Workload pattern analysis

Abstract

In a service oriented world, performance plays a vital role in the success of any IT system. For an application running in a production environment, whenever there is a change in the workload or workload pattern, utilization of major server resources like cpus, disks, memory, network etc. will also change. In this paper, we are extending our methodology to estimate the server resource utilization for any given workload pattern by extracting the optimal information from the historic production logs (application logs and resource utilization or system monitoring logs) using a specifically designed genetic algorithm. Across all experimental validations, we find the average absolute error in estimating utilization of server resources was less than 15%. Unlike traditional approaches to estimate overall resource utilization, method presented here, neither requires to estimate service demands for each individual application functions nor does it require to benchmark individual business transactions. Only necessary input to the model is the application logs containing the information about the throughput (for example an access log in case of web application) and system monitoring logs containing aggregate resource utilization information.