This paper describes a newly developed Data Science major that is jointly offered between two departments (Computer Science and Statistics) and across two colleges (College of Engineering and College of Natural and Architectural Sciences). The paper explores the inspiration and motivation for creating the undergraduate Data Science program, the program curriculum, and the governing body of the program. We aim to recruit undergraduates, specifically women, into the new major which are underrepresented in Computer Science at our University. We expect that this Major in Data Science will serve as a general framework for establishing similar interdisciplinary programs.

Obtaining lower confidence bounds on the reliability of a series system remains a problem of interest. Engineers and practitioners desire methods with good properties to obtain lower confidence bounds when only component-level binomial test data are available. With numerous methods proposed in the literature-none of which are uniformly superior-it can be an overwhelming task to select the method which best handles the scenario at hand. We develop a software tool in R (available in the “serieslcb” package) through which users can discover the methods that best suit their situation. The tool runs user-defined simulations and then ranks the best performing methods based on objective comparisons utilizing a delta coverage metric. This article outlines the methods and comparison strategy implemented in the package. It also discusses the design of the software tool and illustrates it with an example.

Accurate clock synchronization is essential in the operation of Sensor Networks. Algorithms for clock synchronization generally rely on estimators of the relative offset and frequency of two different clocks. Sensor Networks have limited power resources so it is important that these estimators not be too computationally intensive. For this reason a number of least squares-based estimators have been proposed, however they are not appropriate for sensor networks with significant propagation latency. For these applications, the Time Synchronization for High Latency (TSHL) protocol was developed. We seek to improve the accuracy of TSHL and thus propose, study, and demonstrate the effectiveness of several alternative least squares-based estimators.

Availability is an important attribute of optical transport networks, and is enabled through many alternative restoration schemes. Different restoration schemes tradeoff network cost, as measured by provisioned capacity, and network reliability to match diverse application needs. We provide a systematic quantification of these tradeoffs for commonly employed restoration schemes and illustrate them for several networks. In this paper, we analyze a dedicated capacity 1+1 protection scheme and two commonly employed shared capacity mesh restoration schemes: path and link based restoration

We classify system failures into three categories: hardware failures, software failures, and hardware-software interaction failures. We develop a unified reliability model that accounts for failures in all three categories. Hardware, and software failures are accounted for with well-known modeling approaches. In this paper, we propose a modeling methodology using Markov processes to capture hardware-software interaction failures. We illustrate the combined hardware & software modeling approach by applying it to a real telecommunication system

Information Discovery and Analysis Systems (IDAS) are designed to correlate multiple sources of data and use data mining techniques to identify potential significant events. Application domains for IDAS are numerous and include the emerging area of homeland security.Developing test cases for an IDAS requires background data sets into which hypothetical future scenarios can be overlaid. The IDAS can then be measured in terms of false positive and false negative error rates. Obtaining the test data sets can be an obstacle due to both privacy issues and also the time and cost associated with collecting a diverse set of data sources.In this paper, we give an overview of the design and architecture of an IDAS Data Set Generator (IDSG) that enables a fast and comprehensive test of an IDAS. The IDSG generates data using statistical and rule-based algorithms and also semantic graphs that represent interdependencies between attributes. A credit card transaction application is used to illustrate the approach.

Delay measurements from timing message exchanges between two clocks produce a maximum-likelihood estimator (MLE) of the clock offset when fixed delays in each direction are equal and unknown, and variable delays in each direction have an exponential distribution with an unknown mean. It is shown that the MLE corresponds to a previously proposed estimator of clock offset. The ML interpretation of the estimator provides further insight and motivation for its use.

Software test environments are often different from field environments. Using test data exclusively to estimate a field failure rate will not usually give a very accurate estimate. In this paper, we extend an empirical calibration methodology for adjusting the failure rate estimate obtained from analysing test data. In addition to scaling the estimated failure rate of a fault, we propose scaling the estimated number of residual faults as well. We also derive likelihood ratio tests to formally determine (from previous releases of the software) if test, and field environments are significantly different. We illustrate our new results with two telecommunications case studies. The combination of the likelihood ratio test, and the calibration methodology offers a practical way to extend the application of software reliability growth models to less formal test environments.

A generalized blocking model is derived for a switch access system that was investigated by Doshi, Dravida, Jeske, Murti and Samadi ( 1998) by relaxing the assumption that the line unit is a cross bar switch. The need for this arises from the fact that in some switch access architectures, the number of servers in a server group is smaller than the number of timeslots in the bus connecting the line units to the server groups. Lower and upper bounds on the blocking probability are derived to incorporate the architecture changes. The bounds are then applied to practical switch access systems and shown to be tight and useful.

We study the delay characteristics of voice packets that share a transmission medium with bursty data traffic. In order to model the data traffic accurately, we used real data traffic traces upon which we superimposed simulated voice sources generating real time traffic at a constant rate. Our simulation environment allows us to easily develop and compare different adaptive algorithms for the play-out delay of voice streams. We motivate and compare the performance of three algorithms based on exponential smoothing techniques. We test the algorithms in the context of a shared 10 Mbps Ethernet LAN, using the CSMA/CD protocol.

The signal-to-interference ratio (SIR) is an important metric of communication link quality. For wireless data systems SIR estimates are used in several control actions and for optimal allocation of radio resources. This paper first develops SIR estimators based on pilot symbols and then extends the results to estimators based on data symbols. It is shown that long-term smoothing of the interference variance estimates improves the overall estimation accuracy. Since both pilot and data symbols are available, estimators that optimally combine estimates from both types of symbols are also considered. In all cases, approximate analytic results are provided and validated using simulations. The analytic expressions serve two purposes. Firstly, improved estimators based on bias correction become apparent from the expressions and secondly, optimal weights for combining pilot and data based estimators can be evaluated. Results show that modest to substantial improvement in estimation accuracy is possible through optimal combining.

