Parallelization speeds x-ray tube design
Simulating the design of a new x-ray tube is a time-consuming and costly process. In developing new tubes, designers usually use finite-element-based CAD packages that run on high-performance workstations. A number of companies including Philips (Hamburg, Germany) use the Scala CAD package from Vector Fields Limited (Aurora, IL) in the development of such tubes. In operation, the Scala package models the steady-state behavior of charged particle-beam devices using finite-element analysis.
Because such packages use finite-element analysis, the process of simulating a new x-ray tube can take from one to two days. Shortening this to a few hours would reduce total development time and cost. But to do so requires porting established computer-aided-design code to parallel architectures. Working in conjunction with Philips, Oxford Parallel (Oxford, England) developed a parallel version of Scala to run on a Sun Sparc workstation already installed in Philips` Hamburg facility. Using Oxford Parallel`s BSP library, a tool that simplifies the process of creating parallel programs in C and FORTRAN on parallel computing systems, Oxford optimized the Vector Fields scalar package to run on parallel processors.
To test the results, the company developed two sets of x-ray tube data--one that required 480 rays to be simulated and one that required 4000 rays. These data were then run on an IBM SP2 processor and the results tabulated. "The 480 ray simulation scales reasonably up to four, while the 4000 ray simulation scales reasonably up to a larger number of machines," says Bob McLatchie of Oxford Parallel. Using parallel processors, the simulation time for such designs can be reduced by factors of up to six. "Although this could not be cost-justified for the Phillips` Hamburg x-ray development group, it could be for TV CRT tube development at Philips` Eindhoven facility," says McLatchie. This group, which currently uses in-house software, is considering using commercial software.
"They are currently evaluating Scala and comparing the results with their own software. Having a parallel version of Scala could be an advantage, because the present run time of 10 hours for a typical problem could be reduced to 40 minutes using 24 processors," McLatchie adds. Philip`s TV and CRT developers are also considering buying a parallel computer because estimates of design times for a new CRT model range from three to six months using all their available workstations. For more information, contact Oxford Parallel at email@example.com or Vector Fields at firstname.lastname@example.org.