Enterprise JavaBeans (EJB s) represent one of the most widely-used server-side component architectures for developing the business logic of enterprise-scale applications. Because of their runtime properties, such as scalability, security, and transactional integrity, Enterprise JavaBeans are also an excellent tool for building software agents. This paper outlines an architecture for developing and deploying EJB -based agents. The presented approach is designed in a way that employs all the benefits of EJB s, while, at the same time, it hides the underlying complexity from agent developers.
I.2.11 [Computing Methodologies]: Distributed Artificial Intelligence—Multiagent systems
Extensible Java EE-based Agent Framework (XJAF ) [
A recent improvement of XJAF , named SOA-based MAS
(SOM ) [
SOM can be easily used by a wide variety of external clients. Because it is a specification of web services, their functionalities and interactions, SOM can be implemented using many modern programming languages and platforms. The default implementation is still provided in Java EE , including the same set of benefits as the original XJAF .
SOM is accompanied by an agent-oriented programming
language named Agent Language for SOM (ALAS ) [
This paper presents recent improvement of the ALAS compiler that enables the full utilization of Enterprise JavaBeans (EJB ). SOM agents are now defined as stateless session EJB s, and then passed on to the enterprise application server. In this way, many advanced EJB functionalities, such as scalability and object pooling, can be exploited with a minimum programming effort. At the same time, the ALAS programming language hides this complexity from the agent developer.
The rest of the paper is organized as follows. Section 2 provides an overview of the EJB technology. Recent performance improvements of SOM and its agents are described in Section 3. Finally, general conclusions and future research directions are given in Section 4. 2.
Enterprise JavaBeans (EJB ) represent one of the most
important Java EE technologies for developing server-side
components. EJB s incorporate the business logic of
“distributed, transactional, secure and portable” [
In general, there are two categories of EJB s (or, simply,
beans): message-driven, and session beans. Message-driven
beans are used in the context of Java Message Service JMS
[
Session beans can further be categorized into singleton, stateless, and stateful. As its name suggests, there is a single instance of a singleton bean per Java EE application. Concurrent access is managed by the EJB container and can be fine-tuned by the developer. Stateless beans maintain no conversational state between distinct invocations. They are well-suited for operations that can be executed in a single method call. A stateful bean, on the other hand, is used when the conversational state between different method calls needs to be preserved.
When it comes to the runtime efficiency of Java EE applications, stateless session beans offer the best performance. This is because modern enterprise application servers offer the EJB pooling technique. The server maintains a pool of pre-initialized stateless beans. Once a client request is made, the server selects an instance from the pool, executes the method, and then returns the bean instance back to the pool. Therefore, an instance is recycled from the pool for each new request, rather than having to be (re)allocated, used, and then deallocated. The pool size is also automatically adjusted to fit the number of concurrent requests. In this way, only a small number of stateless EJB s can be used to serve a large number of external clients.
To achieve the best performance, and utilize the object pooling technique, SOM agents are based on stateless session EJB s.
Originally, SOM (and its predecessor XJAF ) featured a
single stateless session EJB class named AgentHolder [
This approach was convenient from the point of view of agent developers. They were able to develop agents as POJO s, without any knowledge or understanding of EJB s and the whole enterprise application architecture. An obvious disadvantage, however, is the extra allocation/deallocation of the agent object – the exact step EJB pooling avoids.
With the recent development of ALAS , this disadvantage can be alleviated. The language compiler has now been modified to output an EJB directly, and thus avoid the process of “packing” a POJO inside of the AgentHolder bean.
As an example, Listing 1 shows the source code of a simple PingAgent written in ALAS . The agent exposes a single service named ping which outputs the received parameter.
Listing 1: ALAS source code of PingAgent package ex amp les ; agent PingAgent { s e rv i ce void p i n g ( S t r i n g msg ) {
l o g ( msg ) ; } } // o u t p u t s t h e m essage From the given source code, the ALAS compiler produces a stateless session EJB shown in Listing 2. The bean, also marked as local-only features a single handler for all incoming messages.
Listing 2: Auto-generated EJB for PingAgent @ S t a t e l e s s @LocalBean public c l a s s PingAgent implements Agent { public void onMessage ( ACLMessage msg ) { i f ( msg . getCommand ( ) . e q u a l s ( ”Ping ” ) ) { // unm arshal t h e a c t u a l c o n t e n t Ping t a s k = XMLMapper . u n m a r s h a l l P i n g ( message . g e t C o n t e n t ( ) ) ; S t r i n g param = t a s k . getMsg ( ) ; a l a s . s t d l i b . j a v a . common . Log . w r i t e (
param ) ; return ; } } }
This simple example demonstrates how ALAS represents a solution that achieves the best runtime performance, while hiding the complexity of enterprise applications from agent developers. 4.
Enterprise JavaBeans represent one of the most widelyused technologies for developing the server-side business logic of large-scale applications. A category of EJB s, known as stateless session EJB s, provide the best runtime performance due to advanced object pooling and load-balancing techniques offered by modern enterprise application servers. This paper presents a solution for developing and deploying software agents that are based exactly on stateless session EJB s. At the same time, the ALAS agent-oriented programming language hides the complexity of enterprise applications from agent developers, allowing them to focus on problem-solving tasks.
Further improvements of the system will be focused on efficient approaches and algorithms for preserving the agent state between distinct running sessions. Additionally, the existing JMS -based system will be replaced by a new and recommended technique of asynchronous bean invocation. The goal is to continuously improve the system’s performance, while following the modern standards and technologies. 5.
This work is partially supported by Ministry of Education and Science of the Republic of Serbia, through project no. OI174023: “Intelligent techniques and their integration into wide-spectrum decision support.” 6.