Apache Derby er en Open Source-database utviklet i Java. Den er designet som en tradisjonell diskbasert database og er optimalisert for diskaksess. Målet med denne oppgaven er å finne måter å øke ytelsen på databaser der vi antar at hele databasen får plass i minnet. Vi ønsker å optimalisere for aksessering av data i minnet istedenfor på disken og på den måten gjøre databasen mer lik MMDB(Main Memory DataBase)-databaser. Dette har vi gjort ved å identifisere og erstatte flere disk-spesifikke moduler i Derby med nye moduler optimalisert for minneaksessering. Samtidig bruker vi eksisterende moduler for å oppnå persistens av dataene. Resultatene viser at ytelsen på leseaksesser i siste stabile versjon av Derby (10.2.2.0) kan økes med 70-200% og skriveaksesser med 20-100% avhengig av hvor mange klienter som benytter databasen samtidig.

Sammenligner LAESA, List of Cluster og Piecewise Constant Aggregation i henhold til utelse målt via hvor mange objekt sammenligninger som blir utført.

We study the viability of using the map-reduce model and frameworks for NetFlow data processing. The map-reduce model is an approach to distributed processing that simplifies implementation work, and it can also help in adding fault tolerance to large processing jobs. We design and implement two prototypes of a NetFlow processing tool. One prototype is based on a design where we freely choose an approach that we consider optimal with regard to performance. This prototype functions as a reference design. The other prototype is based on and makes use of the supporting features of a map-reduce framework. The performance of both prototypes is benchmarked, and we evaluate the performance of the framework based prototype against the reference design. Based on the benchmarks we analyse and comment the differences in performance, and make a conclusion about the suitability of the map-reduce model and frameworks for the problem at hand.

This thesis focuses on a part of Internet forensics concerned with determining the geographic location of Internet hosts, also known as geolocation. Several techniques to geolocation exist. A classification of these techniques, and a comparative analysis of their properties is conducted. Based on this analysis several novel improvements to current techniques are suggested. As part of an earlier designed Multi-Agent Framework for Internet Forensics (MAFIF), an application implementing two active- measurement geolocation techniques is designed, implemented and tested. Experiments with the application are performed in the Uninett network, with the goal of identifying the impact of different network properties on geolocation. What most clearly set this thesis apart from earlier work, in addition to the use of a multi-agent system, is the analysis of the impact of IPv6 on geolocation, and the introduction of multi-party computation to geolocation. The extensive focus on delay measurements, although not bringing anything new to the field of networking in general, is also new to geolocation as far as we know.

Careful architectural decisions are required in order to create a highly available and scalable search system. This requires an in-depth analysis and understanding of the architecture and context of each deployment. Different requirements placed upon the system by different deployments mean different solutions provide the best case by case result, thus benchmarks provide an invaluable source of information.This thesis provides an overview of common components and important aspects of a distributed search system. It then gives an overview of different partitioning techniques before going into the details of repartitioning and rebalancing in a document-partitioned full-text search system.A processing framework that draws inspiration from flow-based programming literature is introduced, which is shown a valuable tool in creating custom tailored search solutions. The implementation is used to benchmark different repartitioning and rebalancing strategies.In conclusion, the techniques mentioned in the thesis show great promise in creating custom, maintainable and flexible partitions. The processing framework enables each specific deployment to easily compare different partitioning schemes and associated manageability and maintenance costs to determine the best fit for any given situation.

Given a spatial location and a set of keywords, a spatial keyword query locates spatio-textual objects based on both the location of the objects, and the textual relevance of the query keywords to the description of the objects. Spatial keyword queries can be used to answer challenging questions such as finding the nearest spatio-textual object relevant for the query keywords "restaurant sushi".Our focus in this project is on a new type of spatial keyword query that takes a road network into account during query processing. These queries are based on the fact that the distance between two objects in the real-world are constrained by the pre-defined paths that comprise the road network. Different from traditional spatial keyword queries that employ the Euclidean distance, the spatial keyword query on road networks assumes the shortest path between the query location and the objects. Unfortunately, no approach currently exists that supports processing of spatial keyword queries on road networks.In this thesis we address the challenging problem of locating spatio-textual objects in a road network given a spatial location and a set of keywords. We first propose a baseline framework that combines existing state-of-the-art approaches to support processing of keyword-based spatial queries such as range and k-nearest neighbour on road networks. Then, we present a novel framework termed Road Network Indexing (RNI) that permits efficient processing of such queries by indexing the spatio-textual objects in each road segment using inverted files.Moreover, we present algorithms to evaluate keyword k-nearest neighbour and keyword range queries on both the baseline and the RNI framework.Finally, we show through an experimental evaluation using real-world datasets, that our RNI framework performs nearest neighbour queries on road networks in around one order of magnitude faster than the baseline approach in terms of response time and I/O.

