CASE EXAMPLE

Nanyang Technological University (NTU) has adopted E-learning as a complementary tool for teaching and learning, combining it with the conventional classroom lecture and tutorial education. The system used at NTU is " Edventure " (NTU, 2002) supported by "Blackboard". In this environment, lecturers can post the course materials, exercises and project assignments to the forum. Students can download course materials from there, submit assignments and post questions regarding the course they are taking.

Although the " Edventure " system runs smoothly and plays an important role in conducting teaching and learning at the school, there are still some limitations. It is based on centralized client-server paradigms . So, the course materials are hosted at a single server (or dual servers), and the server's CPU has to service all the incoming requests from users. Some course materials in the school are only updated weekly by the lecturers. Within this short period of time, all students registered for the subjects need to download course materials from the Edventure server to prepare themselves for the lecture. The request hits to the server increase tremendously during this period, and not all the requests can be processed by the CPU in the Edventure server immediately. Some requests are queued until the CPU completes the current processing of client requests, which results in long response time for the user to bear with. The computation load at the server is also very heavy since the whole system depends on the centralized server. Under situations where the system administrator needs to perform system maintenance, or when the server is down accidentally , the whole system stops functioning until the server is recovered.

Course Information Distribution

Using the mobile agent-based E-learning system described, we reconstructed the Edventure system based on a highly decentralized network configuration ” distributed mobile agent paradigm. Several proxy servers are needed in addition to the central server. Both the central server and the proxy servers are used to construct faded information field (FIF), in which both the CPU time and network bandwidth are distributed.

For the weekly updating of course materials ” after the materials are posted to the central server by the lecturers, and when the user requests increase ” the system calculates the access effort (Equation 1) of each information based on the number of hits, information lifetime, and size . The information with more hits will be assigned higher priority, and a push agent will push this higher priority information to the neighboring nodes. In this case, a user's request is directed to corresponding proxy nodes, which have a lower network traffic cost than other nodes. In a later period, after most students have downloaded the course materials, user requests will decrease, and the push agents will recalculate the access effort of the course information. This process will likely yield a smaller access effort, showing that users' demand for such course materials has decreased. Therefore, the push agent will erase the course materials from some proxy nodes. When a threshold access effort is reached, all the course materials will be removed from the neighboring hosts .

Customized Course Construction

The course structure in the previous Edventure system is fixed, and students have to take course modules in a fixed sequence. Thus, the reusability of course materials is low. For example, in the department where the authors of this article work, "Network Design and Simulation" and "Network Performance Analysis" are two typical courses. For both courses, the mathematics behind probability and statistics are the foundation for other theories and algorithms. Hence, if a student has taken "Network Design and Simulation", in which he has learned statistics and probability, he must still go through those lessons again when he takes "Network Performance Analysis". The fixed course structure decreases the reusability of course materials; students have to take redundant sub-courses and follow the fixed course structure at a pace not set by themselves.

Using a mobile agent system, the problem is solved by dividing an entire course into sub-courses, and categorizing them according to specific domain knowledge. Each sub-course material is related to a set of non-serialized reference materials, and these materials are available for students' ad-hoc retrieval. Customized course materials may be conducted by retrieving and linking sub-course materials together. In this case, the common parts, such as the probability and statistics in the above example, are separated; these common parts can be reused to construct other courses related to network technology. For each sub-course module, a student can optionally retrieve some reference material on the respective topics. With the structured course materials and unstructured reference materials, the course can be customized according to the student's pace and interest, and the sub-course reusability is improved. The construction of a course from sub-course materials and reference materials by students in order to study at their own pace also relieves the course authoring burden of lecturers.

Interactive User Tracking

It has been shown that the more interaction and cooperation between instructors and learners, the more effective learning will be. User tracking functions provide an interactive environment for lecturers and students to conduct teaching and learning. This environment can be realized through the cooperation of static agents at student machines and static agents that sit on an e-learning central server. These two groups of agents monitor the course status for a particular student, and communicate this information by message passing.

There are two types of tracking. At the lecturers' side, they receive feedback from students regarding the course materials effectiveness. For example, if the user tracking report shows a high access rate for a certain course material, it may mean the difficulty level of this course material is appropriate or that the course is taken by many students. On the contrary, if the access rates of certain courses are low, that may mean the courses are not very popular; either the difficulty levels are too high or the course formats are not userfriendly. With the feedback from students, the course materials can be customized to match students' learning pace and interest.

To test the effectiveness of student learning, after the course is taken, the system also provides some randomized questions for students in the form of quizzes. The quizzes will be graded by the e-learning server, and the server will feedback the grade information to each student. This is realized through the communication of two groups of static agents.

System Security

The Edventure system provides strict user authentication. Each student who has registered for some subjects has an independent account. Each user can only log in to his own account to search for course materials, and the system enforces regular password changes by users. At the server side, information on all transactions performed by an individual user is recorded into a log file. In the event that the system faces any security attack, it is easy to trace the source of attack from the log file.