Performance Evaluation Metrics Models And Benchmarks

This book constitutes the refereed proceedings of the SPEC International Performance Evaluation Workshop, SIPEW 2008, held in Darmstadt, Germany, in June 2008. The 17 revised full papers presented together with 3 keynote talks were carefully reviewed and selected out of 39 submissions for inclusion in the book. The papers are organized in topical sections on models for software performance engineering; benchmarks and workload characterization; Web services and service-oriented architectures; power and performance; and profiling, monitoring and optimization.

This book constitutes the thoroughly refereed proceedings of the SPEC International Performance Evaluation Workshop, SIPEW 2008, held in Darmstadt, Germany, in June 2008 . The 17 revised full papers presented were carefully selected out of 39 submissions for inclusion in the book. The papers are organized in topical sections on models for software performance engineering; benchmarks and workload characterization; Web services and service-oriented architectures; power and performance; and profiling, monitoring and optimization.

Computer and microprocessor architectures are advancing at an astounding pace. However, increasing demands on performance coupled with a wide variety of specialized operating environments act to slow this pace by complicating the performance evaluation process. Carefully balancing efficiency and accuracy is key to avoid slowdowns, and such a balance can be achieved with an in-depth understanding of the available evaluation methodologies. Performance Evaluation and Benchmarking outlines a variety of evaluation methods and benchmark suites, considering their strengths, weaknesses, and when each is appropriate to use. Following a general overview of important performance analysis techniques, the book surveys contemporary benchmark suites for specific areas, such as Java, embedded systems, CPUs, and Web servers. Subsequent chapters explain how to choose appropriate averages for reporting metrics and provide a detailed treatment of statistical methods, including a summary of statistics, how to apply statistical sampling for simulation, how to apply SimPoint, and a comprehensive overview of statistical simulation. The discussion then turns to benchmark subsetting methodologies and the fundamentals of analytical modeling, including queuing models and Petri nets. Three chapters devoted to hardware performance counters conclude the book. Supplying abundant illustrations, examples, and case studies, Performance Evaluation and Benchmarking offers a firm foundation in evaluation methods along with up-to-date techniques that are necessary to develop next-generation architectures.

The author is one of the prominent researchers in the field of Data Envelopment Analysis (DEA), a powerful data analysis tool that can be used in performance evaluation and benchmarking. This book is based upon the author’s years of research and teaching experiences. It is difficult to evaluate an organization’s performance when multiple performance metrics are present. The difficulties are further enhanced when the relationships among the performance metrics are complex and involve unknown tradeoffs. This book introduces Data Envelopment Analysis (DEA) as a multiple-measure performance evaluation and benchmarking tool. The focus of performance evaluation and benchmarking is shifted from characterizing performance in terms of single measures to evaluating performance as a multidimensional systems perspective. Conventional and new DEA approaches are presented and discussed using Excel spreadsheets — one of the most effective ways to analyze and evaluate decision alternatives. The user can easily develop and customize new DEA models based upon these spreadsheets. DEA models and approaches are presented to deal with performance evaluation problems in a variety of contexts. For example, a context-dependent DEA measures the relative attractiveness of similar operations/processes/products. Sensitivity analysis techniques can be easily applied, and used to identify critical performance measures. Two-stage network efficiency models can be utilized to study performance of supply chain. DEA benchmarking models extend DEA’s ability in performance evaluation. Various cross efficiency approaches are presented to provide peer evaluation scores. This book also provides an easy-to-use DEA software — DEAFrontier. This DEAFrontier is an Add-In for Microsoft® Excel and provides a custom menu of DEA approaches. This version of DEAFrontier is for use with Excel 97-2013 under Windows and can solve up to 50 DMUs, subject to the capacity of Excel Solver. It is an extremely powerful tool that can assist decision-makers in benchmarking and analyzing complex operational performance issues in manufacturing organizations as well as evaluating processes in banking, retail, franchising, health care, public services and many other industries.

This book constitutes the reviewed proceedings of the first Conference on Performance Evaluation and Benchmarking, TPCTC 2009, held in Lyon, France, August 24-28,2009. The 16 full papers and two keynote papers were carefully selected from 34 submissions. This book considers issues such as appliance, business intelligence, cloud computing, complex event processing, database performance optimizations, green computing, data compression, disaster tolerance and recovery, energy and space efficiency, hardware innovations, high speed data generation, hybrid workloads or operational data warehousing, unstructured data management, software management and maintenance, virtualization and very large memory systems

Application-level monitoring of continuously operating software systems provides insights into their dynamic behavior, helping to maintain their performance and availability during runtime. Such monitoring may cause a significant runtime overhead to the monitored system, depending on the number and location of used instrumentation probes. In order to improve a system’s instrumentation and to reduce the caused monitoring overhead, it is necessary to know the performance impact of each probe. While many monitoring frameworks are claiming to have minimal impact on the performance, these claims are often not backed up with a detailed performance evaluation determining the actual cost of monitoring. Benchmarks can be used as an effective and affordable way for these evaluations. However, no benchmark specifically targeting the overhead of monitoring itself exists. Furthermore, no established benchmark engineering methodology exists that provides guidelines for the design, execution, and analysis of benchmarks. This thesis introduces a benchmark approach to measure the performance overhead of application-level monitoring frameworks. The core contributions of this approach are 1) a definition of common causes of monitoring overhead, 2) a general benchmark engineering methodology, 3) the MooBench micro-benchmark to measure and quantify causes of monitoring overhead, and 4) detailed performance evaluations of three different application-level monitoring frameworks. Extensive experiments demonstrate the feasibility and practicality of the approach and validate the benchmark results. The developed benchmark is available as open source software and the results of all experiments are available for download to facilitate further validation and replication of the results.