Running
CFD on supercomputers
IITZeus
discusses the usage of Param10000 and Param Padma in a computational
fluid dynamics project for aerospace design
The
primary function of the Indian Institute of Technology (IIT) Zeus Group
which is a research group operating in IIT Bombay has been to integrate
the usage of computational fluid dynamics (CFD) with Computer Aided
Engineering (CAE). The group got a significant opportunity to do a prestigious
project when they bagged the civil aircraft project of Casde (Centre
for Aerospace Systems Design and Engineering) Mumbai.
The project began with the evaluation of the technical data such as
the length of fuselage, diameter of fuselage, and tail aspect ratio
of the aircraft. Other vital information was recorded in different types
of files.
The task was to interpret all the data and make a transition from Computer
Adaptive Design (CAD) to simulation tests within a period of six months.
This would be a four-stage process entailing CAD repair, grid generation,
solution of Partial Differential equations and post-processing work.
The highlight of the project was the use of supercomputers for running
the CFD solutions. The machines were Digital Compaq with an Alpha 64-bit
CPU, 883 MHz clock speed, 4 GB RAM, a IBM Pseries with Power 4+ 32-bit
CPU, 1.45 GHz, 4 GB RAM, and a Pentium 4 16-node cluster with P4 1.5
GHz clock speed, 16+2 GB RAM on a Red Hat 7.3 operating system. The
project also used India’s famous Param 10000 8- noded machine with
Ultra Sparcs II CPU, with 400 MHz, 2.5 GB RAM running on Solaris 2.6,
and the Param Padma supercomputer with Power 4+ 64-bit CPU, with a clock
speed of 1.0 GHz, 8 GB RAM running on AIX operating system.
The group simulated speed from a low range of a few mm per second to
the high speed of several km per second. Likewise, different types of
flow ranging from laminar to turbulent were simulated by the software.
Among the various research-based problem encountered by the group, one
of the most challenging was with reference to low speed flow past a
cube mounted on a floor. The grids were structured and had 17 blocks,
and more than 3,00,000 points.
The ratio of cells was 104. The solution scheme comprised density based
Steger warming 2nd order, accurate with MUSCL and global pseudo compressible
preconditioned solver. When comparing the problem solving efficiency
of the various environments, it was noted that Digital Compaq took 800
seconds for 10 iterations, IBM P-615 took 200 seconds, Pentium-4 16
noded cluster with 8 nodes took 230 seconds, Param Padma with 8 nodes
took 93 seconds, while Param 10000 took 399 seconds.
The group made the following inferences about the working of the machines:
. The Param 1000 and the CFD codes written by
the IIT Zeus Group scale up satisfactorily till the number .....
....CONTD
Next
