Avoid Bottlenecks

If we want to make progress, we are probably suspicious of every bottleneck that might slow down the team. Sometimes, we might install a bottleneck intentionally—for example, having a single integration point is an intended bottleneck for avoiding conflicts. However, most of the time, we want to avoid bottlenecks no matter what the cause.

Bottlenecks in a large team

Here are the most frequent bottlenecks found in large teams.

Coupling

The architecture has to ensure that the subsystems are as loosely coupled as possible. Otherwise, a single change can require the compilation of the whole application because it is very likely that our application will be huge after a while; this can be very time consuming.

This problem can be solved by eliminating strong typing (either by an interpretative or just-in-time programming language like Smalltalk or by generic interfaces supported in Java). The disadvantage is that the type won’t be checked during compile time.

Enterprise-wide object model

Do not insist on an enterprise-wide object model. The value of an enterprise-wide object model is rarely worth the effort of creating it. The higher the effort of creating the enterprise-wide object model, the more we should question its value. It’s probably so difficult because everybody (or every team) has a different view of the objects needed. Instead of forcing everybody to have the same view, it might be more helpful to support different views. The problem with an enterprise-wide or system-wide object model is that it creates a lot of general dependencies not only among the objects but also among all teams. Often, we will find that the model is highly coupled, probably with several cyclic dependencies in place. This usually leads to a general recompilation as soon as something in the model changes.

An enterprise-wide or system-wide data model is most often sufficient and often already available, as it is rare to work on the very first system for a company. A much more typical situation is one where a client company already has a system (or systems) in place and just needs another system (or a new version), which will operate on the already existing data model.