The Standard Performance Evaluation Corporation (SPEC) CPU benchmark has been widely used as a measure of computing performance for decades. The SPEC is an industry-standardized, CPU-intensive benchmark suite and the collective data provide a proxy for the history of worldwide CPU and system performance. Past efforts have not provided or enabled answers to questions such as, how has the SPEC benchmark suite evolved empirically over time and what micro-architecture artifacts have had the most influence on performance?—have any micro-benchmarks within the suite had undue influence on the results and comparisons among the codes?—can the answers to these questions provide insights to the future of computer system performance? To answer these questions, we detail our historical and statistical analysis of specific hardware artifacts (clock frequencies, core counts, etc.) on the performance of the SPEC benchmarks since 1995. We discuss in detail several methods to normalize across benchmark evolutions. We perform both isolated and collective sensitivity analyses for various hardware artifacts and we identify one benchmark (libquantum) that had somewhat undue influence on performance outcomes. We also present the use of SPEC data to predict future performance.

DELAUNAYSPARSE contains both serial and parallel codes written in Fortran 2003 (with OpenMP) for performing medium- to high-dimensional interpolation via the Delaunay triangulation. To accommodate the exponential growth in the size of the Delaunay triangulation in high dimensions, DELAUNAYSPARSE computes only a sparse subset of the complete Delaunay triangulation, as necessary for performing interpolation at the user specified points. This article includes algorithm and implementation details, complexity and sensitivity analyses, usage information, and a brief performance study.

Exponential increases in complexity and scale make variability a growing threat to sustaining HPC performance at exascale. Performance variability in HPC I/O is common, acute, and formidable. We take the first step towards comprehensively studying linear and nonlinear approaches to modeling HPC I/O system variability in an effort to demonstrate that variability is often a predictable artifact of system design. Using over 8 months of data collection on 6 identical systems, we propose and validate a modeling and analysis approach (MOANA) that predicts HPC I/O variability for thousands of software and hardware configurations on highly parallel shared-memory systems. Our findings indicate nonlinear approaches to I/O variability prediction are an order of magnitude more accurate than linear regression techniques. We demonstrate the use of MOANA to accurately predict the confidence intervals of unmeasured I/O system configurations for a given number of repeat runs - enabling users to quantitatively balance experiment duration with statistical confidence.

In recent years, accelerated destructive degradation testing (ADDT) has been applied to obtain the reliability information of an asset (component) at use conditions when the component is highly reliable. In ADDT, degradation data are measured under stress levels more severe than usual so that more component failures can be observed in a short period. In the literature, most application-specific ADDT models assume a parametric degradation process under different accelerating conditions. Models without strong parametric assumptions are desirable to describe the complex ADDT processes. This paper proposes a general ADDT model that consists of a nonparametric part to describe the degradation path and a parametric part to describe the accelerating-variable effect. The proposed model not only provides more model flexibility with few assumptions, but also retains the physical mechanisms of degradation. Due to the complexity of parameter estimation, an efficient method based on self-adaptive differential evolution is developed to estimate model parameters. A simulation study is implemented to verify the developed methods. Two real-world case studies are conducted, and the results show the superior performance of the developed model compared with the existing methods.

Reliability analysis of multicomponent repairable systems with dependent component failures is challenging for two reasons. First, the failure mechanism of one component may depend on other components when considering component failure dependence. Second, imperfect repair actions can have accumulated effects on the repaired components and these accumulated effects are difficult to measure. In this paper, we propose a parametric statistical model to capture the failure dependence information with general component repair actions. We apply the maximum likelihood method to estimate the model parameters by utilizing the historical failure data. Statistical hypothesis tests are developed to determine the dependence structure of the component failures based on the proposed reliability model. The proposed methodology is demonstrated by a simulation study and case studies of a car body assembly process and a forklift vehicle system.

Many consumer products are designed and manufactured so that the probability of failure during the technological life of the product is small. Most product units in the field retire before they fail. Even though the number of failures of such products is small, there is still a need to model and predict field failures for purposes of risk assessment in applications that involve safety. Challenges in the modeling and prediction of failures arise because the retirement times are often unknown, few failures have been reported, and there are delays in field failure reporting. Motivated by an application to assess the risk of failure for a particular product, we develop a statistical prediction procedure that considers the impact of product retirements and reporting delays. Based on the developed method, we provide the point predictions for the cumulative number of reported failures over a future time period, and corresponding prediction intervals to quantify uncertainty. We also conduct sensitivity analysis to assess the effects of different assumptions on failure-time and retirement distributions.

