Reliability-Aware Speedup Models for Parallel Applications with Coordinated Checkpointing/Restart

Goodness-of-Fit (GOF) tests for the non-repairable systems as modelled through exponential, Weibull, normal and lognormal failure distributions are well touched upon in the literature. Substantial efforts have been made in developing GOF tests for non-repairable systems and Non-Homogeneous Poisson Process (NHPP) modelled through Power-Law Process (PLP). However, the literature is found to be limited in developing GOF models for imperfect repair modelled through Generalised Renewal Process (GRP). The paper, besides reviewing six trend tests, also develops a GOF test for repairable systems modelled through GRP based on Kijima I (KI) virtual age concept. KI takes into consideration the effect of repair effectiveness along with the shape and scale parameters. The developed GOF test model is a modification of the present Cramer-Von Mises (CVM) GOF test model available for PLP. The efficacy of the model is demonstrated with the help of three failure data sets.

A return on investment analysis of applying health monitoring to LED lighting systems

This paper presents an experimental work investigating the failure data of oil-impregnated paper under pulsating DC voltage. The failure data of the oil-impregnated paper specimens were obtained through constant voltage stress tests; the Weibull distribution was used to estimate the statistical distribution of failure data. The exponential function (EF) and the inverse power function (IPF) were used to modify the Weibull distribution of failure data, so as to obtain two dual-variable Weibull failure distributions (WFDs), which were used to analyze failure data of oil-paper specimens obtained under pulsating DC and individual DC and AC voltages. Their adequacies were evaluated by using failure data obtained through the constant stress tests under various voltages. Both EF-modified and IPF-modified WFDs show more adequacy for failure data of oil-paper insulation under DC and pulsating DC voltages than AC voltages. The EF-modified WFD was more adequate than IPF-modified WFD in estimating the statistical distribution of failure data.

This article revisits the optimization problems of age-based replacement models for exponential failure distributions. First, we show the steps how an age replacement time goes to an infinite value when the lifetime of a unit has an exponential distribution, and then, we give two discussions of possible replacement times, by introducing a new concept of cost named as deviation cost per unit of time between replacement and failure. Second, similar discussion is given for a random replacement model when the age replacement time is a random variable with a uniform distribution. Third, the models of age-based replacement first and last with deviation costs are discussed when a random replacement is taken into account for an age replacement model. Numerical examples are illustrated to show the results of the analytical discussions. Finally, necessary proofs are provided in the Appendixes.

A hard real-time control system, such as a fly-by-wire system, fails catastrophically (e.g., lose stability) if its control input is not updated by its digital controller computer within a certain time limit called the hard deadline. To assess and validate system reliability by using a semi-Markov model that explicitly contains the deadline information, we propose a path-space approach deriving the upper and lower bounds of the probability of system failure. These bounds are derived by using only simple parameters, and they are especially suitable for highly-reliable systems which must recover quickly. Analytical bounds are derived for both exponential and Weibull failure distributions, which have proven effective through numerical examples, while considering three repair strategies: repair-as-good-as-new, repair-as-good-as-old, and repair-better-than-old.

On the impact of process replication on executions of large-scale parallel applications with coordinated checkpointing

The study of the performance of parallel applications may have different reasons. One of them is the planning of the execution architecture to guaranty a quality of service. In this case, it is also necessary to study the computer hardware, the operating system, the network and the users behavior. This paper proposes different models for those based on simulation. First results of this work show the impact of the network and the necessary precision of communication model in fine-grained parallel applications.

A systematic reliability analysis procedure for evaluating fault-tolerant inertial measurement unit (IMU) architectures is described. This reliability analysis procedure is based on modeling the system architecture, the component reliabilities, and the redundancy management. The component reliabilities are based on constant failure rates and exponential failure distributions. The procedure computes both the overall reliability of the IMU and the major contributors to the IMU reliability. This reliability analysis procedure is used to evaluate and compare three state-of-the-art fault-tolerant IMU architectures: the triple triad architecture, the dual quad architecture, and the hexad architecture. >

It is common in applications to ignore the influence of sample size and population size when evaluating confidence lower limit of Mean Time to Failure (MTTF) for products with exponential failure distribution. The conventional method assumes that failure times or lives of different samples are independent identically distributed random variables. This assumption holds reasonably when the population size is much larger than the sample size. However, it will induce substantial inaccuracies when the sample size is not negligible to the population. To this problem, the influence from the sample size to the confidence limit of MTTF is analyzed when the population is determined. First, the conventional method is reviewed. Then, the formulae based on hypergeometric distribution are derived to reflect the relationships among sample size, population size, confidence level and confidence limit. Finally, an example is presented to illustrate the suitability of the proposed method and to show the computation errors from the conventional method.

