ABSTRACT
This chapter deals with the development of parallel algorithms to solve computationally intensive problems in the area of structural engineering. It reviews the potential parallelism of some eigenvalue algorithms and discusses the concept of converting sequential algorithms into parallel procedures. The chapter also discusses the general considerations of stratagem design and problem decomposition. State-of-the-art parallel computer systems may be characterized in three classes: pipelined computers, array processors, and multiprocessor systems. The concurrent tasks with the same operations but with different data are preferable since to seed the processing elements with the same code by broadcasting is more economical than seeding the processing elements with different codes. The distribution of data among the processing elements is in general a function of the operations portion of the tasks. However, when the concurrent task operations are identical, the data distributions are those that ensure an even work load among processing elements and minimize the communication costs.
