It’s often a challenge to come up with an efficient architecture for running WebSphere applications on Linux for zSeries. Such systems consist of zSeries mainframes running VM and multiple Linux guests, each running WebSphere and customer applications. With all of these components involved, it gets a little complicated ensuring that everything works together optimally. This article provides the basics you need to get started as well as additional references.
The methodology provided enables you to calculate memory requirements for the VM and Linux systems. IBM developed this methodology using customer experiences and real-world testing. The methodology and accompanying spreadsheet should work for most customers, but of course some applications may not fit the model.
This article contains a mix of high-level explanations, worksheets, and detailed instructions. We recommend that you:
- Read the article, filling in the sample worksheets provided.
- Download the worksheets from the z/Journal Website at www.zjournal.com (click on the current issue icon and go to the Steve Wehr article).
- Use the data from the spreadsheet to set up your systems.
OVERVIEW OF THE ARCHITECTURE
The architecture discussed is shown in Figure 1. A network dispatcher, part of the IBM Edge Server suite, receives requests from the Internet and “sprays” them between two or more production Linux guests running the WebSphere applications. Edge Server setup is beyond the scope of this article and is not discussed.
The mainframe is configured with one or more z/VM LPARs, which run the Linux guests for production and test WebSphere applications. It shows a typical WebSphere application configuration using JDBC through DB2 Connect to talk to DB2 databases running on a z/OS LPAR.
WHY USE z/VM TO RUN LINUX?
Linux can run natively on the hardware or under z/VM. Running natively eliminates the small amount of CPU overhead that it takes to run VM. This overhead is around 15 percent for a VM system running Linux and WebSphere.
Running Linux natively can reclaim this 15 percent; however, there are disadvantages, too. For example, a z900 allows only 16 LPARs, many of which may already be in use; with several or even dozens of Linux guests, you will quickly run out of LPARs. So, the first argument for running Linux under VM is that it conserves LPARs. More important is VM’s ability to virtualize CPUs and memory and share those resources among many Linux guests. The result is that you can run the same number of Linux guests under VM using fewer CPU and memory resources than it would take to run them natively.