This report compares the characteristics of push/pull load balancing techniques usedin the context of a logging system. The logging system is expected to handle a largevolume of events. The load balancing techniques are evaluated with focus onthroughput during high load. The testing scenarios includes the use of a traditionalload balancer (push-based) and the use of messaging queues (pull-based and indirectlycontext aware) in its place. The ultimate goal of the report is to determine the feasibilityof using a messaging queue rather than a traditional load balancer in a distributedlogging system. Tests were conducted measuring the throughput of multiple setupswith different load balancers. The conclusion of this report is that both messagingqueues and load balancing are equally feasible in a logging context. / Rapporten jämför egenskaper hos lastbalanseringstekniker för användning i ettdistribuerat logghanteringssystem. Systemet förväntas hantera stora volymermeddelanden vid hög belastning. Testscenarion som utförs sker med traditionelllastbalansering där event trycks ut, samt med meddelandeköer som är hämtbaserade.Målet med rapporten är att avgöra om kontextbaserad lastbalansering kan ökastabiliteten i ett system avsett för hantering av loggdata. Testerna som utfördesuppmätte mängden data som gick igenom systemet vid en given tidpunkt, testernakördes med flera typer av lastbalanserare. Slutsatsen som dras är att bådemeddelandeköer och lastbalansering är passande för användning i ett loggsystem.

Matching and integrating ontologies has been a desirable technique in areas such as data fusion, knowledge integration, the Semantic Web and the development of advanced services in distributed system. Unfortunately, the heterogeneities of ontologies cause big obstacles in the development of this technique. This licentiate thesis describes an approach to tackle the problem of ontology integration using description logics and production rules, both on a syntactic level and on a semantic level. Concepts in ontologies are matched and integrated to generate ontology intersections. Context is extracted and rules for handling heterogeneous ontology reasoning with contexts are developed. Ontologies are integrated by two processes. The first integration is to generate an ontology intersection from two OWL ontologies. The result is an ontology intersection, which is an independent ontology containing non-contradictory assertions based on the original ontologies. The second integration is carried out by rules that extract context, such as ontology content and ontology description data, e.g. time and ontology creator. The integration is designed for conceptual ontology integration. The information of instances isn't considered, neither in the integrating process nor in the integrating results. An ontology reasoner is used in the integration process for non-violation check of two OWL ontologies and a rule engine for handling conflicts according to production rules. The ontology reasoner checks the satisfiability of concepts with the help of anchors, i.e. synonyms and string-identical entities; production rules are applied to integrate ontologies, with the constraint that the original ontologies should not be violated. The second integration process is carried out with production rules with context data of the ontologies. Ontology reasoning, in a repository, is conducted within the boundary of each ontology. Nonetheless, with context rules, reasoning is carried out across ontologies. The contents of an ontology provide context for its defined entities and are extracted to provide context with the help of an ontology reasoner. Metadata of ontologies are criteria that are useful for describing ontologies. Rules using context, also called context rules, are developed and in-built in the repository. New rules can also be added. The scientific contribution of the thesis is the suggested approach applying semantic based techniques to provide a complementary method for ontology matching and integrating semantically. With the illustration of the ontology integration process and the context rules and a few manually integrated ontology results, the approach shows the potential to help to develop advanced knowledge-based services. / <p>QC 20130201</p>

A considerably fraction of science discovery is nowadays relying on computer simulations. High Performance Computing  (HPC) provides scientists with the means to simulate processes ranging from climate modeling to protein folding. However, achieving good application performance and making an optimal use of HPC resources is a heroic task due to the complexity of parallel software. Therefore, performance tools  and runtime systems that help users to execute  applications in the most optimal way are of utmost importance in the landscape of HPC.  In this thesis, we explore different techniques to tackle the challenges of collecting, storing, and using  fine-grained performance data. First, we investigate the automatic use of real-time performance data in order to run applications in an optimal way. To that end, we present a prototype of an adaptive task-based runtime system that uses real-time performance data for task scheduling. This runtime system has a performance monitoring component that provides real-time access to the performance behavior of anapplication while it runs. The implementation of this monitoring component is presented and evaluated within this thesis. Secondly, we explore lossless compression approaches  for MPI monitoring. One of the main problems that  performance tools face is the huge amount of fine-grained data that can be generated from an instrumented application. Collecting fine-grained data from a program is the best method to uncover the root causes of performance bottlenecks, however, it is unfeasible with extremely parallel applications  or applications with long execution times. On the other hand, collecting coarse-grained data is scalable but  sometimes not enough to discern the root cause of a performance problem. Thus, we propose a new method for performance monitoring of MPI programs using event flow graphs. Event flow graphs  provide very low overhead in terms of execution time and  storage size, and can be used to reconstruct fine-grained trace files of application events ordered in time. / <p>QC 20150508</p>

