Using Load Dependent Servers to Reduce the Complexity of Large Client-Server Simulation Models

摘要

The process of developing software systems that meet performance objectives has been called Software Performance Engineering (SPE). The goal of SPE techniques is to construct performance models in each phase of the software development cycle. This allows us to detect designs that are likely to exhibit unacceptable performance. Simple models are preferable in early stages of the development cycle because their solutions require less computational effort and time. Hence, our research is oriented to reduce the execution time of large client-server simulation models by replacing some of their components, by less time-consuming models: Load Dependent Servers (LDS). In this chapter we describe the construction of Load Dependent Servers that model server stations with transactional workloads. These LDS also include a representation of the main overhead sources in transactional environments: operating system and DBMS overhead. As case study we model a server station running TPC-C transactions.