Computer performance

Abstract: Measuring Computer Performance sets out the fundamental techniques used in analyzing and understanding the performance of computer systems. Throughout the book, the emphasis is on practical methods of measurement, simulation, and analytical modeling. The author discusses performance metrics and provides detailed coverage of the strategies used in benchmark programmes. He gives intuitive explanations of the key statistical tools needed to interpret measured performance data. He also describes the general 'design of experiments' technique, and shows how the maximum amount of information can be obtained for the minimum effort. The book closes with a chapter on the technique of queueing analysis. Appendices listing common probability distributions and statistical tables are included, along with a glossary of important technical terms. This practically-oriented book will be of great interest to anyone who wants a detailed, yet intuitive, understanding of computer systems performance analysis.

Abstract: Dynamic power management in enterprise environments requires an understanding of the relationship between resource utilization and system-level power consumption. Power models based on resource utilization have been proposed in the context of enabling specific energy-efficiency optimizations on specific machines, but the accuracy and portability of different approaches to modeling have not been systematically compared. In this work, we use a common infrastructure to fit a family of high-level full-system power models, and we compare these models over a wide variation of workloads and machines, from a laptop to a server. This analysis shows that a model based on OS utilization metrics and CPU performance counters is generally most accurate across the machines and workloads tested. It is particularly useful for machines whose dynamic power consumption is not dominated by the CPU, as well as machines with aggressively power-managed CPUs, two classes of systems that are increasingly prevalent.

Abstract: Performance isolation is a key challenge in cloud computing. Unfortunately, Linux has few defenses against performance interference in shared resources such as processor caches and memory buses, so applications in a cloud can experience unpredictable performance caused by other programs' behavior.Our solution, CPI2, uses cycles-per-instruction (CPI) data obtained by hardware performance counters to identify problems, select the likely perpetrators, and then optionally throttle them so that the victims can return to their expected behavior. It automatically learns normal and anomalous behaviors by aggregating data from multiple tasks in the same job.We have rolled out CPI2 to all of Google's shared compute clusters. The paper presents the analysis that lead us to that outcome, including both case studies and a large-scale evaluation of its ability to solve real production issues.

Abstract: Moore's Law states that the number of transistors on a device doubles every two years; however, it is often (mis)quoted based on its impact on CPU performance. This important corollary of Moore's Law states that improved clock frequency plus improved architecture yields a doubling of CPU performance every 18 months. This paper examines the impact of Moore's Law on the peak floating-point performance of FPGAs. Performance trends for individual operations are analyzed as well as the performance trend of a common instruction mix (multiply accumulate). The important result is that peak FPGA floating-point performance is growing significantly faster than peak floating-point performance for a CPU.