Messung und Modellierung von SAP-R-3- und Storage-Systemen für die Kapazitätsplanung

Many companies make use of the standard business software SAP R/3 to carry out their business processes. High response times while processing transactions in an R/3 system and, as a result, a lack of service quality or a poor system availability may have serious consequences for the companies, so that e. g. products cannot be finished in time. Such performance bottlenecks are partly due to a growing number of users, an R/3 release upgrade or an increased workload caused by the installation of additional business components. Therefore the aim must be to make out such scenarios as early as possible in order to avoid performance bottlenecks e. g. by carrying out necessary system upgrades in time. This PhD thesis is meant to present a feasible approach to the capacity planning of SAP R/3 systems, which deals with monitoring, measuring, and analysing R/3 performance data on the one hand, and predicting possible future scenarios on the basis of the data acquired earlier on the other. In the context of data measurement a model of characterizing workload including a concept to define service quality is presented. Quickly solvable mathematical and analytical models (queueing networks) have been employed to predict performance. Concepts, methods, and tools for analysing performance data and modelling SAP R/3 systems have been described in detail and explained on the basis of several case studies. Almost every transaction in an SAP system accesses the central database and, depending on the cache of the database, the corresponding IO-system. Accordingly, the response time per transaction is determined by the CPU performance of the servers as well as by their IO performance. Thus, this PhD thesis also focuses on the analysis and prediction of the performance of IO systems. Concepts to describe the workload of IO activity in interdependence with the load in an SAP R/3 system have been worked out along with an analytical disk model. The latter supports the disk feature Tagged Command Queueing and caching strategies for readand write-requests. Moreover, it takes into account the parameters of the operating system having an influence on IO processing. In contrast to earlier research this model has been evaluated on the basis of real data, i. e. various load scenarios have been examined, which were produced with the help of an IO benchmark and analysed by using an IO monitoring tool.