Introduction

---

Many factors are important to achieving continuous availability in the Web infrastructure. General availability concepts and practices are covered in the HVWS paper Maximize Web Site Availability. This chapter discusses design scenarios for clustering Web and application servers to achieve continuous availability. The scenarios include many of the new and enhanced features of IBM WebSphere Application Server Network Deployment, Version 5. The clustering scenarios cover the features in the WebSphere Web server plug-in and the WebSphere Edge components (formerly Edge Server).

Much is written about clustering Web and application servers. This chapter presents design scenarios presented from the perspective of demonstrating continuous availability, not just during normal operations, but also during system maintenance. This means maintaining continuous, or near continuous, availability, while hardware is replaced and applications are upgraded. Typical clustering scenarios work well during steady state conditions. The real challenge comes in achieving near continuous availability during system maintenance, especially for high-volume Web sites that do not have a true off-peak time, such as global sites.

A consideration in architecting the Web infrastructure is the physical placement of the Web server, the J2EE Web container, and J2EE EJB container processes. It is possible to have each process run on its own physical machine or node, or grouped on one or two nodes. The scenarios described in this chapter follow the practice of placing the Web server process on its own node and the Web container and the EJB container on a second node. When the J2EE containers are placed on the same node, it is also possible to deploy them in the same Java Virtual Machine (JVM) process. A grouping of nodes performing a similar function is referred to as a tier and hence there are two tiers in the scenario depicted in Figure 3-1, the Web tier and the application server tier. For the purposes of this document the Web tier generally serves static content while the application server tier serves dynamic content.

Figure 3-1: Physical separation of Web and Application server tiers

This architecture is optimal and popular for Web infrastructures for several reasons. Two of the main reasons are:

• Placing the Web server process on a physically separate machine provides the option of deploying the Web server in a demilitarized zone (DMZ), while protecting the application server processes behind a firewall.

• Running the Web container and the EJB container in the same process provides optimal performance by removing the overhead of serialization and deserialization of parameters and results that are exchanged between the Web container and the EJB container.

  Complex and large object types can result in significant performance costs when these containers run in separate processes.

There are exceptions that require a different configuration than Figure 3-1. For example, EJB clients other than Web components are not typically deployed in the same process as the EJB modules. Discussion of these exceptions is beyond the scope of this chapter.

---