Enterprise telephony servers for large enterprises in business sectors like finance and healthcare, are complex software systems that require at least five nines of availability. Such systems need guaranteed service with minimal service disruption during failures and are built with availability designed into various aspects of operation. One of the approaches used to improve availability is detecting and predicting failures through analysis of system traces. System trace/debug logs are very textual in nature and as such are not amenable to a fully automated analysis. In this work we develop a log analysis technique for analyzing trace logs of large enterprise telephony systems. We analyze the logs to define normal and failure operational states of a running system and then categorize the current state of the system based on previously learned states. Our analysis shows that the technique is successful in correctly identifying the failure states of the system.