This paper describes and evaluates the use of aggressive static analysis in Jackal, a fine-grain Distributed Shared Memory (DSM) system for Java. Jackal uses an optimizing, source-level compiler rather than the binary rewriting techniques employed by most other fine-grain DSM systems. Source-level analysis makes existing access-check optimizations (e.g., access-check batching) more effective and enables two novel fine-grain DSM optimizations: object-graph aggregation and automatic computation migration.The compiler detects situations where an access to a root object is followed by accesses to subobjects. Jackal attempts to aggregate all access checks on objects in such object graphs into a single check on the graph's root object. If this check fails, the entire graph is fetched. Object-graph aggregation can reduce the number of network roundtrips and, since it is an advanced form of access-check batching, improves sequential performance.Computation migration (or function shipping) is used to optimize critical sections in which a single processor owns both the shared data that is accessed and the lock that protects the data. It is usually more efficient to execute such critical sections on the processor that holds the lock and the data than to incur multiple roundtrips for acquiring the lock, fetching the data, writing the data back, and releasing the lock. Jackal's compiler detects such critical sections and optimizes them by generating single-roundtrip computation-migration code rather than standard data-shipping code.Jackal's optimizations improve both sequential and parallel application performance. On average, sequential execution times of instrumented, optimized programs are within 10% of those of uninstrumented programs. Application speedups usually improve significantly and several Jackal applications perform as well as hand-optimized message-passing programs.

Optimal maintenance policies for systems subject to a Markovian operating environment

A single repairman maintains a series system whose components have exponential failure and repair distributions. It is desired to assign the repairman to maximize the expected total discounted working time or average working time for the system. Whenever the failure rates of all components are equal, the time to repair the system from any initial state is stochastically independent of the operating policy. The Laplace transform of time to repair the system and the long-run average time that the system works are explicitly available. For the two-component series system the repair time is stochastically minimized if the failed component with longest expected lifetime is always under repair. This is true even if the repairman experiences random periods of time during which repair is not allowed. For an n-component series system the policy that always repairs the failed component with longest expected lifetime stochastically minimizes the repair time over the class of policies, that repair failed components in a predetermined order. We apply the results to a queuing system.

To make efficient use of multi-core processors, it is important to understand the performance behavior of parallel applications. Modeling this can enable the use of online approaches to optimize throughput or energy, or even guarantee a minimum QoS. Accurate models would avoid probe different runtime configurations, which causes overhead. Throughout the years, many speedup models were proposed. Most of them based on Amdahl's or Gustafson's laws. However, many of those make considerations such as a fixed parallel fraction, or a parallel fraction that varies linearly with problem size, and inexistent parallelization overhead. Although such models aid in the theoretical understanding, these considerations do not hold in real environments, which makes the modeling unsuitable for accurate characterization of parallel applications. The model proposed estimates the speedup taking into account the variation of its parallel fraction according to problem size, number of cores used and overhead. Using four applications from the PARSEC benchmark suite, the proposed model was able to estimate speedups more accurately than other models in recent literature.

We extend a diagnostic plot for the frailty distribution in proportional hazards models to the case of shared frailty. The plot is based on a closure property of exponential family failure distributions with canonical statistics z and g(z), namely that the frailty distribution among survivors at time t has the same form, with the same values of the parameters associated with g(z). We extend this property to shared frailty, considering various definitions of a “surviving” cluster at time t. We illustrate the effectiveness of the method in the case where the “death” of the cluster is defined by the first death among its members.

Scalability models are powerful analytical tools for evaluating and predicting the performance of parallel applications. Unfortunately, existing scalability models do not quantify failure impact and therefore cannot accurately account for application performance in the presence of failures. In this study, we extend two well-known models, namely Amdahl's law and Gustafson's law, by considering the impact of failures and the effect of fault tolerance techniques on applications. The derived reliability-aware models can be used to predict application scalability in failure-present environments and evaluate fault tolerance techniques. Trace-based simulations via real failure logs demonstrate that the newly developed models provide a better understanding of application performance and scalability in the presence of failures.