Digging into SAP R/3 for Capacity Planning

SAP R/3 is a complex application containing almost a duplicate distributed operating system inside. While working on traditional capacity planning for SAP R/3 relevant features of its internal architecture sometimes need to be considered. This paper presents an example of a validated analytic model of a complex SAP R/3 environment with a twist related to its internal architecture, namely, dispatcher queuing within an SAP instance. 1. Capacity Planning for SAP R/3 – an Overview Before discussing a case study, we present a brief overview of what capacity planning for SAP R/3 would entail. 1.1 Workload characterization This task consists of determining resource utilization by different types of work. The data obtained is used for regular ongoing performance management and capacity planning. In the case of SAP R/3 the categories into which performance metrics can be broken down include: • Applications (FI, CO, SD) • Sub-applications • Users • Tasks (Dialog, Update, Background, etc.) • Tcodes (SAP R/3 Transaction Codes) etc. To obtain such categorization, data from the following sources must be combined: • The operating system (“work processes” doing the application server work are well known) • SAP R/3 proper (CCMS, SAP R/3 Computer Center Management System, is instrumented to measure resource use and other parameters broken down by multiple categories) Information on units of work (dialog steps in SAP) is also available from CCMS. The quality of this data, as well as approaches to its extraction and post-processing are discussed in more detail, e.g., in [THOM97]. 1.2 Sizing using resource utilization profiles If resource utilization for the unit of work (e.g., dialog step in the case of SAP) is available, what-if modeling of system performance under hypothetical conditions could be done. Of course relevant resource profiles must have been previously obtained, usually by means of benchmarking, and an analytic modeling tool needs to be used. 1.3 Application architecture-specific modeling The internal architecture of SAP R/3 is Complex and contains many subsystems, which include services similar to those provided by general-purpose operating systems. Several such areas of architecture can be listed: • memory management including multiple buffers and Extended Memory • task scheduling performed by the dispatcher assigning requests to work processes • locking and access to objects A straightforward capacity planning approach can be taken on the assumption that these subsystems are configured adequately for the load, yet inadequate configuration could skew the results of such an approach. A significant quantity of advice in configuring these subsystems is available from SAP AG and other sources. The challenge, however, is to come up with analytical methods for determining the effects of configuration parameters and consequently for choosing appropriate values for them.