Rule Discovery in Telecommunication Alarm Data

Fault management is an important but difficultarea of telecommunication network management: networksproduce large amounts of alarm information which must beanalyzed and interpreted before faults can be located. So called alarm correlation is acentral technique in fault identification. While the useof alarm correlation systems is quite popular andmethods for expressing the correlations are maturing, acquiring all the knowledge necessary forconstructing an alarm correlation system for a networkand its elements is difficult. We describe a novelpartial solution to the task of knowledge acquisition for correlation systems. We present a methodand a tool for the discovery of recurrent patterns ofalarms in databases; these patterns, episode rules, canbe used in the construction of real-time alarm correlation systems. We also present tools withwhich network management experts can browse the largeamounts of rules produced. The construction ofcorrelation systems becomes easier with these tools, as the episode rules provide a wealth ofstatistical information about recurrent phenomena in thealarm stream. This methodology has been implemented ina research system called TASA, which is used by several telecommunication operators. We briefly discussexperiences in the use of TASA.