High Performance Thermal-aware Distributed Computing

The power consumption of large scale clusters is now a critical design constraint in high-performance clusters. This increased power consumption and system densities have multiple side effects including increased operational costs, higher average operating temperatures which decreased reliability of microelectronics and, this additional heat which must be dissipated by complex cooling systems. For example, the Arrhenius equation states a temperature increase of 10 degrees Celsius results in reliability decrease of an electronic device by 50 percent. In this book, we discuss and analyze answers to common questions like: What parts of my parallel application will benefit from thermal management techniques?, Where do I start optimizing my parallel application to reduce thermals?, Are the thermal properties of my application similar across machines in a cluster?, What and where are the performance effects of thermal optimizations on my application?, How do I regulate thermals within the threshold limits and yet maintain performance? Readers with a background in the computer science, electrical and electronics engineering find this book interesting.