Parallel Density-Based Stream Clustering Using a Multi-User GPU Scheduler

With the emergence of advanced stream computing architectures, their deployment to accelerate long-running data mining applications is becoming a matter of course. This work presents a novel design concept of the stream clustering algorithm DenStream, based on a previously presented scheduling framework for GPUs. By means of our scheduler OCLSched, DenStream runs together with general computation tasks in a multi-user computing environment, sharing the GPU resources. A major point of concern throughout this paper has been to disclose the functionality and purposes of the applied scheduling methods, and to demonstrate the OCLSched’s ability of managing highly complex applications in a multi-task GPU environment. Also in terms of performance, our tests show reasonable improvements when comparing the proposed parallel concept of DenStream with a single-threaded CPU version.

Authors: Ayman Tarakji, Hassani M., Lyubomir Georgiev, Seidl T.
Published in: Proceedings of the 11th Internation Conference on Beyond Databases, Architectures and Structures (BDAS 2015), 26-29 May, 2015 in Ustron, Poland.
Publisher: Springer International Publishing
Sprache: EN
Jahr: 2015
Seiten: 343-360
ISBN: 978-3-319-18421-0
ISSN: 1865-0929
Konferenz: BDAS
DOI:10.1007/978-3-319-18422-7_31
URL:BDAS
Typ: Tagungsbeiträge
Forschungsgebiet: Data Analysis and Knowledge Extraction