Implementation of parallel FP-growth algorithm on cluster computers

With the development of technology, there is an increase in size and complexity of data. The increase bring about new needs in the process of data processing. Data mining, aimed to satisfy these needs, is the process of acquiring useful information from large data warehouses. In data mining, the purpose is usually to find the frequent items which is called frequent itemset mining (FIM). Different algorithms are developed and studied for this purpose. Because of the limited physical capacity, serial algorithms aren?t sufficient when solving large data problems. So, paralel computing becomes important. With parallel algorithms, it is aimed to save on time and sources (like storage and memory). FP-Growth is a data mining algorithm that is developed to find frequent itemsets. FP-Growth is known with performing only two scans while reading database and using frequent pattern tree to find frequent items. Because of this data structure, it becomes more efficient than other frequent itemset mining algorithms. In the study, there are three different parallel approaches implemented in MPI, a library that is used to write parallel programs. The first basis thing for the approaches is whether there is a database on all of the nodes or not. The second thing is how task parallelism is done. For task, there are static and dynamical approaches. These are tested and analysed in two different clusters. According to the results, two of them are better than the other one which is sent database (tree) to the other processors. For both of them, the running time of the parallel algorithms is decreased by one forth by using two nodes. In addition to that, dynamical task parallelism is better than the static one in better algorithms.