Skip to Main Content

Summary

We have experimented with running three electromagnetic (EM) modeling codes on parallel machines: a 3-D integral-equation code, a finite-difference code for axisymmetric models, and a 3-D finite-difference code. All three codes calculate EM responses at several frequencies and then transform the results into a transient response. Our first method of parallelization uses a task farm where the work is divided into independent subtasks, which are distributed across the processors of a parallel computer. Each subtask calculates the response for one frequency, 30 to 40 of which are required to calculate a transient. This method gives excellent speedups on systems ranging from a departmental workstation cluster to a Cray T3D massively parallel supercomputer.

The 3-D finite-difference program was also ported onto a Thinking Machines CM200 and a DEC mpp 12000/SX and run in data-parallel mode. These single-instruction, multiple-data (SIMD) parallel machines employ several thousand simple processors and offer built-in support for simple operations on data arrays. They are easier to program than clusters of larger, more powerful, processors, but their simplicity limits flexibility in programming. With emerging parallel software standards, however, it soon may be possible to run the same software on clusters of workstations or massively parallel supercomputers with little change.

You do not currently have access to this chapter.

Figures & Tables

Contents

GeoRef

References

Related

Citing Books via

Close Modal
This Feature Is Available To Subscribers Only

Sign In or Create an Account

Close Modal
Close Modal