The lubrication and tension levels are two important parameters when running a chainsaw, both impacting the cutting performance and lifespan of the tool. An appropriate level of lubrication and tension leads to maximum performance of the chainsaw and direct benefits for the end-user. This thesis addresses the problem of detecting the lubrication status and the tension level using the information contained in the acoustic emissions captured in the guidebar of the chainsaw. Data was collected by running controlled experiments using an acoustic emissions sensor.Information was extracted from acoustic emissions by using a number of features computed on different frequency ranges.Three machine learning models were trained and evaluated on data corresponding to different combinations of lubrication status and tension levels. The models' performances were evaluated using the well-known metrics accuracy, precision, and recall. A pattern was found for each lubrication and tension setup, and the model that registered the highest performance was the Random Forest. The impact of temperature, guidebar, and chain on acoustic emissions is also analyzed. The detection of different lubrication levels using the information contained by acoustic signals is also addressed, the patterns in data being determined by computing features in the time and frequency domains. The analysis shows that the temperature does not have an impact when the running time is less than 10 minutes, and the chain has a bigger impact than the guidebar for the specific setup of the experiments. Moreover, a pattern dependent on the guidebar and chain combination correlated with the lubrication level was identified. The main contribution of this thesis consists of detecting a pattern representative of lubrication and tension setup in acoustic emission using a number of features computed in different frequency ranges.

The number of IoT devices and their respective data is increasing for each day impacting the traditional architecture model of solely using the cloud for processing and storage in a negative way. This model may therefore need a supporting model to alleviate the different challenges for future IoT applications. Several researchers have described and presented algorithms and models with focus on distributed architecture models. The main issues with these however is that they fall short when it comes to the implementation and distribution of tasks. The former issue is that they are not implemented on actual hardware but simulated in a constrained environment. The latter issue is that they are not considering sharing a single task but to distribute a whole task. The objective of this thesis is therefore to present the different challenges regarding the traditional architecture model, investigate the research gap for the IoT and the different computing paradigms. Together with this implementing and evaluating a future architecture model capable of collaboration for the completion of a generated task on multiple off-the-shelf hardware. This model is evaluated based on task completion time, data size, and scalability. The results show that the different testbeds are capable communicating and splitting a single task to be completed on multiple devices. They further show that the testbeds containing multiple devices are performing better regarding completion time and do not suffer from noticeable scalability issues. Lastly, they show that the completion time drops remarkably for tasks that are split and distributed.

In this thesis, we explore the idea of using binary visualization and image analysis to detect anomalous activity on an Industrial Internet of Things (IIoT) based network. The data is gathered into a pcap file and then fed into our encoder, which uses a space-filling curve to convert the 1-dimensional stream of data into pixels with a specific red, blue, and green gradient value.  The pixels create an image which is then given to an image analysis system based on a Convolutional Neural Network, which classifies if the traffic supplied is malicious or not. The results show that using a Binary and Multiclass classifier approach to the image analysis both work well reaching an accuracy of 100% and 94% respectively. While the binary classifier is more accurate both succeed at separating Malicious from Benign traffic. The choice of space-filling curves in our binary visualization ended up having little to no impact on overall classification accuracy.

Detta arbete utvärderar nyttan av att konstruera ett kluster med ett företags eller en organisations befintliga datorer. Klustret baserat på vanliga persondatorer kan blisom ett alternativ att driftsätta applikationer och tjänster. Alternativet är i stället föratt införskaffa serverar och högprestandadatorer. Detta kan bidra till att minska kostnader och återanvända resurser som redan finns. Arbetet mäter klustrets prestanda och effektivitet, därefter jämför resultatet med andra datorer och system. Arbetet utförs hos Syntronic och Högskolan i Gävle, där arbetet använder sig av de resurser som både parterna har och arbetet konstruerar varsitt kluster att mäta med.Resultatet visar att ett kluster byggd av datorer från Syntronic och högskolan presterar jämförbart med konsumentklassade processorer som kan hittas i nya datorer. Resultatet visar dock att bägge klustren har varken någon ekonomisk eller ekologisk fördel jämfört med nya datorer. Det är mer gynnsamt att sälja av Syntronics och Högskolans datorer i andrahandsmarknaden, för att sedan införskaffa nya datorer. / This thesis evaluates a cluster according to energy efficiency and performance. A cluster made of computers that a company, or an organization already owns, without purchasing any computers. The cluster based on older commodity computersacts as an alternative for operating and running software instead of buying server or high-end hardware. Reason for creating a cluster out of existing computers is mainly for reducing expenses and reuse hardware that otherwise considers as waste. This thesis measures the clusters performance and efficiency, then compares the results with other hardware and systems. This thesis uses computers from Syntronic and University of Gävle. This thesis creates a cluster each, one at Syntronic, one at University of Gävle and then measures them. The result shows that a cluster made of Syntronics computer performs equally amongst systems with a consumer grade processor. This thesis also shows that both clusters are not a viable option from an economic and environmental perspective. It is more beneficial to sell those computersfrom Syntronic and the University, for then to purchase new hardware.

