Understanding the CO4 behaviour

• CO4 and KAPI
• Addressing
• Storage
• Web communication

CO4 and KAPI

The CO₄ protocol takes advantage of the KAPI messaging services, and like KAPI uses an Agent Name Server (ANS) supported by an HTTP server to manage clients as agents. It thus requires KAPI to be installed. The CO₄ package is made of two libraries, libCo4, the main one, and libKapi2, built on top of KAPI which adds some functions like mailboxes, so KAPI (libkapi.a) is still required. The package also contains 6 cgi-scripts, to deal with ANS requests:

• Co4_register and Co4_unregister to manage agent.
• Co4_deleteFile and Co4_writeFile to manage saved files.
• Co4_addMail to save mail
• Co4_directory to retrieve existing repositories
• Co4_routingScript to route HTTP requests from a browser.

Addressing

A CO₄ repository is a process running as a demon, exchanging KQML messages with other CO₄ repositories, for example a demand of registration. A TCP port ("KQML port" in the remainder) is dedicated to this repository to communicate with its peers and the corresponding URL (host and port) is recorded by the ANS at the creation of the repository. This allows KQML messages to only deal with repository nicknames, without knowing their physical address.

Before each emission of a message, the ANS is requested for the URL of the target, as shown in . Several repositories can run on the same host, by choosing different KQML ports.

Figure . Communication scheme of the Co4 protocol. Two repositories B1 (located on host1) and B2 (located on host2) communicate through an ANS (located on ANS-host). B1 requests the URL of B2, so the ANS retrieves in the files of B2 its URL, tcp://host2:1000, and returns it to B1. On its side, B2 asks to save data, so the ANS returns to it the result of the corresponding script execution (here done).

Storage

The ANS also deals with repository management requests, because it stores repository data, by the way of the cgi-scripts presented above. The ANS used by CO₄ has been enriched to store all data relating to the repositories. So, let a HTTP server be on a host, with a scripts directory located in $HTTP_BIN_ROOT, and a documents directory located in $HTTP_DOC_ROOT. CO₄ installation will copy the 6 Co4_ scripts in $HTTP_BIN_ROOT, and make a directory named CO4 in $HTTP_DOC_ROOT. Each registered repository will have an account in this directory CO4, to store:

• a file named url containing the URL used by KAPI (see KAPI documentation for more information);
• a file named infos containing all the information needed by CO₄ for a successful wake up;
• a file named mbox containing pending mails;
• a file named active, whose presence indicates it the repository is active or not;
• a directory named knowledge where all the content is stored.

Figure . The various directories and files that can be found in the ANS HTTP server.

• -- a CREATE will create a new directory named after the new repository, and create url and active.
• -- a DELETE will erase all the repository directories.
• -- a SHUTDOWN will delete the active file, save a new infos file, and save all the knowledge files.
• -- a WAKEUP will read infos, read and delete mbox if it is not empty, load knowledge files, and create active.

Web communication

CO₄ provides a HTML interface to visualise the repository and to manipulate the CO₄ primitives. This repository behaves as a server for some HTTP clients (web browsers). It uses another TCP port (server port in the remainder), different from the KQML port, to communicate with its clients, via a routing script, named Co4_routingScript and located in $HTTP_BIN_ROOT. This script receives a URL, extracts the server port from it, sends and receives data from the repository through that port, and finally, returns HTML code to the browser to be displayed, as shown in .

Several servers can run on a same host, by choosing different server and KQML ports.

Figure 4. A click on the HTML interface of B1 displayed by a browser causes a call to the routing script, that retrieves the server port of B1 (tct://lotse:6000) from the URL, then sends data to B1, gets some HTML code from it, and finally returns this code to the browser to be displayed. The KQML port (tcp:5551) is always active in order to let B1 receive KQML messages send by other repositories.