Handling batch processing overloads in ERP

Users of Microsoft Dynamics AX ERP leverage batch jobs for processing large amounts of data. This kind of batch job processing is expected to cause high resource load on the system. Parallel execution of more than one such resource-intensive batch job can lead to overload of system resources. This results in poor interactive user experience due to high response times. This master's thesis introduces resource overload control solution that reduces number of parallel resource-intensive jobs by performing admission control. Because admission control distributes execution times of resource-intensive jobs over time, resource consumption also gets distributed. The introduction part of the thesis begins with a brief overview of Microsoft Dynamics AX system and batch job use. That is followed by analysis of system resources and batch job workload on the test environment. In analysis resources are represented as operating system resources and data resources that are defined as groups of database tables. In order to evaluate the proposed solution an artificial workload is needed, which is defined based on the results of workload analysis. The central part of the thesis presents how the standard algorithm for scheduling of batch jobs in AX works. The scheduling algorithm is then extended with admission control that uses information about tasks in combination with resource consumption in order to decide whether to select task for execution or not. Admission control needs information about state of resources, thus a monitoring mechanism is introduced. This monitoring implements a dynamic threshold algorithm to eliminate insignificant resource state changes. Final part of the thesis begins with the presentation of results obtained by using the solution on the test environment. First, the results are presented, which were obtained by applying the workload of generic batch jobs that consume a specific resource. This is used to demonstrate that the solution is working according to the design. That is followed by the presentation of the results that were obtained by applying workload of real batch jobs that consume various system resources. With this it is presented how well the solution performs when handling overloads in general. At the end the solution is evaluated by summarizing the results. In the process, the time is used as a criteria of evaluation, in which the resource consumption is above the given reference point