Highly Parallel Computing (The Benjamin/Cummings Series in Computer Science and Engineering)

Revised to provide more coverage of current parallel programming environments and the latest software, this book includes Fortran 90 (-D and -clustered), Express, PVM, Linda and Occam. The book includes treatment of the issues involved in choosing a language for parallelism and developing parallel computational modes. The book discusses the issues around choosing and designing compilers and operating systems for large-scale parallel computers, including a section on parallelism in LAN-connected workstations. It describes applications from both numerical and symbolic processing, including atmospheric modelling for weather prediction, particle calculations, computer-aided design, seismic imaging, database systems and artificial intelligence approaches. The book concludes with three chapters discussing the design trade-offs between SIMD architectures, MIMD architectures and Hybrid architectures. The text includes descriptions of the latest parallel machines from Intel, Thinking Machines, Maspur and Kendall Square Research alongside updates from Ncube and Alliant. It offers extensive coverage of new architectural ideas such as directory-based cache coherence (as exemplified by the DASH and Alewife research prototypes) UMA/NUMA/COMA memory organizations, fat trees (as exemplified by CM-5) and de-populated 3D touruses (as exemplified by the Tera design). It also adds coverage of the latest technologies, including RISC chips, fibre optics, GaAs and BiCMOS and software such as GFortran 90, Express and Linda.