 |
The Seventh International Conference on Performance, Safety and Robustness in Complex Systems and Applications PESARO 2017 |
| Tutorials |
T1: Autonomic Management and Security for Cloud and Internet of Things
Carlos Becker Westphall, University of Santa Catarina, Brazil
Carla Merkle Westphall, University of Santa Catarina, Brazil
T2: From Vehicular Ad hoc Networks to Internet of Vehicles
Eugen Borcoci, University "Politehnica"of Bucharest (UPB), Romania
T3: Statistical Methods for System Dependability: Reliability, Availability, Maintainability and Resiliency
Andy Snow, Ohio University, USA
T4: Unraveling Stories from Your Massive Datasets
Venkat N Gudivada, East Carolina University, USA
T5: Dependability Analysis through Monte Carlo Methods: The Case of Rare Events
Gerardo Rubino, INRIA, Rennes, France
T6: Bringing ICT into Newborn Monitoring: A Video-Based Approach
Riccardo Raheli, University of Parma, Italy
Detailed information
T1: Autonomic Management and Security for Cloud and Internet of Things
Carlos Becker Westphall, University of Santa Catarina, Brazil
Carla Merkle Westphall, University of Santa Catarina, Brazil
Autonomic Cloud Computing management requires a model to represent the elements into the managed computing process. This tutorial proposes an approach to model the load flow through abstract and concrete Cloud components. Our model has a formal mathematical background and is generic, in contrast with other proposals. It receives new Virtual Machines on the Cloud and organizes them by relocating their placements based on the Multiple-Objectives of the environment. These Objectives are represented by Rules, Qualifiers and Costs, which can be easily added, extended and prioritized. In contrast to existing solutions, that address specific objectives, our framework was devised to be objective-agnostic and easily extensible, which enables the implementation of new and generic prioritized elements.
Our work proposes an autonomic intrusion response technique that uses a utility function to determine the best response to the attack providing self-healing properties to the environment. Cloud computing allows the use of resources and systems in thousands of providers. This paradigm can use federated identity management to control user's identification data, but it is essential to preserve privacy, while performing authentication and access control.
This tutorial describes a model where the cloud consumer can perform risk analysis on providers before and after contracting the service. We motivate the use of risk-based access control in the cloud and present a framework for enforcing risk-based policies.
T2: From Vehicular Ad hoc Networks to Internet of Vehicles
Eugen Borcoci, University "Politehnica"of Bucharest (UPB), Romania
Prerequisites: general knowledge on IP networking architectures, protocols, introductory knowledge on VANET, SDN, NFV and Cloud computing.
The vehicular communications have been constantly developed in the last two decades, as part of the Intelligent Transport System (ITS). Specific technologies like Dedicated Short-Range Communications (DSRC) and architectural stacks like Wireless Access in Vehicular Environments (WAVE) are components of the emerging ITS market.
The traditional ITS has evolved, including vehicle to vehicle (V2V), vehicle to road (V2R), or more general vehicle to Infrastructure (V2I) communication types, commonly denoted as Vehicular Adhoc Networks (VANET). However, VANETs have limitations; in spite of their good potential to contribute in solving safety and traffic management problems with low operational cost, they did not attract very high commercial interest. The limitations are related to pure adhoc network architecture (in V2V case), unreliable Internet service, incompatibility with personal devices, low or non-cooperation with cloud computing, low accuracy of the services, and operational network dependency.
A recent novel solution will be the Internet of Vehicles (IoV), seen as a global network; it can be considered as special case of Internet of Things. The IoV objectives include vehicles driving (classic goal - in VANET), but also others - like vehicle traffic management in urban or country areas, automobile production, repair and vehicle insurance, road infrastructure construction and repair, logistics and transportation, etc. The IoV is not a clean slate solution; it includes many of the previous concepts and technologies of the VANETs. Generally, it is estimated that smart-cities systems will include a strong IoV component.
This tutorial will provide a high level overview of transition from VANETs to IoV: applications, architecture and protocols proposed for IOV, novel communication types, network models and heterogeneous access infrastructures (WiFi, 4G, 5G), etc. Several recent technologies are considered as candidate to support or cooperate with IoV, i.e., Cloud/Fog computing, Software Defined Networking and Network Function Virtualization. However functional harmonization of these components raises several challenges, from conceptual, architectural and design point of view. These are outlined in this tutorial.
T3: Statistical Methods for System Dependability: Reliability, Availability, Maintainability and Resiliency
Andy Snow, Ohio University, USA
I. Overview
The goal of this tutorial is to provide attendees a working knowledge of the statistical methods used to assess complex system dependability. The dependability attributes reliability, availability, maintainability, and resiliency, are covered. Empirical techniques of assessing and forecasting these attributes are provided along with examples of using actual outage data.
II. List of Tutorial Modules and Content
A. Hazards
Threats (natural and manmade)
Vulnerabilities
Faults Taxonomy
Service Outages
Single Points of Failure
Over-Concentration
Risk as a f(Severity, Likelihood)
Protection through fault prevention, tolerance, removal, and forecasting
Best Practices
B. Dependability Attributes: Reliability, Availability, Maintainability and Resiliency
Point Processes:
Homogeneous Poisson Process (HPP)
Renewal Processes (RP)
Branching Poisson Processes (BPP)
Non-Homogenous Poisson Process (NHPP)
Reliability – f (MTTF)
Maintainability – f (MTTR)
Availability – f ( MTTF, MTTR)
Resiliency – f ( MTTF, MTTR, Severity)
User and Service Provider Perspectives of Dependability
D. Statistical Dependability Assessments & Forecasting
Data Collection Requirements
Outage Cause Classification and Analysis (trigger, direct, and root causes)
Trend Assessments (Graphical, Laplace, Lewis-Robinson, Mil-Handbook tests)
Poisson Regression, Simulation and Artificial Neural Networks
IT and Telecom Case Studies
Audience
The tutorial will be very useful for those looking to understand the statistical methods for assessing complex system dependability. University professors, graduate students, and industry professionals are likely to benefit from this tutorial.
