Search

Performance testing modeling is a relative new research field. Researches investigating how to apply models to document performance testing information essentially started to be reported in the last decade. Motivated by the lack of a standard to represent performance testing information, our research group, in collaboration with an IT company, proposed a UML approach and lately a Domain-Specific Language (DSL) to support performance testing modeling. The goal of this paper is to show how we support our partner company on the decision process to replace UML by a DSL, hence we designed and conducted an experimental study to provide evidence about the benefits and drawbacks when using UML or DSL for modeling performance testing. In this paper, we report an in vitro experiment, where the subjects designed UML models and DSL models, for the purpose of evaluation with respect to the effort and suitability, from the perspective of the performance testers and the performance engineers in the context of industry and academia for modeling performance testing. Our results indicate that, for performance modeling, effort using a DSL was lower than using UML. Our statistical analysis showed that the results were valid, i.e., that to design performance testing models using our DSL is better than using UML.

There is an increasing number of web data that consists of text and structured data, such as the combination of Wikipedia pages and DBpedia data. To issue queries to such data, we must choose one of the following: (1) submit keyword queries against textual data or (2) submit structured queries written in structured query languages such as SQL and SPARQL against structured data. This paper proposes MorphingAssist, a tool to assist users to gradually transform keywords into structured queries. A feature of our method is to use a hybrid query language that allows a variety of mixtures of keyword and structured query components in a query so that every intermediate step of transformation produces meaningful queries. However, assisting users to transform queries in such a way is a challenge. This paper explains how MorphingAssist generates and shows such hints to the user to facilitate his/her transforming simple keyword queries into more precise queries.

In this paper, we report on successful chaining of two unique model checkers, namely DIVINE and PRISM, which, as a whole, allows for practical verification of multi-threaded C++ programs that may choose input and other actions according to a given discrete probabilistic distribution. In the paper, we discuss technical details of the extensions of the DIVINE model checker that were required to enable the chaining, in particular, we report on combination of dynamic τ+reduction used within the DIVINE state space exploration engine with the probabilistic choice operator. We also give preliminary experimental evaluation of our approach, discuss some possible applications for the tool chain, and finally, we plot some of the future steps to be done.

This paper proposes a new method called Virtual Run for extracting the representative trajectories from a set of similar trajectories by readjusting the sequence of movements of vehicles. This method uses convex hull and Euclidean minimum spanning tree to move the trajectory sets, so that it can generate the representative trajectories. Those representative trajectories are used to generate the digital road map which is the final goal of this research. Generally, the well-known Frechet distance measure is applied to estimate the similarity between a pair of trajectories. However, in this paper, we present that our new method is more effective than the previous methods on the basis of an experiment and evaluation. Also we show the application of generating the digital road map using the presented method. For the evaluation, we collected a set of actual trajectory data from 3,000 taxis in the Gangnam and Seongnam areas. The experimental results show that our approach can solve the problems associated with the special case of loops, which are not clearly solved by the previous approaches. The results also show that our method is capable of generating a digital road map from a set of trajectories.

This paper introduces SAMSON (Self-Adaptive Middleware for wireless SensOr Networks), an instance of a Reference Architecture (RA) that was designed to manage wireless sensor networks (WSN) in an autonomic way by enabling runtime adaptation of nodes' behavior according to context changes, application requirements, and predefined adaptation policies. In this paper, we propose a process, supported by model-driven transformations to generate SAMSON from the RA specification. The model-driven transformations are used to map each element of the RA specification to software components and to generate the source code to be deployed in a target WSN platform. We evaluate our proposal by presenting four application scenarios where SAMSON was employed to optimize the network lifetime and to react to runtime events according to adaptation policies previously defined by software architects and network administrators. Evaluation results shown that SAMSON was able to dynamically adapt the behavior of sensor nodes, to efficiently use the hardware resources, and to increase the WSN lifetime.

Over the past few years, nonvolatile memory has actively been researched and developed. Therefore, studying operating system (OS) designs predicated on the main memory in the form of a nonvolatile memory and studying methods to manage persistent data in a virtual memory are crucial to encourage the widespread use of nonvolatile memory in the future. However, the main memory in most computers today is volatile, and replacing highcapacity main memory with nonvolatile memory is extremely cost-prohibitive. This paper proposes an OS structure for nonvolatile main memory. The proposed OS structure consists of three functions to study and develop OSs for nonvolatile main memory computers. First, a structure, which is called plate, is proposed whereby persistent data are managed assuming that nonvolatile main memory is present in a computer. Second, we propose a persistent-data mechanism to make a volatile memory function as nonvolatile main memory, which serves as a basis for the development of OSs for computers with nonvolatile main memory. Third, we propose a continuous operation control using the persistent-data mechanism and plates. This paper describes the design and implementation of the OS structure based on the three functions on The ENduring operating system for Distributed EnviRonment and describes the evaluation results of the proposed functions.

