PeerPressure for automatic troubleshooting

Technical support contributes 17% of the total cost of ownershipof today’s desktop PCs [3]. An important element of technical sup-port is troubleshooting misconfigured applications. Misconfigura-tion troubleshooting is particularly challenging, because configura-tion information can be shared and altered by multiple applications.Maintaining healthy configurations of a computer platform with alarge installed base and numerous third-party software packageshas been recognized as a daunting task [1]. The considerable num-ber of possible configurations and the difficulty in specifying the“golden state” [4], the perfect configuration, have made the prob-lem appear to be intractable.In this paper, we address the problem of misconfiguration trou-bleshooting. There are two essential goals in designing such a trou-bleshooting system:1. Troubleshooting effectiveness: the system should effectivelyidentify a small set of sick configuration candidates with ashort response time;2. Automation: the system should minimize the number of man-ual steps and the number of users involved.To diagnose misconfigurations of an application on a sick ma-chine, it is natural to find a healthy machine to compare against [7].Then, the configurations that differ between the healthy and the sickare misconfiguration suspects. However, it is difficult to identify a