Failure Prediction

With ever-growing complexity and dynamicity of computer systems, proactive fault management is an effective approach to enhancing availability. Online failure prediction is the key to such techniques. In contrast to classical reliability methods, online failure prediction is based on runtime monitoring and a variety of models and methods that use the current state of a system and, frequently, the past experience as well. This survey describes these methods. To capture the wide spectrum of approaches concerning this area, a taxonomy has been developed, whose different approaches are explained and major concepts are described in detail.

Forecasting precipitation in arid and semi-arid regions, in Jordan in the Middle East for example, has particular importance since precipitation is the unique source of water in such regions. In this study, 1-month ahead precipitation forecasts are made using artificial neural network (ANN) models. Feed forward back propagation (FFBP), radial basis function (RBF) and generalized regression type ANNs are used and compared with a simple multiple linear regression (MLR) model. The models are tested on monthly total precipitation recorded at three meteorological stations (Baqura, Amman and Safawi) from different climatological regions in Jordan. For the three stations, it is found that the best calibrated model is FFBP with respect to all performance criteria used in the study, including determination coefficient, mean square error, mean absolute error, the slope and the intercept in the best-fit linear line of the scatter diagram. In the validation stage, FFBP is again the best model in Baqura and Amman. However, in Safawi, the driest station, not only FFBP but also RBF and MLR perform equally well depending on the performance criterion under consideration.

Background-Prognosis and mode of death in heart failure patients are highly variable in that some patients die suddenly (often from ventricular arrhythmia) and others die of progressive failure of cardiac function (pump failure). Prediction of mode of death may facilitate decisions about specific medications or devices. Methods and Results-We used the Seattle Heart Failure Model (SHFM), a validated prediction model for total mortality in heart failure, to assess the mode of death in 10 538 ambulatory patients with New York Heart Association class II to IV heart failure and predominantly systolic dysfunction enrolled in 6 randomized trials or registries. During 16 735 person-years of follow-up, 2014 deaths occurred, which included 1014 sudden deaths and 684 pump-failure deaths. Compared with a SHFM score of 0, patients with a score of 1 had a 50% higher risk of sudden death, patients with a score of 2 had a nearly 3-fold higher risk, and patients with a score of 3 or 4 had a nearly 7-fold higher risk (PϽ0.001 for all comparisons; 1-year area under the receiver operating curve, 0.68). Stratification of risk of pump-failure death was even more pronounced, with a 4-fold higher risk with a score of 1, a 15-fold higher risk with a score of 2, a 38-fold higher risk with a score of 3, and an 88-fold higher risk with a score of 4 (PϽ0.001 for all comparisons; 1-year area under the receiver operating curve, 0.85). The proportion of deaths caused by sudden death versus pump-failure death decreased from a ratio of 7:1 with a SHFM score of 0 to a ratio of 1:2 with a SHFM score of 4 (P trend Ͻ0.001). Conclusions-The SHFM score provides information about the likely mode of death among ambulatory heart failure patients. Investigation is warranted to determine whether such information might predict responses to or costeffectiveness of specific medications or devices in heart failure patients. (Circulation. 2007;116:392-398.)

Since the seminal work of Pawlak (International Journal of Information and Computer Science, 11 (1982) 341 -356) rough set theory (RST) has evolved into a rule-based decision-making technique. To date, however, relatively little empirical research has been conducted on the e cacy of the rough set approach in the context of business and ÿnance applications. This paper extends previous research by employing a development of RST, namely the variable precision rough sets (VPRS) model, in an experiment to predict between failed and non-failed UK companies. It also utilizes the FUSINTER discretisation method which neglates the in uence of an 'expert' opinion. The results of the VPRS analysis are compared to those generated by the classical logit and multivariate discriminant analysis, together with more closely related non-parametric decision tree methods. It is concluded that VPRS is a promising addition to existing methods in that it is a practical tool, which generates explicit probabilistic rules from a given information system, with the rules o ering the decision maker informative insights into classiÿcation problems. ? (M.J. Beynon). This is evidenced by the recent application of neural networks (NNs), recursive partitioning algorithm (RPA) and case based reasoning to this issue . A key advantage of these contemporary methods over their traditional counterparts (such as MDA and logit analysis) is that they do not require pre-speciÿcation of a functional form, nor the adoption of restrictive assumptions concerning the distributions of model variables and errors .

