While performance testing a system, we are curious about how would the system perform with a bigger load. To get as real results as possible, we use a synthetic load instead of a fully artificial load. Synthetic load reflects some of the properties of the real load. This master thesis develops a methodology for the analysis of the load of a sample e-health system. The methodology aims to get as much detailed data of real load so the data can be used as a starting point for forming synthetic load, which we usually use in case of stress and load testing of sample systems.
In this master thesis, we first explain the concept of e-health and introduce the sample system, which was used to obtain the sample load. We continue with queueing theory, the definition of basic load metrics, and with a presentation of calculating interarrival times for requests with Poisson and Pareto probability distribution.
The analysis is suggested in two main steps. The first step consists of obtaining and processing of the data in the sample load. The second step consists of the analysis of the processed data which represents the groundwork for defining synthetic load.
In the first step, we focus on capturing and processing of HTTP requests. User actions play a key role in this master thesis. They consist of consecutive HTTP requests. For the purpose of processing action data, we develop our own tool, written in the Java programming language. The result of capturing and processing the data is a frequency distribution of interarrival times for actions that the user performed in the sample system. The distribution is graphically shown on a histogram.
In the second step, we focus on metrics of interarrival times of requests in the serving system. In detail, we analyze the dynamics of interarrival times of requests and actions. We compare the distribution of interarrival times of actions with Poisson and Pareto probability distributions. The decision for the distribution which largely coincides with our distribution of interarrival times of actions is based on the results of the Jensen-Shannon divergence. The final result of this step is a definition of the equation to calculate interarrival times, which can be used as the groundwork for the definition of a synthetic load.
|