As we have previously seen in this chapter, our clients were defined by using interfaces. The main motivation for this added level of abstraction is so that we can substitute a remote implementation for a local one. Let's say that component A depends on component B. If these components were deployed on separate servers, we would want to use an implementation of B's client that makes remote calls to the component. If, however, both components co-existed within the same JVM, using a remote client would incur unnecessary network latency. Substituting the client for one that directly invokes component B's Java implementation ensures that no networking takes place, thus reducing latency.