The goal of this study is to show an alternative method to corporate failure prediction. In the last decades Artificial Neural Networks have been widely used for this task. These models have the advantage of being able to detect non-linear relationships and show a good performance in presence of noisy information, as it usually happens, in corporate failure prediction problems. AdaBoost is a novel ensemble learning algorithm that constructs its base classifiers in sequence using different versions of the training data set. In this paper, we compare the prediction accuracy of both techniques on a set of European firms, considering the usual predicting variables such as financial ratios, as well as qualitative variables, such as firm size, activity and legal structure. We show that our approach decreases the generalization error by about thirty percent with respect to the error produced with a neural network.

Recent developments in the aircraft industry towards substantially improving fuel economy and extending flight range have accelerated interest in the use of advanced composites as primary structural materials. The airframes of next-generation airliners will have substantial parts made of light-weight composites. This means that the engineering demands on the performance of fiber-reinforced composites will become greater. There is therefore a need to better understand and predict the multiple complex failure mechanisms in composite structures, and to devise more reliable failure theories and damage progression models. There is a large body of literature on progressive damage analysis in composites, much of which employs damage mechanics and material stiffness degradation methods. This article reviews some of the more recent work in this area and describes the issues pertinent to application in composite structures. The authors' ongoing research efforts in modeling and prediction ...

Improved methods are proposed for disk drive failure prediction. The SMART (Self Monitoring and Reporting Technology) failure prediction system is currently implemented in disk drives. Its purpose is to predict the near-term failure of an individual hard disk drive, and issue a backup warning to prevent data loss. Two experimentally tests of SMART showed only moderate accuracy at low false alarm rates. (A rate of 0.2% of total drives per year implies that 20% of drive returns would be good drives, relative to ≈1% annual failure rate of drives). This requirement for very low false alarm rates is well known in medical diagnostic tests for rare diseases, and methodology used there suggests ways to improve SMART.

Abstract We are focusing on three alternative techniques that can be used to empirically select predictors for failure prediction purposes. The selected techniques have all different assumptions about the relationships between the independent variables. Linear ...

An experimental platform capable of measuring forces in process during an incremental forming procedure is described and some of the earliest measurements of forces in incremental forming with the changes induced on the measured load are reported. Using a table type force dynamometer with incremental forming fixture mounted on top, three components of force were measured throughout the forming process. They were found to vary as the parts were made. The reported experimental test program was mainly focused on the influence of four different process parameters on the forming forces: the vertical step size between consecutive contours, the diameter of the tool, the steepness of the parts' wall and the thickness of the sheet metal being formed. The effect of lubrication and the geometry of the test part in the incremental forming process were investigated by a set of initial experiments. Part failure prediction based on the shape of the force curve is explained. For the tested materials, analytical results demonstrating the relationship between the respective process parameters and the induced forces are presented in this paper.

Why do some businesses succeed and others end up bankrupt? There is great discrepancy in the literature as to which variables do in fact lead to success, thus, there currently is no theory. To move the field in that direction, this study tests the Lussier 15-variable business success versus failure prediction model in Chile with a sample of 234 small businesses-131 failed and 103 successful. Results support the model's validity in Chile. Thus, the model has been tested with significant results in three very different parts of the world; first in United States (North America), then in Croatia (Central Eastern Europe), and now in Chile (South America). The model will reliably predict a group of businesses as failed or successful more accurately than random guessing in all three countries over 96 percent of the time.