T4: Unraveling Stories from Your Massive Datasets
Venkat N Gudivada, East Carolina University, USA
This tutorial is aimed at educators, students, and software professionals who are interested analyzing and interpreting massive datasets using various machine learning techniques. According to IBM, every day we create over 2.5 quintillion bytes of data. The magnitude of this data is so much that 90% of the data in the world today has been created in the last two years alone. This big data is creating unprecedented opportunities for scientific discoveries and business innovation.
This two-hour tutorial will provide an overview of various machine learning algorithms and techniques. Also, these techniques will be demonstrated on real datasets using R, which is open source software for statistical computing and machine learning. The structure of the tutorial is as follows:
- Introduction and roadmap (5 min)
- Supervised, semi-supervised, and unsupervised learning (10 min)
- Parametric and non-parametric learning (5 min)
- Linear and nonlinear learning algorithms (5 min)
- Ensemble learning, bagging, and random forest (10 min)
- Bias and variance trade-off (5 min)
- Linear regression (15 min)
- Break (5 min)
- Classification (15 min)
- Tree-based methods (15 min)
- Support Vector Machines (15 min)
- Unsupervised learning (15 min)
- Summary and conclusions (5 min)
Participants are encouraged to bring their laptop computers for hands-on exploration. Install R system (https://www.r-project.org/) and the following packages: AppliedPredictiveModeling
(https://cran.r-project.org/web/packages/AppliedPredictiveModeling/index.html), mlbench (https://cran.rproject. org/web/packages/mlbench/index.html), and datasets
(http://stat.ethz.ch/R-manual/R-devel/library/datasets/html/00Index.html).
Please note that this step is optional.
T5: Dependability Analysis through Monte Carlo Methods: The Case of Rare Events
Gerardo Rubino, INRIA, Rennes, France
To analyze dependability properties of complex systems, engineers use models on which specific metrics are evaluated. For this purpose, different types of techniques exist, usually referred to as analytical ones, including approximations (bounds, for instance), numerical ones, also including approximations, and simulation (here also called Monte Carlo). With respect to the range of models that can be analyzed with them, by far the most powerful methods are the latter, but there is a main drawback associated with them: if the event of interest is rare (if it has a very small probability), then Monte Carlo procedures are in trouble, if they are used in a straightforward manner. Face to rare events, the analyst must use much more sophisticated approaches, that make the topic a more complex and open one.
This tutorial will first describe the context, that is, the most used dependability metrics, and it will continue with a brief description of the available analytical and numerical approaches to evaluate them on models. Then, we will focus on Monte Carlo methods in general, and we will introduce the rare event problem. The tutorial will then concentrate on specific techniques that have been proved efficient to attack the problem, including Importance Sampling and its hottest subfamily, the zero-variance approach, Splitting methods, and Recursive Variance Reduction ones. Both static and dynamic models will be considered, corresponding to system without and with repair capabilities.
T6: Bringing ICT into Newborn Monitoring: A Video-Based Approach
Riccardo Raheli, University of Parma, Italy
In both clinical and domestic environments, newborns deserve continuous attention from the medical personnel or the caring parents. In a Neonatal Intensive Care Unit (NICU), neonates affected by perinatal diseases are at risk of neonatal seizures, which are the most common sign of acute neurological dysfunctions and must be promptly and accurately recognized in order to establish timely treatments. In a domestic scenario, respiration disorders and the possible occurrence of apnoea episodes may be related with a potential risk of Sudden Infant Death Syndrome (SIDS) and should be immediately reported to a pediatrician. All these potential events may occur unexpectedly and with low rate, causing a non negligible risk of letting their initial occurrence go unnoticed with possible detriment to the health of the newborn.
Continuous wide-scale newborn monitoring by caring personnel is of course unfeasible, even if we restrict our interest to a subset of the population which may exhibit a higher risk of disorders. As a consequence, there has been interest in devising automatic real-time low-cost systems, based on Information and Communication Technology (ICT), capable of continuously monitoring a newborn and prompting the attention of the caring personnel in a timely and reliable manner. Among various options, one that appears particularly convenient and appealing, both from the scientific and application viewpoints, is the use of one or multiple video cameras positioned around the cradle and framing the newborn, equipped with proper signal processing algorithms designed to detect the occurrence of possible disorders and alert the caring personnel.
This tutorial will provide an overview of video signal processing methods for newborn monitoring which have been the subject of research and experimentation in recent years. As a specific case study, we focus on monitoring infants potentially at risk of diseases characterized by the presence or absence of rhythmic movements of one or more body parts, such as the limbs, the chest or the abdomen. In fact, a specific category of neonatal seizures, named “clonic”, are characterized by repetitive movement patterns of some body parts, whose possible presence can be detected automatically by a video processing system. Likewise, respiration monitoring can be considered in which possible apnoeas can be detected by the temporary absence of repetitive movement patterns.
After introducing the subject and providing an overview of earlier work, we shall present the principles underlying the extraction of relevant information content from video signals. We shall then present specific video-based solutions to newborn monitoring, their performance and the results of initial experiments in a real NICU environment. Finally, we shall discuss potential applications of these methods to respiration monitoring in a domestic environment employing low-cost devices, such as smartphones or tablets.
