Using Semantic Nets to Model Troubleshooting's Knowledge, part 1

By Marc-Henri Poget. Published with permission of the author.

---

EDITORS NOTE: Era 4 tools are expert systems whose foundation and work flow are based primarily on a valid troubleshooting process.

---

Semantic Nets

For example, imagine typical concepts related to birds, these concepts can be: Birds, Feathers, Wings, Worms, Trees. These concepts by themselves don't give so much information. Additional information is contained in the following sentences:

• Birds fly using their Wings.
• Their Wings are made of Feathers.
• Birds eat Worms.
• Birds build their nests in the Trees.

figure 1: Birds related concepts

Applying Semantic Nets to Troubleshooting

figure 2: Web application with database.

Although the figure 2 presents the architecture of a typical CGI application, it lacks some information that may be useful when troubleshooting it. This figure doesn't show that the Web browser requires some dynamic link libraries (DLL) or that the CGI program is written in Perl. Semantic nets aren't restricted to an architectural view of the system, but they can show all the Web application related concepts and the relationships which express either dependancies between these concepts or the action of a component on another. Therefore, semantic nets allow to build better models than system diagrams do.

Semantic nets creation is an iterative process. First create a very simple semantic net that can be easily extended when new information about the system is gained. To create the semantic net for the Web application example depicted in figure 3, I followed these steps:

• Identify the concepts. They may be either physical component (as the Web browser) or some more abstract entities (as the TCP / IP protocol).
• Identify the relationships between these concepts. In the Web application example, one can identify the following relationships between the concepts:
• The Web application is accessed through the Web browser.
• The Web application runs on the Web server.
• The Web browser communicates using HTTP and TCP / IP.
• The Web browser needs Dynamic link libraries.
• The Web server communicates using HTTP and TCP / IP.
• When using HTTP, TCP / IP, one gets host names using the Domain Name Server.
• The Web server is configured by Configuration files.
• The Web server launches the CGI program.
• The CGI program is written in Perl.
• The CGI program gets results from the database.
• The CGI program formats the results in HTML.
• The Database is configured by Configuration files.
• Draw the semantic net. Each concept is mapped to a box and each of the sentences above is mapped to an arrow linking the related boxes.
• When new information is gained, new concepts and relationships can be added to refine the semantic net.

As a troubleshooting's guide, this semantic net helps to see how things that may fail are related. Thus, it is a good tool to support the 6th step of the Universal Troubleshooting Process (Narrow it down to the root cause).

To navigate from one concept to another, the Troubleshooter must know which tests to do. For example, starting from Web application, one needs to know how to test whether the problem is on the Web browser or on the Web server side. This testing information may be associated to concepts and typical Web browser related tests may be: try accessing other pages, try using another browser. Assuming that all these tests run perfectly, one will conclude that the problem is located in the Web server.

---

EDITORS NOTE: The test described above is impossible given the preceding Semantic Net because the Web browser and Web server share http and TCP/IP on the preceding Semantic Net. Fortunately all that's required to make the test possible is to refine the preceding Semantic Net to break TCP/IP into components that can be put squarely on either the browser or the server side, probably with relationships connecting them. Those relationships can then be strongarmed to narrow the scope. This is very much like troubleshooting an audio amplifier with a feedback loop from output to input.  That feedback loop must be temporarily broken to draw meaningful information from strongarming the input and observing the output.

This illustrates the real power of Semantic Nets. It provides the Troubleshooter with a straightforward tool to improve insufficient Mental Models during the hot persuit of a solution.

---

Depending on the problem description he's received, the Troubleshooter can start directly from the appropriate concept. Imagine that the problem description is: "I can't access the Web application from my browser, but I'm able to access any other page on this server.". Given this description, an experienced Troubleshooter will immediately suspect the CGI program and navigate in the semantic net from this concept. A less experienced Troubleshooter will have to start from another concept, like for instance the Web server . Navigation through the semantic net is the Troubleshooter's choice, he also can explore several paths in parallel. Thus the Troubleshooter's efficiency higly depends on the semantic net navigation capabilities. The navigation software shall provide this flexibility.

Conclusion

---

EDITORS NOTE: Marc needs to go broader and deeper to fulfill the promise of his article. I have issued Marc a important series of challenges in the next article.

---

References Mentioned in this Article

• http://www.cm.cf.ac.uk/Dave/AI2/AI_notes.html provides a very good introduction to knowledge representation formalisms, including semantic nets.
• http://www.troubleshooters.com/tpromag/199901/199901.htm january 1999 issue of the Troubleshooting Professionals Magazine.
• http://www.troubleshooters.com/tpromag/9811.htm#GeneratingWebDecisionGraphsusingPerl former author's article about decision graphs.

Marc-Henri Poget is a Swiss software engineer, working in a hospital computer department. His main interests are: network and Web applications, network management, and troubleshooting. He can be reached at mpoget@hospvd.ch.