A proactive handling of faults requires that the risk of upcoming,failures is continuously,assessed. One of the promising approaches is online failure prediction, which means,that the current state of the system is evaluated in order to predict the occurrence,of failures in the near future. More specifically, we focus on methods that use event-driven sources such as errors. We use Hidden Semi- Markov Models (HSMMs) for this purpose and demonstrate effectiveness based on field data of a commercial,telecom- munication,system. For comparative,analysis we selected three well-known failure prediction techniques: a straigh t- forward method that is based on a reliability model, Dis- persion Frame Technique,by Lin and Siewiorek and the eventset-based method,introduced by Vilalta et al. We as- sess and compare the methods in terms of precision, recall, F-measure, false-positive rate, and computing time. The ex - periments suggest that our HSMM approach,is very effec- tive with respect to on...

This paper presents a low-cost method to realize a real-time condition monitoring and a predictive-maintenance system of an electrolytic capacitor used in uninterruptible power supplies (UPSs). This method consists in detecting the changes in real time of the equivalent series resistance and the capacitance C values of the electrolytic capacitors. Simulation and experimental results are presented to illustrate the proposed monitoring technique. The proposed method can be used in UPS where waveforms are continuously varying in amplitude, frequency, and temperature. The proposed online failure prediction method has the merits of using only the existent resources in UPS and with the use of known algorithms.

A firm in the early stages of financial distress exhibits characteristics different from those of healthy firms. As the economic condition of a firm worsens, its financial characteristics shift toward those of failed firms. Practitioners in the financial sector have long been interested in the early detection of a firm's slide toward insolvency. Several models have been developed with this purpose in mind, but these older models are static in nature. Therefore, a need exists for the development of business failure prediction models that assess the financial condition of firms sequentially over time. This article addresses this need by presenting a sequential business failure prediction model.

Few studies have addressed the performance of smaller unquoted companies involved in take-overs, especially in the Continental European countries. Therefore this study addresses the post-take-over financial characteristics of privately held companies involved in 143

Gompertz curve has been used to estimate the number of residual faults in testing phases of software development, especially by Japanese software development companies. Since the Gompertz curve is a deterministic function, the curve cannot be applied to estimating software reliability which is the probability that software system does not fail in a prefixed time period. In this article, we

BACKGROUND. The American Society for Therapeutic Radiology and Oncology (ASTRO) definition of biochemical failure (BF) incorporates backdating, resulting in an artificial flattening of Kaplan-Meier curves and overly favorable estimates when follow-up is short. The nadir 1 2 ng/mL (Nadir 1 2; Phoenix) definition reduces these artifacts. The objective of the current study was to compare ASTRO and Phoenix BF estimates as determinants of distant metastasis (DM), cause-specific mortality (CSM), and overall mortality (OM).

Purpose: Past research has found that social, academic, and behavioral problems are linked with depression during childhood and adolescence. The present study tests a longitudinal cascade model of adolescent problems predicting depression into adulthood, while additionally testing for gender differences.

The A-Score Method 34 2.4.1 Introduction 34 2.4.2 Argenti A-Score 35 2.4.3 Cbnclusions to A-Score 37 2.5 Summary and Conclusions 37 CHAPTER 3: THE TECHNIQUE OF DISCRIMINANT ANALNSIS • • 39 3.1 Introduction 39 3.2 The General Cbncept of MDA 41 3.3 The Mathematical Theory Behind MDA • .. • • 43 3.4 Derivation of Computational Formula • • • • 47 3.5 Statistical Cbmputer Package and its use. . 3.6 Summary and Cbnclusions 55 CHAPTER 4: DATA COLLECrICN AND FORMATION OF THE DATA GRCUPS 4.1 Introducticn 4.2 Data Analysis 4.3 Formation of the Data Groups 4.3.1 The "Failed 11" and "Test Failed 11" Groups 4.3.2 The "Non-Failed 20" Group 4.3.3 The "Taken-Over 9" Group 4.3.4 The "Cbntinued 70" Group 4.3.5 The "All 90" Group 4.4 Summary and Cbnclusions CHAPTER 5: THE DEVELOPED MODEL AND REGRESSION CHARACTERISTICS CHAPEER 7: RESULTS FRal THE NICUIS 113 7.1 Introduction 113 7.2 Results Obtained from the "-Failed 20" and "Failed 11" Groups 114 7.2.1 Introduction 114 7.2.2 The Results of the "Non-Failed 20" and "Failed 11" Groups 114 viii "Non-Failed 20" Using Multiple Predicting

The failure prediction in sheet metal forming is typically realized by evaluating the so called forming limit curves (FLC). The standard experimental method is the Nakajima test, where sometimes also the Marciniak test setup is used. Up to now, the FLC determination was performed with failed specimens and an one-directional intersection line method or by manually analyzing the estimated strains

A special type of interface element was created to represent the resin-rich layers between plies in a finite element model of a composite structure. Yielding of the element depended on an interactive stress-based criterion and final failure depended on a mixed-mode ...