Context: The activities related to Requirements engineering (RE) are some of the most important steps in software development, since the requirements describe what will be provided in a software system in order to fulfill the stakeholders' needs. In this context, the ACM Symposium on Applied Computing (SAC) has been a primary gathering forum for many RE activities. When studying a research area, it is important to identify the most active groups, topics, the research trends and so forth. Objective: This study aims to investigate how the SAC RE-Track is evolving, by analyzing the papers published in its 8 previous editions. Method: We adopted a research strategy that combines scoping study and systematic review good practices. Results: We investigated the most active countries, institutions and authors, the main topics discussed, the types of the contributions, the conferences and journals that have most referenced SAC RE-Track papers, the phases of the RE process supported by the contributions, the publications with the greatest impact, and the trends in RE. Conclusions: We found 79 papers over the 8 previous SAC RE-Track editions, which were analyzed and discussed.

Enterprise applications are used for managing operational data and are aimed at improving business efficiency. Many enterprise applications have been developed over the past three decades and are often referred to as legacy systems. Usually they are monolith, inflexible, poorly documented and hard to maintain. These issues can be addressed by improving their design using decomposition. The purpose of this paper is to introduce a method for decomposition of large scale enterprise applications and to present initial practical evaluation results. The paper systematically defines the general decomposition process including its tailoring options. The general process is adapted and applied for application in the industrial case study investigated as part of the university-industry collaboration research project.

Markov Decision Processes (MDPs) are extensively used to model long-term sequential decision-making in stochastic situations. MDPs have received a lot of attentions from quantitative model checking [3], operations research [9], machine learning [4] and management science communities [7]. In order to reason efficiently about quantitative properties of systems modelled as MDPs, theoretically and practically efficient decision algorithms are of main interest. In this paper we consider completely ergodic Markovian decision processes with no discounting. We give a novel reduction from given MDPs to generalized flow networks which enables us to reformulate the corresponding LP as an LP with embedded network structure. The proposed reduction enables a combinatorial approach towards designing a strong polynomial time algorithm for finding strategies that minimize undiscounted limiting mean payoff over an infinite time horizon. The strong polynomial results can speed up the computation significantly in case of large-scale MDPs. The question whether the proposed problem in this paper can generally be solved in strongly polynomial time remains still open [9].

Similarity measurements in images have always been a challenging task in the field of pattern recognition techniques. Shape based features is a widely adopted method in Content Based Image Retrieval (CBIR) for similarity measurement. In this paper we proposed an enhanced version of Mesh Distance Fourier Descriptor (MDFD) previously developed in our lab for the similarity measurement. Two extra levels of filters have been added to the output of MDFD so that the final output is more refined and the most similar image to the query image is selected from the database. The first level filter includes processing of images retrieved based on ratio of area of the image to the area of minimum bounding rectangle enclosing that image. The second level filtering includes calculation of average of absolute difference of global features like eccentricity, convexity and solidity of the query image and retrieved image. Adding these two extra filtering levels, the matching ratio has been increased from 84% to 88% which shows adding filters enhances the results of MDFD. In this paper we have used binary images extracted from region of interest (ROI) of mammogram which are classified into single objects using known classification methods such as K-means and SVM algorithms.

This paper presents the main features and architecture of MHARS (Mobile Human Activity Recognition System), a pervasive system for monitoring the daily movement activities of patients in the context of Ambient Assisted Living. MHARS is based on smartphones, and allows data gathering from the different sensors usually available on this class of devices, as well as from external sensors, either the ones comprising a body sensor network as well as from ambient deployed sensors. MHARS uses accelerometer data for inferring the patient performed activity and her/his heart rate for computing its intensity. It also allows to correlate the inferred data with other context data (such as altitude, body and environment temperature, etc) for detecting user defined situations related with the patient current health status and provides a decision making engine for defining an action plan (set of actions) that must be executed whenever relevant health situations are detected. This paper emphasizes the MHARS service responsible for inferring the patient movement activity. We present a comprehensive set of experiments used for building the movement activity classifier and the evaluation of its performance.

The presentation of multimedia content requires devices with proper computational resources to guarantee intermedia synchronization requirements. When a multimedia system is executed in resource-constrained devices, some synchronization mismatches may occur. This problem becomes harder to solve when considering support for features like user interaction, systemic resilience, and distributed multimedia processing. This paper discusses the separation of concerns between multimedia presentation orchestration and multimedia content decoding and rendering. A multimedia system architecture and its implementation are also presented in this paper, where the separation of concerns goes beyond modularization by being deeply realized as process isolation. This design decision not only assures improved system resilience to the proposed solution but also enables it to be applied in distributed multimedia processing/presentations. Experiments carried out have indicated that the presentation orchestration does not require much resource, and may be run on lightweight devices. On the other hand, multimedia decoding/rendering is suitable to run on devices that are more resourceful.

