Because of the diverse growth of various networks, however, many problems have emerged. Symptoms of problems in internetworking include the inability to make a connection between hosts, unintentional disconnection, slow transmission, and network congestion such as Ethernet meltdown and broadcast storm. One major drawback of these networks is that only a few experts are able to isolate and resolve such problems. Moreover, network problem is difficult to resolve because different problems often show the same symptom, because some problems are considered only potential problems, and because networks are intrinsically distributed, that is, hosts attached to them are geographically and functionally distributed.
To cope with these problems, we have developed a diagnostic knowledge-based system, called LODES (Large internetwork Observation and Diagnostic Expert System) that carries out automatic, systematic, and effective network troubleshooting and management. This system, attached to a network, attempts to determine the cause of a problem and, if possible, fix it automatically. Even if the problem cannot be fixed automatically, the system supports human experts by automatically supplying important data collected from the network.
LODES is a knowledge-based system for internetwork problems. The main features of this system are automatic detection and problem diagnosis based on its knowledge of protocols, troubleshooting experience, and management data. Each network segment has a copy of this system and together they isolate a problem autonomously or cooperatively depending on the type of problem. Furthermore, LODES has knowledge needed to recognize the seriousness of the problem, thus preventing frequent warnings of small problems. Of course, a number of difficult and complicated problems may be undiagnosable. Still, LODES is valuable because in actual situations, simple problems occur frequently, and such problems have to be automatically resolved.
This system is also used as a testbed for research on distributed artificial intelligence.