All damage and failure models, describing either the evolution of microvoids, the development of shear bands or local rupture, rely on the knowledge of the hardening function at large plastic strains which, then, becomes an essential prerequisite for any failure prediction.

Log preprocessing, a process applied on the raw log before applying a predictive method, is of paramount importance to failure prediction and diagnosis. While existing filtering methods have demonstrated good compression rate, they fail to preserve important failure patterns that are crucial for failure analysis. To address the problem, in this paper we present a log preprocessing method. It consists of three integrated steps: (1) event categorization to uniformly classify system events and identify fatal events; (2) event filtering to remove temporal and spatial redundant records, while also preserving necessary failure patterns for failure analysis; (3) causality-related filtering to combine correlated events for filtering through apriori association rule mining. We demonstrate the effectiveness of our preprocessing method by using real failure logs collected from the Cray XT4 at ORNL and the Blue Gene/L system at SDSC. Experiments show that our method can preserve more failure patterns for failure analysis, thereby improving failure prediction by up to 174%.

Some volcanoes exhibit at times accelerating strain and seismicity before eruptions. Different authors have used this behavior to predict eruptions and to develop materials-science models. Here, a linear Kelvin-Voigt viscoelastic model is proposed to describe the observed behavior. This model also has applications to other failure phenomena, such as landslides. The strain rate of a variety of materials under constant stress after an instantaneous elastic response is found to be well described by a general relaxation law:

Cohesive zone models have had notable success in the prediction of damage from notches in engineering materials loaded in tension. They have also been used to determine the growth of fibre microbuckling from a hole in a composite laminate under compression, (Soutis C, Fleck NA, Smith PA. Failure prediction technique for compression loaded carbon fibre/epoxy laminate with an open hole. J Compos Mater, 1991;25(11):1476-1498). The usual strategy is to replace the inelastic deformation associated with plasticity or microbuckling with a linecrack and to assume some form of stress-displacement bridging law across the crack faces. For instance, in the Dugdale analysis of plastic deformation in metals from the root of a notch (Dugdale DS. Yielding of steel sheets containing slits, J Mech Phys Solids, 1960;8:100-104), it is assumed that the bridging normal traction across the crack faces equals the tensile yield strength of the solid. The material response elsewhere in the cracked structure is assumed to be linear elastic. In this paper, a new method is presented for the calculation of the fracture energy G c associated with fibre microbuckling, a parameter required in the Soutis et al. linear softening cohesive zone model for the prediction of the open hole compression strength. Theoretical results are found to be in good agreement with experimental data for several carbon fibre-epoxy laminates.

As the scale of cluster computing grows, it is becoming hard for long-running applications to complete without facing failures on large-scale clusters. To address this issue, checkpointing/restart is widely used to provide the basic fault-tolerant functionality, yet it suffers from high overhead and its reactive characteriristic. In this work, we propose FT-Pro, an adaptive fault management mechanism that optimally chooses migration, checkpointing or no action to reduce the application execution time in the presence of failures based on the failure prediction. A cost-based evaluation model is presented for dynamic decision at run-time. Using the actual failure log from a production cluster at NCSA, we demonstrate that even with modest failure prediction accuracy, FT-Pro outperforms the traditional checkpointing/restart strategy by 13%-30% in terms of reducing the application execution time despite failures, which is a significant performance improvement for long-running applications.