Designing modern web applications involves a wide spectrum of choices when it comes to deciding where the different tiers of application and framework code that constitute these distributed applications should be placed. These system design choices affect programmer productivity, ease of deployment, security, and performance, particularly with respect to latency and scalability. In this paper, we propose and evaluate a design choice in which not only all application logic executes server-side, but most presentation logic as well. The client browser is reduced to a rendering and I/O engine, similar to a "thin client" or "dumb terminal," but retains the full expressiveness of rich, modern Internet applications. We have developed CloudBrowser 2.0, a system that implements this distribution model using a scalable multiprocess approach. In this paper, we perform an evaluation of the benefits and costs of this approach when compared to both more traditional approaches as well as emerging alternatives. We focus on programmability and systems aspects including performance and latency.

Novel technologies represent a promising way for boosting learning quality in scientific sectors, by engaging students more effectively into experimental activities. Mobile devices allow increasing learning effectiveness significantly, thanks to their diffusion, especially amongst youngsters. Built-in sensors can enable easily-deployable, low-cost and large-scale sensing networks, which exploit the Mobile Crowd-Sensing (MCS) paradigm and involve their owners directly into scientific monitoring activities. For such reasons, in this paper we present a MCS-based noise-sensing platform for engaging students in learning acoustics and sound metering. The proposed platform can be included into traditional didactic activities about those topics. This capability is fostered also by the Bring Your Own Device (BYOD) approach, which further eases the platform diffusion across students. Moreover, BYOD paves the way for the platform application as a MOOC (Massive Online Open Course) enabler. The paper presents the adopted design and implementation methodologies, as well as the platform prototype and its technical validation. We believe that our approach offers students new possibilities for learning how to use professional sound level meters profitably and how to perform measurements correctly.

A fundamental challenge in robotics is to reason with incomplete domain knowledge to explain unexpected observations, and partial descriptions of domain objects and events extracted from sensor observations. Existing explanation generation systems are based on ideas drawn from two broad classes of systems, and do not support all the desired explanation generation capabilities for robots. The objective of this paper is to first compare the explanation generation capabilities of a state of the art system from each of these two classes, using execution scenarios of a robot waiter assisting in a restaurant. Specifically, we investigate KRASP, a system based on the declarative language Answer Set Prolog, which uses an elaborate system description and observations of system behavior to explain unexpected observations and partial descriptions. We also explore UMBRA, an architecture that provides explanations using a weaker system description, a heuristic representation of past experience, and other heuristics for selectively and incrementally searching through relevant ground literals. Based on this study, this paper identifies some key criteria, and provides some recommendations, for developing an explanation generation system for robots that exploits the complementary strengths of the two classes of explanation generation systems.

International audience; The paper is devoted to effective check of the existence and derivation of adaptive distinguishing sequences (distinguishing test cases) for possibly nondeterministic partial Finite State Machines (FSMs). The complexity of these problems for nondeterministic FSMs remains unknown, however the length of the corresponding test case is shown to be exponential. In this paper, we address FSM classes that allow to derive specific FSM projections for which the problem of checking whether a given FSM is adaptively distinguishing or not can be performed in polynomial time. In order to estimate the length of distinguishing test cases for FSMs of these classes we improve the upper bound on the length of an adaptive distinguishing test case for partial deterministic FSMs. The listed contributions make it possible to apply the proposed techniques for testing 'real' technical systems which behavior is decribed by (partial) nondeterministic FSMs

Multiple bio-inspired approaches have been developed aiming at achieving adaptive and autonomous behavior in the cloud. However, no such effort was invested in studying biomorphic-inspired methods. In this paper, biomorphism is presented as a central evolutionary processes where cellular differentiation take place. To the best of our knowledge, the only state-of-the-art biomorphic model applied to cloud computing was reported in our previous work [28]. The biomorphic model devised in [28] and called timing algorithm was simplistic and aimed to illustrate the concept of biomorphism for cloud scaling scenarios. In this respect, one of its major disadvantage is the timer randomness which could lead to long reaction time, possibly spawning additional unnecessary VMs. In this paper, we present an enhanced biomorphic algorithm which builds on the concept of fixed timer intervals and random seeds so that to coordinate the transformation procedure. The advantage of using a different method allowed us to achieve instant reactions and a more predictive nature of the system due to the fixed time spawning intervals.