Association rules mining is one of the the most relevant techniques of data mining. It has been also applied in the domain of text mining, but the results are hard to interpret. In this matter, an Association Network is an structure to represent as a graph the relationships mined as association rules. The goal of this project was to provide a methodology to build association networks from concepts extracted from a collection of documents, as well as the study of the mathematical properties of the association networks to prove that they are not random graphs and that they exhibit small-world properties.

Today's games are social on a level that could only be imagined before. With modern games putting a stronger emphasis on social networking than ever before, the identity in the game often becomes on par with the one in real life. Yet many games lack the really strong networking tools, especially when networking regarding players of different games is concerned.Geelix is a project which tries to enhance the social gaming aspect by providing sharing of what has been aptly named: gaming experiences. The motivation for this goal is to enable stronger support for letting friends take part in your online, or even offline, experiences. It is the belief that sharing gaming experiences is a key element in building strong social networks in games. This master thesis was written in relation to the Geelix project, where the focus was on enhancing the Geelix live sharing experience with advanced methods for video compression and streaming.

Similarity search is very useful in many applications. Because of the complex-ity and expensive nature of such search operations, many existing methods re-quire special access methods and cannot be directly integrated with commercialDBMSs(Database Management Systems). NTNUStore is a framework to aid re-search in this eld and focuses on disk-based metric indexing to keep it compatiblewith commercial DBMSs.This project has implemented and experimented with a version of the LAESA(Linear Approximating and Eliminating Search Algorithm) in NTNUStore. Theresults are close to what was represented in Erik Bagge Ottesen's master thesis. The biggest contribution in this project is that the new algorithm solves KNNsearch without the need of providing a range parameter in the query.

Tradisjonelle løsninger for publisering av produktkataloger på CD eller papir innebæ-rer ofte store kostnader knyttet til distribusjon og printing av dokumenter. Dette kanmedføre at aktørene fokuserer mye på dokumenthåndteringen, på bekostning av egnemålsetninger om skape nærmere tilknytning mellom kunde og produkt.Dagens vekst innenfor teknologi skaper muligheter for løsninger man tidligere anså somdrømmetenking. Teknologiveksten i seg selv er også med på å skape et behov eller etønske om å dra nytte av de mulighetene som _nnes innenfor områder som marketing ogeksponeringstjenester. Ved å implementere en distribuert løsning ved bruk av mellomvareteknologi,vil man kunne få en løsning som administreres sentralt. Dette vil bidramed fordeler som alltid oppdatert produktportefølje, mindre krav til klientplattform, liteeller ingen administrasjon på klientsiden, kunde-/forhandlertilpasset innhold og bedrekostnadse_ektivitet. Både ved hyppige oppdateringer av porteføljen og høy grad av markedstilpasning,vil behovet for en distribuert løsning øke.I dagens teknologijungel _nnes det mange former for mellomvare til ulike formål. Noe avproblemet er å vite hvilken mellomvareteknologi som egner seg til hvilket formål. Ved åvelge rett vil man kunne høste frukter som økt e_ektivitet og forbedret brukeropplevelse.Vårt hovedfokus har derfor vært å se på aktuelle mellomvareteknologier som kan benyttesi en distribuert interaktiv produktkatalog, i den hensikt å skulle kunne avgjøre gjennome_ektivitetstester hvilken teknologi som vil være mest e_ektiv. Vi har valgt ut .NET Remoting,Java RMI og Web Service for å benytte i implementasjoner med funksjonalitetsom vi anser som sentral i en slik løsning. Vi har deretter testet funksjonaliteten i deulike implementasjonene ved hjelp av verktøy for enhetstesting.Gjennom en helhetlig vurdering av e_ektivitet basert på våre testresultater og gra_skgrensesnitt konkluderer vi med at .NET Remoting over TCP/BIN er det beste valget.Den har en overlegen tidsbruk på oppretting av et objekt med XML håndtering, oghenter ett bilde raskest av samtlige mellomvarer på en ADSL linje som er reelt for endistribuert produktkatalog. Samtidig viser vår evaluering av de gra_ske grensesnittene at.NET brukergrensesnittene er raskere enn Java grensesnittene.