We chose the dual-node category of SPECjAppServer2002 for our case study, as that is the simplest setup to show case performance of the application on an application server. We chose a system based on 4P Intel® Xeon® processors for the driver/emulator and a system based on 4P Intel® Itanium® processors for the back end. We focus on performance on the application-server middle tier (also a system based on 4P Itanium processors).
We chose Windows* as the platform for analysis, although most of the methods of analysis presented here can be applied to other operating systems as well. For the application server and the emulator, we chose BEA WebLogic* 8.1, although this study is relevant to other application servers as well. For the JVM* running on the application server, we chose BEA JRockit*, as it is highly optimized for Intel® architecture. We chose a high-speed switched network connection (1Gbps) as a typical production environment for such applications to be deployed in data centers. Figure 1 illustrates the configuration we have described.
Figure 1. Workload configuration.
In our case study, we stayed within the scope of design choices as illustrated in Figure 1, but we have opportunities to use different platforms (e.g. different processors, different processor frequencies, different caches and memories) to study their performance impacts. For the database back-end, we have choices of Oracle 9i* and Microsoft SQL Server*, as well as using or not using RAID disk arrays. The choice of the database back-end systems affects the choice of the JDBC* layer for the application server also. For the application server, we have different versions of BEA WebLogic and JRockit at our disposal.
The initial baseline configuration is composed of a driver/emulator based on 4P Intel Xeon processors at 2GHz with 1MB L3 Cache and 4GB RAM, and application and database servers based on 4P Itanium processors at 1GHz with 3MB L3 Cache and 8GB RAM. We used WebLogic 7 and JRockit 1.3 for our initial configuration. Over the course of the performance study, our final configuration consisted of a platform based on 4P Intel Xeon processors at 2.8GHz with 2MB L3 Cache and 4GB RAM, and application and database servers based on 4P Itanium processors at 1.7GHz with 9MB L3 Cache and 16GB RAM. We used WebLogic 8.1 SP2 and JRockit 1.4.2 SP3 for our final software configurations. We chose to stay with Microsoft Windows 2003 Server, Enterprise Edition, RTM 3790 through the study on all platforms.
As you can see between the initial baseline and the final configurations, we have gone through significant changes in the hardware and software components. As one might expect, simply upgrading the software and hardware components may not directly increase performance. Any such upgrade may expose a new performance bottleneck and thus require a systematic top-down data-driven approach to overcome the bottleneck. Over the course of our study, we obtained a net performance gain greater than four-fold, due to the combinations of software/hardware upgrades and the application of the performance methodology.
We summarize the key findings of performance optimization of our workload in the following sections.