In recent years, crowdsourcing has been actively used for many people to promote their products in online social networks (OSN), e.g., Twitter. The promotion activity in Twitter is typically performed by retweeting tweets by some accounts, called workers who earn monetary profits by deliberately promoting products for a requester. In this process, a fundamental question is whether the promotions by workers are really effective or not. To answer this question, in this paper, we analyze topological characteristics for the networks of workers, and evaluate the effect of promotions. Our empirical study demonstrates that the network of workers tends to be connected as a small-world network and their promotions show low impacts in the network of workers.

Cloud computing is rapidly reshaping the server administration landscape. The widespread use of virtualization and the increasingly high server consolidation ratios, in particular, have introduced unprecedented security challenges for users, increasing the exposure to intrusions and opening up new opportunities for attacks. Deploying security mechanisms in the hypervisor to detect and stop intrusion attempts is a promising strategy to address this problem. Existing hypervisor-based solutions, however, are typically limited to very specific classes of attacks and introduce exceedingly high performance overhead for production use.In this paper, we present Slick (Storage-Level Intrusion ChecKer), an intrusion detection system (IDS) for virtualized storage devices. Slick detects intrusion attempts by efficiently and transparently monitoring write accesses to critical regions on storage devices. The low-overhead monitoring component operates entirely inside the hypervisor, with no introspection or modifications required in the guest VMs. Using Slick, users can deploy generic IDS rules to detect a broad range of real-world intrusions in a flexible and practical way. Experimental results confirm that Slick is effective at enhancing the security of virtualized servers, while imposing less than 5% overhead in production.

Stack Overflow has been providing question and answering service for 7 years. It has become a tremendous knowledge repository for developers' thoughts and practices. Hyperlinks in discussion threads of Stack Overflow are essential knowledge entities for programming on the Web, such as a software library, an API documentation, a code example, or a tutorial. Tens of millions of hyperlinks are disseminated in Stack Overflow, while wisdom on what web resources have been highly recognized by the community is implicit in millions of discussion threads. In this paper, we develop the WisLinker framework that extracts knowledge from discussion, then turns knowledge into wisdom by learning through the knowledge dissemination history. With this wisdom, for a specific hyperlink that users are concerned with, WisLinker can recommend web resources highly recognized by the Stack Overflow community. We evaluate the validity of WisLinker in an open-ended setting using Stack Overflow data dump. We also implement a browser extension for live recommendation of web resources while users browse web pages. WisLinker could enable more efficient exploratory search and information discovery of programming-related web resources.

Microblogs have emerged as a popular platform for sharing information and expressing opinion. Twitter opinion retrieval is now recognized as a powerful tool for finding people's attitudes on different topics. However, the short length and the informal language of tweets make Twitter opinion retrieval very challenging. In this paper, we propose a stylistic-based method that leverages simple and specific stylistic variations to retrieve tweets that are relevant and opinionated on a particular topic. Results show that the proper handling of textual meta-communications can significantly improve opinion retrieval performance in Twitter.

The paper describes a method for monitoring operating systems and applications in a virtual machine. We present a framework based on plugins, receiving information from the running system. The framework can be used for debugging, dynamic analysis, intrusion detection, etc.

In cloud computing, multi-tenancy is concomitant with scalability in the sense that sharing a single deployment instance between many customer organizations (tenants) maximizes the utilization of the available resources. However, this also introduces the need to customize the application to the (slightly) different requirements of different tenants. In the context of workflow-based SaaS offerings, this is not straightforward to accomplish without compromising scalability or manageability of the offering. In this paper, we present a middleware for multi-tenant customization of workflows that enables software providers (i) to decrease coupling of multi-tenant customization concerns and workflow design for better manageability, (ii) to activate tenant preferences at runtime, but (iii) without incurring a scalability penalty. We validate a prototype implementation of our middleware in the context of a realistic SaaS application, evaluate its scalability, and extensively compare the results with related work confirming that the proposed middleware indeed supports customization of workflows in a significantly more scalable fashion. ispartof: pages:432-439 ispartof: Proceedings of the 2016 ACM Symposium on Applied Computing vol:04-08-April-2016 pages:432-439 ispartof: ACM Symposium on Applied Computing location:Pisa, Italy date:4 Apr - 8 Apr 2016 status: published

A major challenge of the existing attack detection approaches is the identification of relevant information to a particular situation, and the use of such information to perform multi-evidence intrusion detection. Addressing such a limitation requires integrating several aspects of context to better predict, avoid and respond to impending attacks. The quality and adequacy of contextual information is important to decrease uncertainty and correctly identify potential cyber-attacks. In this paper, a systematic methodology has been used to identify contextual dimensions that improve the effectiveness of detecting cyber-attacks. This methodology combines graph, probability, and information theories to create several context-based attack prediction models that analyze data at a high- and low-level. An extensive validation of our approach has been performed using a prototype system and several benchmark intrusion detection datasets yielding very promising results.

