Please use this identifier to cite or link to this item:
|Title:||Distributed Similarity Queries Search in Metric Space in IoTSystems|
|Keywords:||Big IoT data; Metric space; Similarity queries search; B3CF-tree; Clustering;DBSCAN; Parallelism; Coefficient of Variation method; ThresholdDistance method; QCCF-tree|
|Abstract:||In recent years, the large amount of continuous and heterogeneous datagenerated by the Internet of Things (IoT) sensors and devices madetheir record and the query search tasks much more difficult. Most of thestate-of-the-art methods have failed to deal with the new IoT requirements. In this thesis, the kNN search method combined parallelismwas used for similarity queries search in proposed methods developedin metric space in the Fog-Cloud architecture. The first proposition isthe Binary tree based on Containers at the Cloud-Clusters Fog computinglevel (B3CF-tree) which is an index constructed by combiningDBSCAN clustering and parallelism. The simulation results of the indexconstruction and the parallel kNN query search showed that theB3CF-tree surpassed those in literature. The second proposition isthe Coefficient of Variation (CV) method which was developed for indexingcontinuous IoT data stream. In this method, the first datastream is grouped into clusters using the DBSCAN algorithm. Datain these clusters are directly indexed in parallel. After the clusteringof the arrival data stream, the data in clusters are inserted in existing indexes or new indexes are constructed basing on the coefficientof variation value. This method has proven its efficiency in term ofthe indexes construction and the parallel kNN query search comparedwith two other methods representing the two utmost cases namely theCreation of a New Index (CNI) and the Insertion in an Existing Index(IEI) methods. The third proposition is the Threshold Distance(TD) method which looked like the CV method. However, in the TDmethod, the arrival clusters are indexed or inserted in existing indexesbasing on the comparison of the distance between their centers andthe first clusters centers with a threshold distance TD. This methodoutperforms the Creation of a New Tree (CNT) method in terms oftrees construction and parallel kNN search however, it is quite insufficientcompared with the results of the CV method. The experimentalresults showed that Both methods surpassed some indexing methodsin literature and could be considered as an alternative method for indexingcontinuous IoT big data. The last proposition is the Quad treebased on Containers at the Cloud- Fog computing level (QCCF-tree) in which data are directly indexed without clustering. The comparisonof the experimental results of the index construction and the parallelkNN search in the index nodes with some indexes in literature showedthat the QCCF-tree is more efficient than these indexes. This made ita candidate as an alternative method for big IoT data indexing eventhough it presented a weakness in front of the B3CF-tree.|
|Appears in Collections:||Thèses de Doctorat|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.