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.