Electroencephalogram (EEG) signals play an important role in e-healthcare systems, especially to recognise the mental state of patients that could need a special care. This paper presents an EEG-based emotion recognition approach to detect the emotional state of patients for Ambient Assisted Living (AAL). The proposed approach combines wavelet energy, modified energy and wavelet entropy features to classify four different classes of emotions. Three different classifiers are used (quadratic discriminant analysis (QDA), k-nearest neighbor, and support vector machines) to recognise the emotion of patients robustly. The new approach is tested using the EEG database "DEAP" using four electrodes. It shows high performance compared to existing algorithms. An overall classification accuracy of 83.2% is obtained.

In this paper we take an important step towards better understanding the existence and extent of entity-centric language-specific bias in multilingual Wikipedia, and any deviation from its targeted neutral point of view. We propose a methodology using sentiment analysis techniques to systematically extract the variations in sentiments associated with real-world entities in different language editions of Wikipedia, illustrated with a case study of five Wikipedia language editions and a set of target entities from four categories.

Parsing Expression Grammars (PEGs) are a formalism for specifying the syntax of programming languages which can be viewed as a description of a recursive descent parser for the language it describes. An issue that users of PEG-based parsers face is poor reporting of syntax errors on the part of PEG-based parsers. Labeled failures are an extension to PEGs that aims to address this issue by annotating a PEG with labels corresponding to syntax errors, improving the quality of error messages generated by a PEG-based parser. An alternative formulation of PEGs gives the meaning of a PEG by translation into a program of a virtual parsing machine. In this paper, we present an updated semantics for PEGs with labeled failures, and we show how to extend the parsing machine to support them. We also prove the correctness of the translation from PEGs with labels to programs of this extended parsing machine, and we evaluate the use of labeled failures on PEG-based parsers which implement their own ad-hoc error reporting strategy.

Many user studies of software development use screen-capture software to record developers' behavior for post-mortem analysis. However, extracting behavioral patterns from screencaptured videos requires manual transcription and coding of videos, which is often tedious and error-prone. Automatically extracting Human-Computer Interaction (HCI) data from screen-captured videos and systematically analyzing behavioral data will help researchers analyze developers' behavior in software development more effectively and efficiently. In this paper, we present BPMiner, a novel behavior analysis approach to mine developers' behavior patterns from screencaptured videos using computer vision techniques and exploratory sequential pattern analysis. We have implemented a proof-of-concept prototype of BPMiner, and applied the BPMiner prototype to study the developers' online search behavior during software development. Our study suggests that the BPMiner approach can open up new ways to study developers' behavior in software development.

Recent computing paradigms like cloud computing and big data have become very appealing to outsource computation and storage, making it easier to realize personalized and patient centric healthcare through real-time analytics on user data. Although these technologies can significantly complement resource constrained mobile and wearable devices to store and process personal health information, privacy concerns are keeping patients from reaping the full benefits. In this paper, we present and evaluate a practical smartwatch based lifelog application for diabetics that leverages the cloud and homomorphic encryption for caregivers to analyze blood glucose, insulin values, and other parameters in a privacy friendly manner to ensure confidentiality such that even a curious cloud service provider remains oblivious of sensitive health data. ispartof: pages:119-124 ispartof: Proceedings of the 31st Annual ACM Symposium on Applied Computing vol:1 pages:119-124 ispartof: Symposium on Applied Computing location:Pisa, Italy date:4 Apr - 8 Apr 2016 status: published

While search engines sometimes return different documents containing contradictory answers, little is known about how users handle inconsistent information. This paper investigates the effect of search expertise (defined as specialized knowledge on the internal workings of search engines) on search behavior and satisfaction criteria of users. We selected four tasks comprising factoid questions with inconsistent answers, extracted answers that 30 study participants had found in these tasks, and analyzed their answer-finding behavior in terms of the presence or absence of search expertise. Our main findings are as follows: (1) finding inconsistent answers causes users with search expertise (search experts) to feel dissatisfied, while effort in searching for answers is the dominant factor in task satisfaction for those without search expertise (search non-experts); (2) search experts tend to spend longer completing tasks than search non-experts even after finding possible answers; and (3) search experts narrow down the scope of searches to promising answers as time passes as opposed to search non-experts, who search for any answers even in the closing stage of task sessions. These findings suggest that search non-experts tend to be less concerned about the consistency in their found answers, on the basis of which we discuss the design implications for making search non-experts aware of the existence of inconsistent answers and helping them to search for supporting evidence for answers.