Existing reliability models for repairable systems with a single component can be well applied for a range of repair actions from perfect repair to minimal repair. Establishing reliability models for multi-component repairable systems, however, is still a challenge problem when considering the dependency of component failures. This paper focuses on a special repair assumption, called partially perfect repair, for repairable systems with dependent component failures, where only the failed component is repaired to as good as new condition. A parametric reliability model is proposed to capture the statistical dependency among different component failures, in which the joint distribution of the latent component failure time is established using copula functions. The model parameters are estimated by using the maximum likelihood method, and the maximum likelihood function is calculated based on the conditional probability. Based on the proposed reliability model, statistical hypothesis testing procedures are developed to determine the dependency of component failures. The developed methods are illustrated with an application in a cylinder head assembling cell that consists of multiple stations.

The relative failure frequency among different failure modes of a production system is referred to as failure profile in this paper. Identification of failure profile based on failure-time data collected in the production phase of a system can help pinpoint the bottleneck problems, and provide valuable information for system design evaluation and maintenance management. Major challenges of effective failure profile identification come from time-varying and limited failure-time data. In this paper, the failure profile is estimated by using the maximum likelihood method. In addition, statistical hypothesis testing procedures are proposed to inspect the existence of a dominating failure mode, and possible changes of failure profiles during a production period. The developed methods are illustrated with an automation system of a high throughput screening (HTS) process, and a production process for cylinder heads.

This paper explores the performance of Chinese small and medium sized enterprises (SMEs) on Business-to-Business (B2B) electronic marketplaces (EMs). Based on a content analysis of 155 cases of high performing online Chinese vendors, this paper explains the success of SME online B2B vendors within a Motivation-Capability framework. This first generation of SME B2B online vendors proved highly motivated to increase sales and developed a set of Internet leveraged organizational capabilities to compete online, including capabilities for online marketing, product innovation, eCommerce management, etc. This study differs from traditional wisdom that online marketplaces will render Guanxi (a Chinese cultural phenomenon defined as close and pervasive interpersonal relationships, Yang, 1994) irrelevant since online marketplaces are perceived to be impersonal. In fact, Guanxi still matters online, but it takes new forms. This research offers important managerial implications for B2B SME online vendors on how to leverage EMs for higher performance.

Accelerated life tests (ALTs) are often used to make timely assessments of the lifetime distribution of materials and components. The goal of many ALTs is the estimation of a quantile of a log-location-scale failure time distribution. Much of the previous work on planning accelerated life tests has focused on deriving test-planning methods under a specific log-location-scale distribution. This paper presents a new approach for computing approximate large-sample variances of maximum likelihood estimators of a quantile of a general log-location-scale distribution with censoring, and time-varying stress. The approach is based on a cumulative exposure model. Using sample data from a published paper describing optimum ramp-stress test plans, we show that our approach and the one used in the previous work give the same variance-covariance matrix of the quantile estimator from the two different approaches. Then, as an application of this approach, we extend the previous work to a new optimum ramp-stress test plan obtained by simultaneously adjusting the ramp rate with the lower start level of stress. We find that the new optimum test plan can have a smaller variance than that of the optimum ramp-stress test plan previously obtained by adjusting only the ramp rate. We compare optimum ramp-stress test plans with the more commonly used constant-stress accelerated life test plans. We also conduct simulations to provide insight, and to check the adequacy of the large-sample approximate results obtained by the approach.

The failure probability of a productF(t), and the life time quantiletpare commonly used metrics in reliability applications. Confidence intervals are used to quantify thes-uncertainty of estimators of these two metrics. In practice, a set of pointwise confidence intervals forF(t), or the quantilestpare often plotted on one graph, which we refer to as pointwise ldquoconfidence bands.rdquo These confidence bands forF(t) ortpcan be obtained throughs-normal approximation, maximum likelihood, or other procedures. In this paper, we compares-normal approximation to likelihood methods, and introduce a new procedure to get the confidence intervals forF(t) by inverting the pointwise confidence bands of the quantiletpfunction. We show why it is valid to interpret the set of pointwise confidence intervals for the quantile function as a set of pointwise confidence intervals forF(t), and vice-versa. Our results also indicate that the likelihood-based pointwise confidence bands have desirable statistical properties, beyond those that were known previously.