Under the EJB 3.0 API, the business interface of an enterprise bean is a plain Java interface, not an EJBObject or EJBLocalObject interface. However, usage of the earlier EJBObject and EJBLocalObjectinterface types continues to be supported under EJB 3.0.
With EJB 2.0, session beans and entity beans can expose their methods to clients through two types of interfaces: a remote interface and a local interface. The 2.0 remote interface is identical to the remote interface used in the 1.1 architecture, whereby, the bean inherits from RMI, exposes its methods across the network tier, and has the same capability to interact with distributed clients.
However, the local interfaces for session and entity beans provide support for lightweight access from EJBs that are local clients; that is, clients co-located in the same EJB container. The EJB 2.0 specification further requires that EJBs that use local interfaces be within the same application. That is, the deployment descriptors for an application’s EJBs using local interfaces must be contained within one ejb-jar file.
In the EJB 1.1 architecture, session and entity beans have one type of interface, a remote interface, through which they can be accessed by clients and other application components. The remote interface is designed such that a bean instance has remote capabilities; the bean inherits from RMI and can interact with distributed clients across the network.
The local interface is a standard Java interface. It does not inherit from RMI. An enterprise bean uses the local interface to expose its methods to other beans that reside within the same container. By using a local interface, a bean may be more tightly coupled with its clients and may be directly accessed without the overhead of a remote method call.
In addition, local interfaces permit values to be passed between beans with pass by reference semantics. Because you are now passing a reference to an object, rather than the object itself, this reduces the overhead incurred when passing objects with large amounts of data, resulting in a performance gain.