The amount of online shops is growing daily and many Web shops focus on the same product types, like consumer electronics. Since Web shops use different product representations, it is hard to compare products among different Web shops. Duplicate detection methods aim to solve this problem by identifying the same products in differentWeb shops. In this paper, we focus on reducing the computation time of a state-of-the-art duplicate detection algorithm. First, we construct uniform vector representations for the products. We use these vectors as input for a Locality Sensitive Hashing (LSH) algorithm, which pre-selects potential duplicates. Finally, duplicate products are found by applying the Multi-component Similarity Method (MSM). Compared to original MSM, the number of needed computations can be reduced by 95% with only a minor decrease by 9% in the F1-measure.

C is the dominant programming language for developing embedded software, operating systems, and device drivers. Unlike programs written in managed languages like Java, C programs rely on explicit memory management and are prone to memory leaks. Existing (static or dynamic) detectors only report leaks, but fixing them often requires considerable manual effort by inspecting a list of reported true and false alarms. How to develop on-demand lightweight techniques for automated leak fixing without introducing new memory errors remains challenging. In this paper, we introduce AutoFix, a fully automated leak-fixing approach for C programs by combining static and dynamic program analysis. Given a leaky allocation site reported by a static memory leak detector, AutoFix performs a graph reachability analysis to identify leaky paths on the value-flow slices of the program, and then conducts a liveness analysis to locate the program points for inserting fixes (i.e., the missing free calls) on the identified leaky paths. We have implemented AutoFix in LLVM-3.5.0 and evaluated it using five SPEC2000 benchmarks and three open-source applications. Experimental results show that AutoFix can safely fix all the memory leaks reported by a state-of-the-art static memory leak detector with small instrumentation overhead.

It is commonly believed, that implicit user feedback such as clicks from search logs, can be a strong indicator of user preferences during web search. Moreover, proper click log data analysis can lead to accurate document relevance estimation. As a result, an effective approach to properly interpret users' search behaviour was the development of user click models, designed to reduce position bias and produce more attractive users' search needs. Here, in this paper we present a novel click model based on a Bayesian network, for estimating document relevance from click-through data. In contrast with previous works, our model focuses only on the estimation of actual document relevance and not on perceived relevance, which we take to be inferred directly from document popularity. We design the model to be scalable and incremental according to modern computational needs and experiments show that our proposed model achieves better scores in log-likelihood and predicts relevance more accurately than current state-of-the-art.

In this paper, we present NCaaS (Network Configuration as a Service), a centralized cloud service for creating Virtual Networks (VNets) in SDN-based cloud architectures. NCaaS relieves tenants from the burdens and complexities of VNet creation and management by supporting dynamic provisioning operations based on Quality of Service (QoS), pricing, privacy, reliability, and energy constraints set by the tenant. NCaaS provides a unified interface through which tenants' network specifications and constraints are fed into the service provisioning algorithms. These algorithms, in turn, handle the negotiation with the different SDN Network as a Service (NaaS) providers and dynamically apply the necessary VNet creation, partitioning, and migration mechanisms to ensure the satisfaction of the tenants' preferences. To the best of our knowledge, NCaaS is the first centralized cloud service for creating, managing, and provisioning VNets dynamically on behalf of tenants and based on their requirements. A proof of concept test bed implementation of NCaaS is provided on top of the OpenVirteX network virtualization platform.

Dead variable reduction is a well-known optimization used to reduce state space. In this paper we present two novel reductions for explicit-state code model checking. These reductions are designed to efficiently handle multi-threaded heap-manipulating programs. We implemented the reductions in Java PathFinder and demonstrated their efficiency by verification of several non-trivial programs. We also formally show correctness of the approach.

During software development, maintenance and evolution, we can assess the quality of code, design and architecture to capture hints of possible problems and face them as soon as possible. We can capture signals of architecture erosion in different ways by exploiting several tools with the aim to identify architectural violations, architectural smells or other relevant features. In this paper, we provide our experience report in using three known tools to capture information that can be useful to identify and evaluate the architectural debt of an application. We outline the main differences among these tools and the results they produce.

The teaching and learning of operating systems are hampered due to the complex nature of concepts involved. It is necessary not only theoretical studies, but importantly the laboratory practices, which are not always possible due to the lack of appropriate software instruments. This paper presents the NKE, a Nanokernel-based OS developed for ARM microprocessors and oriented to the teaching support for practical activities in operating systems classes. The NKE is developed in C language and has been successfully used in laboratory practices as part of OS and embedded systems classes. The results obtained so far are very exciting, given that the NKE promoted richer teaching and learning experiences, providing to the students a real environment for embedded OS development.

To cluster a data stream is a more challenging task than its regular batch version, having stricter performance constraints. In this paper an approach to this problem is presented, based on WiSARD, a memory-based artificial neural network (ANN) model. This model functioning was reviewed and improved, in order to adapt it to this task. The experimental results obtained support the use of this system for the analysis of data streams in an informative way.

