Network element and service configuration is currently a manual process. System administrators configure devices and services by editing files, using text based interactive shells, or through some forms based graphical user interface. Each of these methods entails different risks. For example, a manually edited file may contain syntax errors, and changes through a shell or a graphical interface may not be easily reversible. State of the art installations apply a combination of version control software and device/service-specific syntactical and semantic checkers (e.g. DNS lint) to maintain correctness through change.
The current methodology is not without its risks since users may subvert the version control system, or push through changes that were flagged as erroneous by the checker (the entire .com domain was inaccessible for over 24 hours when a Network Solutions employee ignored warnings and pushed through a corrupted DNS root record). Moreover, the syntax and semantic checkers must themselves be configured to enforce the local domain policies. Since each checker is custom developed for each device or service, administrators must learn several different languages for expressing configuration and policy constraints.
More importantly, this whole approach fails when configuration constraints must be enforced across devices or services. Currently, there is no alternative but to depend on the expertise of system administrators who must constantly maintain these high level constraints when making configuration changes. Sometimes, through human error, or an unforeseen new interaction between services, multi-device/service configuration changes may result in partial, or total network failure. Network administrators are then forced to undo these changes by individually restoring each configuration file. If the correct order is not maintained, some changes are not undone, or due to side effects which have made the previous network state unattainable or unstable, normal network service may still not be restored.
The goal our research is to provide core technologies for automating those aspects of network configuration. Our approach focuses on :
More background on the project, which is called NESTOR (NEtwork Self managemenT and ORganization) may be found in the URL :
http://www.cs.columbia.edu/dcc/nestor/