Patterns for Parallel Software Design

Foreword

The steady increases in processor speeds associated with Moore's Law has improved software performance for decades without necessitating significant changes in software designs or practices. Over the past several years, however, the exponential growth in CPU speed has stalled. Increases in software performance now stem largely from exploiting parallel processing to exchange data reliably and scalably across high-speed interconnects, dynamically balance workload in computation grids, and efficiently synchronize access to shared resources. Researchers and practitioners rely on parallel processing to accelerate scientific discoveries and deliver value to users in a wide range of application domains, including high-performance scientific computing, weather forecasting, financial services, animation rendering, text mining, homeland security and enterprise content management.

Although parallel processors and interconnects continue to improve, it remains tedious and error-prone to develop complex application and infrastructure software that can meet challenging - and changing - user requirements. This situation has yielded a 'parallel software crisis', in which the hardware becomes ever more capable but the software remains hard to develop, debug, optimize and evolve. Much of the effort expended on parallel software is spent rediscovering core concepts such as coordination, communication, and synchronization, and reinventing common components such as active objects, dynamic ...