next up previous

5 High-Performance Optimization Programming

Future developments in the field of optimization will undoubtedly be influenced by recent interest and rapid developments in new technologies --- powerful vector and parallel machines. Indeed, their exploitation for algorithm design and solution of ``grand challenge'' applications is expected to bring new advances in many fields, such as computational chemistry and computational fluid dynamics.

Supercomputers can provide speedup over traditional architectures by optimizing both scalar and vector computations. This can be accomplished by pipelining data as well as offering special hardware instructions for calculating intrinsic functions (e.g., ,), arithmetic, and array operations. In addition, parallel computers can execute several operations concurrently. Multiple instructions can be specified for multiple data streams in MIMD designs, whereas the same instructions can be applied to multiple data streams in SIMD prototypes. Communication among processors is crucial for efficient algorithm design so that the full parallel apparatus is exploited. These issues will only increase in significance as massively parallel networks enter into regular use.