Building Distributed Application with Aglet*

Major existing paradigms for building distributed applications can be classified into two groups. Examples of the first class include traditional RPC and most recently its object cousins RMI and COBRA. For this class of paradigms, functionality of applications are partitioned among participating nodes. Different participants use message-passing to coordinate distributed computation. Computation itself is partitioned, participants exchange intermediate results and other synchronization information. For the second class of paradigms, computation is migrated toward resources. This type of paradigm is especially useful for applications requesting immediate reaction to incoming streams of real-time data and distributed applications that are very tightly coupled. In this project, we try to experiment with one such paradigm--Aglet. Aglet is the shorthand for agent plus applet. It provides us an infrastructure for building second-class distributed applications.

The following part of the paper is divided into four sections. For the first section, we will describe aglet's concepts and architecture. For the next section, we will define the application. And in the following section we will give details of our design and implementation of a distributed stock information system. And finally We will summarize the common working mechanism of Aglet-based applications.

The application we choose is a simple but complete one. It activates most mechanism under the hood of aglet. It also illustrates a model which is suitable for Aglet-based distributed computing. e.g. We will rely on aglet to do intensive remote I/O; and for complicated user requests, we will request aglet to process/filter intermediate results in our stead; user's requests may only be satisfied by collecting data from several databases and some of these databases may be temporally unavailable because of broken links. The scenario poses great challenges to application developers who work in traditional RPC-like framework.

On the server side of our applications are aglet hosts and databases which store all the stock information. For the client side, stock is the major aglet. Stock is a static aglet, at user's requests, it will dispatch slave aglets to server side and fetch stock information.

We download a popular freeware database engine mysql-3.20.16-beta and configure a SQL server on magpie. We build different databases for simulated exchange sites. Conceptually, these databases are geologically distributed all over the world. We generate scripts to automatically load initial data into our databases.

We use gweMysqlJDBC V0.9.2 as our JDBC driver to connect to our mysql databases.

The major part of the application is done on the client side. We will first of all describe the visual components of the client-side user interface.

When the user presses go button, Stock aglet creates a slave called StockSlave, which is subclassed from ibm.aglets.patterns.Slave, and passes the destination and arguments of the query to the StockSlave. StockSlave is the labor aglet that really go to the remote side. Upon arrival at the remote site, doJob member function of StockSlave is called. doJob does what the real work we assign to the slave. Here, it extracts the arguments and connect to the database. It then calls JDBC SQL statement to do the query for the stock information the user is interested in. If there is more than one destination in its itinary, it might dispatch itself to another destination after it finishes work on one site. After finishing the itinary, the slave brings back the query results through TransferInfo class which implements serializable interface.

When the StockSlave returns to the original site, member function callback of the static master aglet Stock is activated. The result is passed as an argument of the callback member function. Stock aglet will extract the result and display on the canvass.

When implementing our applications, we encountered a few problems. The first problem is the caching of byte-code. We modify our stock database by inserting new records into the tables. All the insertion works fine for a stand-alone java application program. But when we tried to use aglet to do the insertion for us, we will get a duplicate key error from the SQL server.

Example:
	String query = "INSERT INTO jdbc VALUES (17, 20000,8000000,
               'This is a variable string')";
	ResultSet rs = stmt.executeQuery(query);

	The first field 17 is a primary key, if we modify it to another key 
	and perform the same insertion, we get a "duplicated primary key error".