Introduction

This study is about building and running Oracle RAC 10g Release 2 on Novell SUSE Linux™ Enterprise Server (SLES 10) SP2, with an IBM® System z10®. The system setup is presented in detail. Several tests are conducted, varying various parameters, to determine optimal values and performance trade-offs.

Oracle Real Application Clusters (Oracle RAC) is a clustered variation of the Oracle relational database, called RDBMS. An Oracle RAC installation is made up of a set of clustered servers or nodes that share a single set of data stored on disk or other media and have an infrastructure to orchestrate the way that the nodes work together on the shared database to ensure data consistency.

Each instance or node of the cluster is a unique and fully operational database server. The shared data can be accessed by all the instances simultaneously, even at high levels of utilization. The technique that enables shared access to the stored data is called cache fusion. A service named Global Cache Service (GCS) maintains the cache on all the servers, and coordinates them to act as a single cache. The data is updated between the nodes with a fast interconnect device dedicated to that purpose.

This study is about installing and running Oracle RAC on Linux on IBM System z®, with exploration of performance-related topics. The results of the study are based on experience obtained when building and using Oracle RAC as a test system in the IBM Linux Integration Test facility at IBM, specifically using Oracle RAC 10g Release 2 on Novell SUSE Linux Enterprise Server 10 (SLES 10) SP2 on an IBM System z10.

The goal of the study was to analyze the performance of an Oracle RAC environment running under Linux on IBM System z, with a transaction processing workload. The workload balanced mode was used with shared tables, in order to identify what factors influence cluster contention and how network tuning, especially on the interconnect, influences the performance. In this mode both nodes are actively used, the alternative is the failover mode where one node is used only in a failure situation.

This document is intended for people who are considering Oracle RAC as a clustered data base on Linux on IBM System z, and are interested in information derived from first hand experience. It is not presumed that the reader has previously installed or used Oracle RAC, or is familiar with the installation process or tuning choices.

This study focuses on features that are unique to IBM System z, including HiperSockets™, LPAR technology, and large disk storage units, so that the content can be useful to someone who is familiar with Oracle on other platforms and wants to know about the product on Linux on IBM System z.

This study describes the system setup, by stepping through the installation process that was followed, emphasizing features that are specific to IBM System z.