Information extraction for retrieval in dynamic system administration

Computer networks, and in particular hosting server farms, have become very complex systems, and the requirement for 24/7 service requires fast correction of any network and server failures. Correcting network and server failures is a time-consuming, knowledge-intensive process that requires in-depth knowledge of the network setup and the relevant monitoring tools. System administrators need the ability to diagnose the problem from any information available (trouble tickets, diagnostic reports, log file data), in order to identify and perform quickly the required corrective actions. In practice, network and web hosting server system administrators accumulate experience over time, but documenting their experience for others is time consuming and therefore impractical. It is estimated that most of the troubleshooting time is spent on two steps of the troubleshooting process, (1) Identification of the root cause, and (2) Planning the resolution. The cost savings to a typical ISP company that employs thousands of technical support staff will be in the tens of millions of dollars if a solution can be found to reduce the troubleshooting time by 30%. The objective of the NetPal-2 project is to develop a second-generation troubleshooting assistant that: (1) automatically extracts and saves into a system-specific knowledge base the relevant information to problem cases already resolved; (2) retrieves cases from the knowledge base that are potentially relevant to an unresolved problem, thereby helping the system administrator to identify causes and locate a solution much faster than looking up generic Web resources.