Capacity Planning

Capacity planning is another key long-term operation. Many of the real-time technologies I have discussed implement policies that describe how the resources are divided among a set of competing services. Capacity planning ensures that there are enough resources in the future to make the resource allocation strategies workno real-time strategy works effectively when its resources are over-subscribed.

The goals of capacity planning are similar to those for proactive managementto give the management team sufficient time to take the necessary actions to prevent a service disruption. In the case of real-time operations, the lead time is measured in minutes, whereas capacity planning works on the scale of weeks or months.

Capacity planning is considerably more complex than in the early client-server days. In early client-server designs, the basic environment was a server attached to a router. Performance problems were usually addressed through boosting server performance or increasing the Internet connection speed. Most of the time the problem was solved, or at least postponed for some time.

Today's environments are not amenable to the blanket upgrade approach; adding resources to every element is simply too expensive. Even when the funding for large-scale over-provisioning is available, there is no guarantee that it will actually solve the problem. Having excess resources helps with loading fluctuations, but often the resource enhancements that actually contribute to any improvement are hard to pinpoint.

Planners need to understand the sensitivitiesthe factors that have the most influence on behaviorin their environments. The dynamics of complex systems depend on the relationships between resources and the changing distance between operating and inflection points. Understanding sensitivities to user volume, transaction volume, service mix, and other factors helps focus on the areas where the highest return will be realized.

Identifying the resources that are the first to be over-subscribed and congested enables specific interventions that produce the largest improvement for the least investment. This represents another chance to take a short breather before the process begins again. Some other resource becomes a new problem area, and the planning and evaluation of alternatives is repeated.

It's important to note that the different constituent domains of the services delivery environmentsoftware, hardware, servers, networking elements, and peopleall scale somewhat differently. As a result, there's no single capacity planning methodology that extends across multiple domains. There are also important dependencies among themserver scaling is obviously a product of the demands of the application or applications that the server will be running.

Note that more mature applications, such as databases and enterprise resource planning (ERP), are fairly well characterized, and they have the instrumentation to support analysis of capacity horizons. Newer applications, such as directories and application servers, are less straightforward. In both cases, there's a fair amount of literature that addresses planning for different subsystems. However, given the expertise required for each domain, effective capacity planning will remain an art of collaboration among such experts for the foreseeable future.