With the development of the Social Network and the Internet of Things (IOT), the big data era has come, in which large amounts of data that could not otherwise be imagined are generated. Data analysis is essential to discover and use value from large ...
With the development of the Social Network and the Internet of Things (IOT), the big data era has come, in which large amounts of data that could not otherwise be imagined are generated. Data analysis is essential to discover and use value from large amounts of data. In order to do this, an ETL (Extract Transformation Loading) task to load the data to be analyzed into DW or Big Data System should be preceded. ETL schedules hundreds or thousands of jobs on the limited system resource into a sequential relationship and keeps the pre and post-relationships of the tasks at a right time. In addition, it is difficult to optimize the overall performance of the ETL, because the scheduled ETL is performed by a complex number of operations. Therefore, it is necessary to increase or decrease the data capacity due to the business change. The performance may be delayed due to the complex load. For these reasons, professional engineer is needed to optimize and operate it, but it is a realistic difficulty. Because it requires not only a lack of experienced engineers but also a high cost.
In order to optimize the performance of the ETL task and maintain the optimized performance on the various changing situation, this study has developed the optimization method by the genetic algorithm which is a kind of meta - heuristic technique and artificial intelligence.
Optimizing ETL performance means you can deliver data at the right time in your business, without delay. To do this, optimizing the performance of the ETL unit job is also important, but it is important to optimize the performance of batch jobs that contain the whole unit job. It is also important to minimize the number of failures due to abnormal performance loads. To do this, we used minimization of ETL batch execution time and load minimization of server CPU resources as a fitness function of genetic algorithm. ETL batch end time and number of concurrent execution tasks were used as constraints.
The data of this study were used to simplify the 3 - month average CPU usage and execution time of the 260 ETL units that are being performed in the business. Also, a mathematical model is defined for genetic algorithm implementation and implemented using Java program and Maria DB. Experiments were repeated to change the parameters of the implemented program in order to obtain optimal results.
This study is expected to be an important foundation for constructing and maintaining a stable big data system by loading the explosive data of the Big Data era. It is a meaningful attempt to automatically optimize the loading of ETL data of the information system.