The peripheral nervous system can be seen as a huge network of neurons that prop-agates signals across the human body. In fact, as seen in [1] , "All the information streaming in the peripheral nervous system (PNS) is transmitted along axons byelectro-chemical signals called action potentials". The ordinary conduction of such nervous signals, or action potentials, can be prevented however due to nerve damage. In this context, the accurate passage of information to an intended destination or partwithin the organism is obstructed. Admittedly, it is understood that physiotherapy can be quite helpful in regaining correct functionality over a damaged part of the peripheral nervous system. Yet, it is still hard to visualize the nervous activity as it is achieved inside the human body. Simulating the nervous system would providea platform to visualize how the system works and how a damaged nerve can affect the PNS. Indeed, the purpose of this study is to simulate a virtual network that im-itates a general topology of the human peripheral nervous system e.g(simulate the nervous structure and behaviour of the human arm) that shows how the signals canbe routed to their correct destination and showcase how can the simulator created simulate biological nerve damage in its system.

Fraud is a threat that most online service providers must address in the development of their systems to ensure an efficient security policy and the integrity of their revenue. Amadeus, a Global Distribution System providing a transaction platform for flight booking by travel agents, is targeted by fraud attempts that could lead to revenue losses and indemnifications. The objective of this thesis is to detect fraud attempts by applying machine learning algorithms to bookings represented by Passenger Name Record history. Due to the lack of labelled data, the current study presents a benchmark of unsupervised algorithms and aggregation methods. It also describes anomaly detection techniques which can be applied to self-organizing maps and hierarchical clustering. Considering the important amount of transactions per second processed by Amadeus back-ends, we eventually highlight potential bottlenecks and alternatives.

As commissioned by Bontouch AB, this project contains the attempt to create an Augmented Reality application for smartphones, where a parcel is visualised in order for a user to get a comprehensive image of the parcel size. The main focus of the project was to create an eectively running en- gine that can process images for computer vision, in order to determine the pose of the smartphone camera. The experi- ment explored the possibilities of using gyroscope measure- ments between input images in order to formulate valid as- sumptions of upcoming images using homographies. These assumptions were meant to unburden the computer vision engine, creating very high performance. The ultimate goal of the engine was to track the motion of a known reference object in the image, with high precision. The proposed method performed adequately, improving reliability of the motion tracking algorithms. The resulting mobile application, run on an iPhone 5S, could perform camera pose estimation at up to 60 times per second, at a video camera feed resolution of 1280x720 pixels. This high performance resulted in a very stable rendition of the parcel. / Som ett uppdrag från Bontouch AB innehåller det här pro- jektet ett försök att skapa en Augmented Reality-applikation för smartphones, där postpaket ska visualiseras för att ge användare en förstärkt bild av paketets storlek. Huvudfo- kus i projektet var att bygga en eektiv motor som kan behandla bilder för datorseende, för att beräkna positionen och riktningen på mobilkameran. Experimentet utforskade möjligheterna för att använda gyroskopdata mellan bilder för att formulera giltiga antaganden hos kommande bilder med hjälp av homografier. Dessa antaganden vad ämna- de att avlasta motorn för datorseende, för att åstadkomma hög prestanda. Målet var att motorn skulle följa rörelsen hos ett givet referensobjekt i bilden, och att göra det med hög precision. Den föreslagna metoden presterade tillräckligt, och för- bättrade pålitligheten hos algoritmerna för att följa objek- tets rörelser. Mobilapplikationen som byggdes, och kördes på iPhone 5S, kunde beräkna kamerans position of riktning upp till 60 gånger per sekund, när videokameran försåg mo- torn med bilder med 1280x720 pixlars upplösning. Den höga prestandan resulterade i en väldigt stabil bild av paketet.