A scalable k-means clustering algorithm on Multi-Core architecture

In the recent years, major CPU designers have shifted from ramping up clock speeds to add on-chip multi-core processors. Algorithms and applications must be tuned to allow multi-core processors to exploit their inherent parallelism. An experiment is carried out with data mining (DM) algorithms, to explore the potential of quad-core hardware architecture with OpenMP API (application programming interface). The concept of memory mapped files is widely supported by most of the modern operating systems. Performance of memory mapped files on multi-core processor is also explored. In this experiment, a popular clustering algorithm k-means with serial versions and parallel versions are used. The later is experimented with static and dynamic threads of OpenMP. Memory utilization of k-means algorithm is also analyzed. Experimental results with both simulated and real data demonstrate the scalability of our implementation and effective utilization of parallel hardware, which is beneficial to data mining (DM) problems, which involves large data sets. The capacity of RAM (random access memory) affects the mmap() performance.