As failure criterion for sheet metal forming, conventional forming limit diagrams (FLD) are often used. The FLD is a strain based criterion, which evaluates the principal deformations at failure. Different investigations show that the FLD is dependent on the forming history and strain path. However, this is not the case for forming limit stress diagrams (FLSD). For this failure criterion, the principal stresses at failure are determined by FEM simulation of the Nakazima-test. Both, the FLD and the FLSD experimental investigations provide the basis for the sheet metal failure criterion. In contrast, continuum damage mechanics describe the damage evolution in the microstructure with physical equations, so that crack initiation due to mechanical loading can be predicted. By using the Gurson–Tvergaard–Needleman (GTN) damage mechanical model, a failure criterion based on void evolution was examined in this work. The parameter identification for the damage model will be discussed. The investigations demonstrate that FLD is inapplicable for complex forming processes with strain path changes. The FLSD is better suitable than the FLD for multi step forming processes. Micro-mechanical damage modelling with the GTN model also shows acceptable formability predictions, in spite of its generally insufficient consideration of the effective deviatoric stress fractions. The quality of failure prediction using continuum damage mechanics models is able to be increased by applying a more suitable damage models and a modern overall-scale modelling approach.

We present a case study of a difficult real-world pattern recognition problem: predicting hard drive failure using attributes monitored internally by individual drives. We compare the performance of support vector machines (SVMs), unsupervised clustering, and non-parametric sta- tistical tests (rank-sum and reverse arrangements). Some- what surprisingly, the rank-sum method outperformed the other methods, including SVMs. We also show the utility

A growing number of structural applications and repairs require strong adhesively bonded joints between composite parts. The viability of the finite elements method (FEM) for the design and analysis of such joints has been shown by several researchers (Composites 13(1982) 29, mechanics and mechanisms of damage in composites and multi-materials, ESIS11, Mechanical Engineering Publications, London, 1991, Structural adhesive joints in engineering, Elsevier Applied science Publishers, Amsterdam, 1984). Most of the proposed plane stress models for adhesively bonded joints have traditionally relied on effective lamina properties for the composite adherends because one of the primary goals was the study of adhesive stresses and failure criteria (J. Adhes. 42 (1993)). Such work has also been usually carried out on mainframe computers. However, with the rapid developments in computing, commercial FE codes now exists for PC applications. These present new possibilities, namely the construction of models which are closer in their approximation to the structures they represent. Within this context, a new 2D plane stress approach has been developed which allows ply-by-ply modelling of composite adherends in a joint. It is a development of the traditional 2D plane stress model used by several authors such as Siener (Compos. Mater.: Testing Des. 10 (1992) 444) in investigating of composite repair joints.This paper reports the application of this new approach to the modelling of scarf joints between composite adherends under tensile loading. Various scarf angles were investigated and ultimate joint failure predicted. The results of the analysis were compared to the experimental results published by Adkins and Pipes (Proceedings of the Fourth Japan–US Conference on Composite Materials, 1988, pp. 845–854). The main aim of work was to develop a model which could provided more information than conventional models.The new approach offers the prospect of accurately modelling more realistic joints because, through the retention of the laminated nature of the composite adherends, scope now exists for the application of more physically based failure criteria.This work has shown that it is possible to model adhesively bonded composite joints using an improved 2D plane stress model which retains the laminated nature of the composite adherends. There was good agreement between the predicted failure loads and the experimental results for 3.0°, 6.2° and 9.2° scarf joints. These predictions are dependent on the type of failure criteria used. For the composite adherends, the maximum stress criterion faired better in predicting the 3.0° scarf joint strength.

Management of an ever-growing complexity of computing systems is an everlasting challenge for computer system engineers. We argue that we need to resort to predictive technologies in order to harness the system's complexity and transform a vision of proactive system and failure management into reality. We describe proactive fault management, provide an overview and taxonomy for online failure prediction methods and present a classification of failure prediction-triggered methods. We present a model to assess the effects of proactive fault management on system reliability and show that overall dependability can significantly be enhanced. After having shown the methods and potential of proactive fault management we describe a blueprint how proactive fault management can be incorporated into a dependable system's architecture.