Today, consumers request virtual resources like CPU, RAM, disk (etc.) supplied by the service providers (like Amazon) and they pay on a "pay-as-you-go" basis. Generally, the supervisors adopt virtualization technologies, which optimize resources usage and limit the operating cost. The virtualization technologies are classified in two categories. The first one concerns the heavy virtualization. Each virtual machines (VM) emulates hardware and embeds its own operating system (OS) that is completely isolated from the host OS. The second one concerns the light virtualization, which is based on the management of containers. The containers share the host OS kernel [5] while ensuring isolation. In this paper, we benchmark the performance and the energy consumption of an infrastructure that is based on the software Hadoop regarding the two technologies of virtualization. At first, we will identify the points to be improved concerning Hadoop performances and then we will reduce the deployment cost on the cloud. Second, the Hadoop community finds an in-depth study of the resources consumption depending on the environment of deployment. Our experiments are based on the comparison of the Docker technology (light virtualization) and VMware technology® (heavy virtualization). We come to the point that in most experiments the light technology offers better performances in completion time of workloads and it is more adapted to be used with the Hadoop software.

Large set cardinality estimators and other streaming oriented operations are the cornerstone of big data processing. Cardinality estimators combined with in-memory based storage systems provide a fast framework for keeping valuable application data easily queryable and maintanable. This has a plethora of applications. For instance, a common use case is to maintain a number of counters for monitoring application statistics for real time dashboard purposes. Another such case is large set size estimation for big data systems in internal operations like counting. In this paper is addressed the issue of scaling the computation of a cardinality estimator in the presence of node failures in a distributed setting. Moreover, for the proposed estimation technique eventual consistency is proved, which is adequate for most cases in distributed applications. To the best of the authors knowledge, this functionality is not currently provided by commonly used commercial and open source systems. Additionally, the proposed approach is generic enough to be applied to other algorithms, which can help build a basic framework for more complex operations in the big data field. We demonstrate this with graph metric calculation applications in the large scale biodata mining field.

State machine replication (SMR) is a well-established technique to fault-tolerant systems. In part, this is explained by the simplicity of the approach and its strong consistency guarantees. Recently, several proposals have suggested parallelizing the execution of state machine replicas to achieve high throughput. Concurrent execution of commands has many implications, including the recovery of replicas from failures. Conventional checkpointing techniques, for example, must be revisited in parallelized models. In this paper, we review parallel variations of state machine replication and discuss how checkpointing procedures apply to these models. Moreover, we evaluate the impact caused by checkpointing techniques on recovery through simulations.

Alzheimer's disease is the most common form of dementia. It is among the major causes of death in the world. It has been proved that Alzheimer's is associated with elevated concentrations of homocysteine. Being part of a complex biological system, we can model it to study its behaviour and predict its deregulations earlier. In this sense, mathematical modelling has limitations in behaviour study. In order to respond to these limits, the adoption of computer modelling has become increasingly widespread in systems biology. In this paper, we propose to use Petri nets to model a biological complex system related to Alzheimer's.

Automated program repair is a promising way to reduce costs on program debugging dramatically. Some repair techniques with genetic algorithm have been proposed, and they were able to fix several dozen of actual bugs in open source software. However, existing techniques occasionally take a long time to generate a repaired version of a given program. The dominant factor for that is generating so many programs that do not pass given test cases. In this research, we are trying to generate a repaired program, which passes all the test cases, in less time. Our key idea is using code similarity to select code lines to be inserted into a given program. More concretely, we propose to select code lines in code regions similar to the faulty code regions. In existing techniques, code lines for insertion are randomly selected from a given program. Currently, we are still in an early stage of this research. In this paper, we report some promising results of our pilot study, which show that using code similarity is very effective to generate repaired programs in less time.

Nowadays, the integration of wireless sensor networks with cloud computing is of great interest to healthcare research and industry. In this context, the design of sensor clouds enhances the reliability, extensibility, scalability, and availability of several WSN applications. This paper proposes a novel Sensor Cloud architecture to monitor high-risk patients. On top of the physical sensors, we build virtual sensors with respect to the medical staff's demands. These virtual sensors are dynamically created and updated based on the patient's location and health status, and also depending on the occurrence of sensitive events related to the status of the associated physical sensors. In addition to that, the configuration of virtual sensors is enriched with new functions of monitoring, maintenance, power saving and security to enhance their use. A simulation is conducted to assess the effectiveness of the proposed architecture.

Hybrid storage that uses SSD over disk-based RAID is usual in enterprise environments. One way to exploit an SSD over RAID is using it as data cache and another is using it as LeavO cache that reduces parity update overhead of RAID. In this paper, we combine data caching and LeavO caching schemes and identify optimal cache sizes. To do that, we derive cost models of combined data and LeavO caching. Also, we propose Adaptive LeavO Cache that dynamically adjusts data cache and LeavO cache sizes for evolving workloads. The experiments show that the Adaptive LeavO Cache performs far better than conventional data caching scheme and on a par with the off-line optimal scheme that determines optimal sizes after exploring all possible sizes of data cache and LeavO cache.

