming
Tel : 28578632
E-mail :

[Contents] [Next] [Previous]