Performance and cost estimations for web applications on Amazon elastic compute cloud

Abstract

Web base application plays an important role in supporting online activities, such as E-business, contents distribution and a communication portal. There is a big increasing in number of online activity usage that leads to an important problem to a webserver which is a “traffic overload problem.” At present, the conventional solutions are whether to increase the number of webserver or to increase the capability of the server (more RAM, CPU, Cache, HDD). Unfortunately, these solutions bring about waste resulting from overloading problem that accounts for less than 25 percents of the operation time. In addition, hardware capacity has not been used to it maximum over than 70 percent of the time, but left idle instead. This study proposes a new method to solve this problem on Amazon Elastic Cloud Compute (EC2) service. The proposed solution relies on Auto Scaling where the number of system instances (web server) can be increased and decreased based on some pre-defined condition. In this case, the number of current system users is the main focus. An experiment showed that this method provided great performance. It was also found that focusing on tuning of system crontab variable was the most effective method and easiest for real practice. However, the proposed solution is only suitable for the system which predicted user distribution is in exponential form.