The goal of online failure prediction is to forecast imminent failures while the system is running. This paper compares Similar Events Prediction (SEP) with two other well-known techniques for online failure prediction: a straightforward method that is based on a reliability model and Dispersion Frame Technique (DFT). SEP is based on recognition of failure-prone patterns utilizing a semi-Markov chain in combination with clustering. We applied the approaches to real data of a commercial telecommunication system. Results are presented in terms of precision, recall, F-measure and accumulated runtime-cost. The results suggest a significantly improved forecasting performance.

A new method for predicting failures of a partially observable system is presented. System deterioration is modeled as a hidden, 3-state continuous time homogeneous Markov process. States 0 and 1, which are not observable, represent good and warning conditions, respectively. Only the failure state 2 is assumed to be observable. The system is subject to condition monitoring at equidistant, discrete time epochs. The vector observation process is stochastically related to the system state. The objective is to develop a method for optimally predicting impending system failures. Model parameters are estimated using EM algorithm and a cost-optimal Bayesian fault prediction scheme is proposed. The method is illustrated using real data obtained from spectrometric analysis of oil samples collected at regular time epochs from transmission units of heavy hauler trucks used in mining industry. A comparison with other methods is given, which illustrates effectiveness of our approach.

This paper describes a successful but challenging application of data mining in the railway industry. The objective is to optimize maintenance and operation of trains through prognostics of wheel failures. In addition to reducing maintenance costs, the proposed technology will help improve railway safety and augment throughput. Building on established techniques from data mining and machine learning, we present a methodology to learn models to predict train wheel failures from readily available operational and maintenance data. This methodology addresses various data mining tasks such as automatic labeling, feature extraction, model building, model fusion, and evaluation. After a detailed description of the methodology, we report results from large-scale experiments. These results clearly show the great potential of this innovative application of data mining in the railway industry.

T he ability to integrate distinct functional elements into a single device structure enables the realization of systems with higher-level functionality. Here we report on the design and fabrication of a fibre device structure that contains integrated optical, electrical and thermal elements for self-monitored optical transport. The fibre transmission element uses a hollow-core multilayer cylindrical photonic bandgap structure 1,2 designed to guide high-power radiation at 10.6 μm along the fibre axis 3 . Multiple thermal-detection elements are placed in the vicinity of the hollow core for the purpose of temperature monitoring along the entire fibre length. Metal wires bridged by a semiconductor layer extend along the length of the fibre and deliver an electrical response to the fibre ends on change in the fibre temperature. The multimaterial fibre is drawn at high speeds from a single preform 4 to produce extended lengths of optically and thermally functional fibres. The exponential dependence on temperature of the electrical conductivity of the semiconducting material allows for the discrimination, in real time, between normal transmission conditions and those that are indicative of localized defect formation, thus enabling a self-monitoring high-power optical transmission line for failure prediction and prevention.

We used a new rapid data acquisition system to examine the role of pre-existing crack density in determining the mechanical properties of crystalline rock. Two similar coarse-grained samples, Tsukuba granite (TG) and granitic porphyry (GP), were used in triaxial compression tests. The main difference between the two rocks is that TG contains main large pre-existing cracks, whereas GP is almost crack-free. In our tests, the two rocks showed significantly different behaviors. In GP, before the fault nucleation, AE activity was low and showed increasing b-values with increasing stress. But in TG, a large number of AEs were observed and showed short-term b-value anomalies-mutual fluctuations on a decreasing background. The short-term fluctuations of bvalues are closely correlated with those of the spatiotemporal clustering of AE locations. Our results demonstrate: 1) pre-existing cracks are the most dominant factor of all heterogeneities that govern the fault nucleation process in laboratory rock samples; 2) failure predictability is dependent on the pre-existing crack density.

