Distributed processing solves design engineering model

"There's no arbitrary limit on the number of processors that can be networked, using our software," said Louis Crain, Chief Executive Officer of MARC Analysis. "We chose to work with Sun because they have a perspective on distributed computing that no one else has. Their operating system, network software, and computer systems all work together effectively, creating a virtual parallel computer composed of individual workstations. This solution is amazingly simple, effective, and inexpensive."

"This demonstration shows that, with distributed processing, we can solve bigger problems and do so quickly, without having to buy a supercomputer to do the analysis," said Kelly Gomes Perey, director of technical market development at Sun. "Even smaller companies can afford a small `compute ranch' made up of workstations they already own. This will allow them to satisfy their ever-increasing appetite for more and larger simulations."

The companies claim that the ability to solve these problems faster, and to solve larger and more complex problems, can lower the cost of product design and development, and shorten time to market for new automobiles, airplanes, and other products.

Shared Processing Analysis Results

Finite element analyses often require large amounts of computer resources to simulate complex physical phenomena, particularly for geometrically complex systems. Often the problem is so large that it can require an extremely large and expensive supercomputer to solve. In the revolutionary MARC approach, an analysis model is broken into parts called domains, each of which can be run on a separate computer in a network of Sun workstations. These computers communicate over a network or across a switch to synchronize the results during an analysis.

One of the smallest models analyzed tested the ability of a part to handle temperature and mechanical loads and remain symmetrical around an axis of rotation. The results showed the excellent scalability of the distributed solution. On one processor the probem took just under three hours to run. It ran in 30 minutes - six times faster - on two dual-processor workstations, a scalability factor of 6 on four processors.

A medium-size problem (42,000 elements, 198,000 degrees of freedom), designed to measure a part's three-dimensional response under a complex loading condition, took 26 minutes on one processor. On the full network of four dual-processor workstations the problem took just three minutes to complete, a full 8.5 times speedup on eight CPUs.

The largest problem was also a three-dimensional analysis of elastic properties, with 336,000 elements and 1.3 million degrees of freedom. The problem could not be completed on a single processor; in the past, it would have been sent to a "big iron" server to be solved, and the analysis would have taken several hours. On the four-workstation network, the problem was solved in 38 minutes.

Network Configuration

MARC ran the problems on both a Sun HPC 450 server with four processors and on four networked Sun Ultra 60 workstations with dual 360 MHz processors. The workstations were connected using a 100 BaseT network and a Lantronix LSW 8FA Fast Switch.

The 360MHz Ultra 60 workstation is ideal for customers who require raw power for the most demanding scientific and technical applications. With up to two 360MHz UltraSPARCTM II processors with 4MB of cache each and 1.9GB/second of system throughput, the Ultra 60 workstation delivers blazing application performance. Pricing for the 360MHz Ultra 60 workstation with 19" monitor begins at US $14,695.

The Sun HPC 450 delivers excellent performance for almost any mix of compute-, data-, or I/O-intensive applications. It features up to four 250/300-MHz UltraSPARC II processors, a 1.6-GB/sec UPA interconnect, and a 600-MB/sec PCI I/O subsystem for outstanding performance and data throughput. It is available with up to 4 GB of main memory, 84 GB of fast hot-swap UltraSCSI internal storage, and over 6 TB of storage capacity.

Finite Element Analysis Increases Product Reliability, Reduces Costs

Companies use finite element analysis (also known as linear and nonlinear structural analysis) in the the design of products and components to determine whether the design can meet required criteria such as deformation, stress, vibration, and temperature restrictions. FEA can reduce product costs significantly because it identifies critical areas in the design or areas in which a material reduction may be possible. The number of prototypes required before finalizing a production version of the design can usually be reduced, which also lowers product costs.

The eight models used in the MARC analysis looked at turbine engine components, simulating the behaviour of these parts under various operating conditions (such as extremes of speed, temperature, and pressure). With the use of the analysis, an engineer can decide whether the design will perform according to its specifications, or whether it needs modifications. An engineer can also simulate the steps required to manufacture the component, a concept called Virtual Manufacturing.