A Test Based Multidimensional Performance Model for a Mission Critical System Server

This paper presents an approach to performance modeling of servers managing radio networks, applied in Motorola Government & Public Safety Department. The primary goal of creating the performance model for mission-critical systems is to verify if strict requirements are met even under the highest possible load. By determining capacity of various hardware configurations, the model serves also as a cost optimization tool for a customer. In large radio systems the cost of devices is tremendous, therefore the method of determining the most economical hardware capable of handling the intended load, is vital. The last but not least purpose of performance tests, being the key part of modeling, is creating a product enhancement plan by finding bottlenecks to widen and indicating parameters that need tuning. The presented method includes the following steps:(1) creation of a predictive model, (2) planning and execution of the set of tests to get real performance measurements, (3) model validation by applying mathematical regression to compare the test results with the theoretical model. In the industry practice the product is under continuous development, so the predictive model for every release is based on experience from previous releases. Many detected weaknesses and performance problems are postponed to the next release. The performance model covers them all and evolves together with the system. The paper describes the full cycle of modeling of a group of Unix-based servers controlling a radio network system. The paper is founded on interesting cases from real-life industrial testing.