A realistic estimate of the field failure rate of software is essential in order to decide when to release the software while maintaining an appropriate balance between reliability, time-to-market and development cost. Typically, software reliability models are applied to system test data with the hope of obtaining an estimate of the software failure rate that will be observed in the field. Unfortunately, test environments are usually quite different from field environments. In this paper, we use a calibration factor to characterize the mismatch between the system test environment and the field environment, and then incorporate the factor into a widely used software reliability model. For projects that have both system test data and field data for one or more previous releases, the calibration factor can be empirically evaluated and used to estimate the field failure rate of a new release based on its system test data. For new projects, the calibration factor can be estimated by matching the software to related projects that have both system test data and field data. In practice, isolating and removing a software fault is a complicated process. As a result, a fault may be encountered more than once before it is ultimately removed. Most software reliability growth models assume instantaneous fault removal. We relax this assumption by relating non-zero fault removal times to imperfect debugging. Finally, we distinguish between two types of faults based on whether their observed occurrence would precipitate a fix in the current or future release, respectively. Type-F faults, which are fixed in the current release, contribute to a growth component of the overall failure rate. Type-D faults, whose fix is deferred to a subsequent release, contribute to a constant component of the overall software failure rate. The aggregate software failure rate is thus the sum of a decreasing failure rate and a constant failure rate.

When software is tested according to the customer's operational profile, many techniques exist in the literature for using the resulting failure data to estimate the customer-perceived failure rate. Unfortunately, the overhead associated with defining and using the customer's operational profile during system test can be significant. As a compromise, many projects test the software by executing it over a sustained period of time using simulated load scenarios that represent informed judgments of what the customer environment is likely to be. By itself, the load test data is suspect for estimating the customer-perceived failure rate since the load scenarios may not be accurate representations of the customer's usage patterns. In addition, the load test data is limited by the relatively small amount of time that is set aside for load testing. In many cases the software being tested is an evolution of a previous release already in the field. The field experience with the previous release along with a characterization of the difference between the two releases are natural sources of information for building a prior distribution of the failure rate of the new release. The load test data can then be used to obtain a posterior distribution of the failure rate of the new release. The posterior distribution inherently reflects the customer's operational profile through its dependence on the field data, and effectively enlarges the set of data from which inference about the failure rate of the new release is to be made. We develop a model for the prior distribution of the failure rate, and obtain the corresponding posterior distribution. We illustrate our modeling approach using real data collected from one of our software products.

The software reliability literature contains several models for estimating software reliability at the end of system test. However, to enable better planning, project management often needs reliability predictions prior to system test. The paper presents a software reliability prediction model that extends the known results on software predictions to the case of a distributed software architecture. The model predictions are based on software code characteristics and the size of code as distributed on the system processors. In particular, the implications of having overlapping code executing on multiple processors are examined. It is shown that the predictions, under suitable assumptions on reliability growth during system test, can be used by project management to plan their test and release time intervals. To illustrate the ideas, the prediction model is applied to a telecommunications software project. The paper explains different methods of obtaining the values of software characteristics that are needed as input to the model.

Consider m offices with each operating a duplexed unit. Suppose a centrally located warehouse is to be provisioned with k spares to be used when offices experience failures. Assume that failed units are instantaneously replaced with spare units, provided a spare is available at the warehouse and the failed units are repaired and used to replenish the spare inventory. We consider two approaches for determining a suitable value for k. In the first approach, a network emergency is declared whenever the number of failed units in the network exceeds k. A simple expression for the mean time to a network emergency (MTNE) is obtained using a birth and death process model. The value of k is selected to make MTNE satisfactorily large. In this approach, it is conservatively assumed every network emergency will be service affecting. However, only when the number of failures exceeds k AND at least one office has two failures is the incident service affecting. Our second approach accounts for this distinction by evaluating the mean time to a service affecting incident (MTSI). The two approaches are compared by contrasting the different valves of k each requires to achieve a specified objective for MTNE and MTSI, respectively.

We introduce the concept of proxy failure times for situations where system test data only consists of the fraction of test cases that fail for a set of execution scenarios. We show how proxy failure times can be simulated if external information about the user frequency of the test cases is available. We develop statistical inference procedures for fitting the Jelinski-Moranda model (Z. Jelinski and P. Moranda, 1996). In particular, we present a graphical diagnostic for testing goodness of fit and show how it suggests appropriate transformations of the failure times that would improve the fit. Influential observations are also identified by the diagnostic and moreover, it provides regression estimators of the model parameters as a quick alternative to the maximum likelihood estimators. Formulas for likelihood based confidence intervals for the model parameters are provided. The simulation of proxy failure times and the statistical inference procedures for the Jelinski-Moranda model are illustrated with an example.

The distribution of the cumulative downtime for a highly reliable component over an interval of time is approximated using a compound Bernoulli process. Given a set of observed cumulative downtimes, the maximum likelihood (ML) and uniformly minimum variance unbiased (UMVU) estimator of the approximate cumulative downtime distribution are derived under the assumption of exponential repair times and are compared to the nonparametric estimator. The ML estimator is more efficient than the UMVU estimator which itself is more efficient than the nonparametric estimator.