Proactivity in maintenance, which is mainly materialized by degradation-based anticipation, becomes essential to avoid failure situation with negative impact on product and/or system conditions. It leads to make emerging the E-maintenance philosophy to move from "fail and fix" maintenance practices to "predict and prevent" strategies. Within these new strategies, the anticipation action is fully supported by prognosis business process. Indeed it analyses the degradation impact on the component itself but also on the global performances of the production system in order to predict future failures of the system and investigate (future maintenance) actions. However, only few research works focuses on generic and scalable prognostic approach. Existing methods are generally restricted on component view and for solving the failure prediction issue. Consequently, the contribution presented in this paper aims at developing a global formalization of the generic prognosis business process. This generic process can be used after, from an instantiation procedure, to develop specific prognosis processes related to particular application such as shown in this paper with the case of E−maintenance platform developed within DYNAMITE Project.

Previous experimental evidence has shown scale sensitivity of adhesives both in terms of ultimate stress and strain to failure; this sensitivity may be modelled using Weibull statistics. The implications of this on the sensitivity of strength predictions to mesh density around singularity points and small changes in local geometry is investigated using an idealised model of an enclosed corner in an adhesive joint. The sensitivity of two other failure criteria are also compared to the Weibull method. It is seen that there is a critical relationship between the Weibull shape parameter, m, and the singularity strength, . This relationship marks the transition between mesh insensitivity and mesh sensitivity, even with a singular stress (or strain) field present. If the singularity is removed by localised rounding, then the Weibull method of failure prediction shows much less sensitivity to the exact local geometry variation than other methods of failure prediction.

The authors examined relations between implicit and explicit Hope of Success (HS) and Fear of Failure (FF) and memory and liking for successful and unsuccessful peers. Implicit motives were expected to predict memory and explicit motives to predict liking of peers. Results from 106 American and 79 Singaporean students supported the implicit-explicit motive distinction as well as a link between HS and the successful peer and between FF and the unsuccessful peer. In both samples, explicit HS predicts liking while implicit HS predicts memory about the successful peer, and implicit FF predicts memory about the unsuccessful peer. Findings indicate achievement motives affect interest in and reaction towards peers and provide evidence for differential validity of implicit and explicit motives.

As the scale of high performance computing (HPC) continues to grow, application fault resilience becomes crucial. In this paper, we present FT-Pro, an adaptive fault management approach that combines proactive migration with reactive checkpointing. It aims to enable parallel applications to avoid anticipated failures via preventive migration, and in the case of unforeseeable failures, to minimize their impact through selective checkpointing. An adaptation manager is designed to make runtime decisions in response to failure prediction. Extensive experiments, by means of stochastic modeling and case studies with real applications, indicate that FT-Pro outperforms periodic checkpointing, in terms of reducing application completion times and improving resource utilization, by up to 43%.

Despite great efforts on the design of ultra-reliable components, the increase of system size and complexity has outpaced the improvement of component reliability. As a result, fault management becomes crucial in high performance computing. The advance of fault management relies on effective failure prediction. Despite years of research on failure prediction, it remains an open problem, especially in large-scale systems. In this paper, we address the problem by presenting a dynamic meta-learning prediction engine. It extends our previous work by exploring dynamic training, testing and prediction. Here, the "dynamic" part is from two perspectives: one is to continuously increase the training set during the system operation; and the other is to dynamically modify the rules of failure patterns by tracing prediction accuracy at runtime. Our case study indicates that the proposed predictor is promising by being capable of capturing more than 70% of failures, with the false alarm rate less than 10%.

To facilitate proactive fault management in largescale systems such as IBM Blue Gene/P, online failure prediction is of paramount importance. While many techniques have been presented for online failure prediction, questions arise regarding two commonly used approaches: period-based and event-driven. Which one has better accuracy? What is the best observation window (i.e., the time interval used to collect evidence before making a prediction)? How does the lead time (i.e., the time interval from the prediction to the failure occurrence) impact prediction arruracy? To answer these questions, we analyze and compare period-based and event-driven prediction approaches via a Bayesian prediction model. We evaluate these prediction approaches, under a variety of testing parameters, by means of RAS logs collected from a production supercomputer at Argonne National Laboratory. Experimental results show that the periodbased Bayesian model and the event-driven Bayesian model can achieve up to 65.0% and 83.8% prediction accuracy, respectively. Furthermore, our sensitivity study indicates that the event-driven approach seems more suitable for proactive fault management in large-scale systems like Blue Gene/P.