IBM BS029ML Self Help Guide - Page 30
The building blocks of an architecture, Planning for the Peak
View all IBM BS029ML manuals
Add to My Manuals
Save this manual to your list of manuals |
Page 30 highlights
X, the actual anticipated estimated rises to 7,500 concurrent clients after two years time, which then increases the percentage to 18.75%. Normally, it is common for business requirements to state that a Portal should be able to handle X number of clients concurrently. It is important to distinguish between concurrent clients and concurrent active clients; as such, terminology is often misinterpreted between different parties. Concurrent active clients have both an active connection to the HTTP server as well as at least one thread of execution running in the application server. At any point in time, many of the clients connected to the Portal are not active; they may be thinking, reading, or even drinking coffee. These are considered as inactive concurrent clients, or more generically as concurrent clients. Based on our experience, a good starting point is to assume that for every single concurrent active client there are approximately 10 concurrent inactive clients (1:10). Theoretically, therefore, an application server capable of supporting 100 active clients will support approximately 1000 concurrent clients (active + inactive). This assumption breaks down somewhat when the characteristics of WebSphere Portal Server shift away from that of being a traditional Web-based solution. For example, a Portal performing more back-end work will effectively shift the assumed work pattern from that of being a traditional Web-based solution to that of an On-Line Transaction Processing (OLTP) solution. Such an OLTP solution will place greater demands on system resources, with a reduced supporting ratio of approximately 1:5 or less. A further point of debate between different parties is the understanding of Peak Load or Arrival Rate. It is important to recognize that it may be necessary to plan for such situations when many users simultaneously access the Portal solution at the same time. This generally breaks any rule of thumb for concurrency and is indicative of such situations as logins, each morning between 8am and 9am, or campaign launches. Under such circumstances, is it only possible to honor each request by Planning for the Peak. Attention: A sizing estimate should only ever be used as an approximation of the hardware resources required to support the proposed implementation. Actual experiences may vary from the sizing estimate for many reasons. The degree of variability can range from small to very significant. As such, there is no substitute for not undertaking a full capacity planning and performance tuning exercise. Failing to implement this critical part of any project plan is planning to fail. 2.2 The building blocks of an architecture When faced with the challenge of architecting a WebSphere Portal Server implementation, it is often useful to take a high-level approach to first define the logical components that comprise the very architecture that is about to be designed. For the experienced IT Architect and Portal Practitioner, this commonly embraces two aspects of design; the component model and the operational model. Component models are typically focused on identifying the components, their responsibilities, and characteristics required to deliver the solution requirements. At a conceptual level, the component model documents the technical architecture at a very high level and does so in a technology agnostic manner. At a specification level, the component model documents the required specifications and corresponding realization of all components, which ultimately will be placed on the operational model, together with a description of their interfaces, dependencies and collaborations. In common terminology, the component model addresses the logical 16 IBM WebSphere Portal V6 Self Help Guide