Spectrum sensing plays a critical role in improving spectrum utilization for cognitive radio networks. The majority of existing sensing approaches aim to detect the existence of a signal on a busy channel without differentiating whether a signal originates from a primary user or not. In this paper, we attempt to solve this challenge by an experimental approach using GNU radio and Universal Software Radio Peripheral (USRP) board. A two-stage sensing strategy is proposed to detect channel states. The first stage detects whether a channel is busy or idle by observing both the energy level and bandwidth. Once the channel is busy, we further determine the channel is occupied by a Primary User (PU) and an Secondary User (SU). From experiments, we observe that a receiver cannot successfully receive a signal if it implements different demodulation schemes with a transmitter. For example, a Binary Phase Shift Keying (BPSK) demodulator cannot detect a Quadrature Phase Shift Keying (QPSK) signal transmitted by a transmitter. Motivated by this observation, we propose that at the second stage, PUs and SUs are negotiated to conduct different modulations. Then SUs are equipped with two demodulators that will be able to demodulate both the PUs signal and the SUs signal. Given that SU and PU perform distinct modulations, once the PU comes to its licensed channel, the SU communicating on that channel will recognize it and withdraw from the channel immediately. Extensive experiments using GNU radio and USRP boards are carried out under various scenarios to evaluate the effectiveness of the proposed approach.

As smartphones become more and more popular, O2O (Offline-to-Online) applications have been spotlighted. They provide users offline data using NFC, QR, and Beacon technologies. But, there are some problems such as the high cost of server maintenance and the private data spills since they are implemented on WWW technologies. This paper suggests PWW (Person Wide Web) platform that provides server-less service on smartphones. In PWW, a smartphone has a PWW browser which is composed of an anchor scanner, a web view, an API linkage library, and internal/cloud storages. When the PWW browser scans an anchor includes a URL, it opens a web document stored in the internal storage of the smartphone or its corresponding cloud. The web document contains the API linkage library to utilize the several devices like a camera, GPS, and so on in the smartphone. In order to show the usefulness of PWW platform, we described the example of handling private data such as business cards. The PWW platform could be applied to various area in real world by developing web-based mobile applications.

Dynamic Time Warping (DTW) is a widely used distance measure for time series data mining. Its quadratic complexity requires the application of various techniques (e.g. warping constraints, lower-bounds) for deployment in real-time scenarios. In this paper we propose a randomized greedy warping algorithm for finding similarity between time series instances. We show that the proposed algorithm outperforms the simple greedy approach and also provides very good time series similarity approximation consistently, as compared to DTW. We show that the Randomized Time Warping (RTW) can be used in place of DTW as a fast similarity approximation technique by trading some classification accuracy for very fast classification.

Built-in data structures are a key contributor to the performance of dynamic languages. Record data structures, or records, are one of the common advanced, but not easily optimizable built-in data structures supported by those languages. Records may be used in an object-oriented fashion or to implement object orientation itself. In this paper, we analyze how records are used in different applications in the Scheme dialect Racket. Based on the data obtained, we suggest the application of existing optimization techniques for records and devise a new one for immutable boolean fields. Most of them can be applied to a wide range of record implementations in dynamic languages. We apply these optimizations to records in Pycket, an implementation of Racket. With one exception, microbenchmarks show a two-to ten-fold speed-up of our implementation over plain Racket.

As the era of Big-data has risen, the importance of big data technologies is also increasing day by day. Especially, Hadoop has become a critical part of the overall Big-data system because of its ability to store, process, and analyze thousands of terabytes of data. A major issue for supporting high performance on Hadoop is managing the growth of data while satisfying high storage I/O request. Hadoop's overall performance is largely influenced by the storage input/output(I/O). However, storage I/O technologies are still very limited. Therefore, now more than ever, studies on improving storage I/O on a distributed file system of Hadoop(HDFS) have been gaining popularity. To this end, latest trend in storage systems is to utilize hybrid storage devices. However, it is not easy to use the information of heterogeneous storage devices in HDFS. This is because, when reading data, HDFS is unable to exploit such heterogeneous storage type information yet. In this paper, we propose a hybrid block-selection method on the HDFS, we consider the storage type such as SSD and HDD when reading data. Using this method, the Hadoop Eco System utilizes the high SSD bandwidth by priority. As a result, we certainly improve the Hadoop Eco System overall performance. In the experiments, we demonstrated that our new method efficiently reduced the execution time of select count(*) query and TPC H benchmark up to 22% and 30% on average.