Troubleshooting is by nature a convergent process where one only repairs a component if it seems not to function properly. Monolithic tools such as Cfengine and its relatives[2,3,4,7,8,17,18] attempt to force every attribute of a given device into compliance with a predefined operational policy. Most of these tools are limited to functioning in environments where one can compile and run programs, and cannot be utilized to maintain closed-source vendor components such as routers, switches, dialup servers, and so on. Babble[10] provides the beginnings of a tool for convergent administration of turnkey network services, but is largely limited to conversing with and controlling one device at a time.
Expanding these tools to cover the problem of network-wide troubleshooting seems impractical. They are already limited by their own size and complexity. Some are perhaps reaching the limits of software complexity from the standpoint of usability, predictability, maintainability, and adaptability to new needs.
In this paper, we study the potential for dividing these currently monolithic administrative processes into smaller pieces that work together to accomplish the same goal. Tools designed to be maintained by a small group of experts may require extensive effort from the programmer who wishes to couple new and reusable convergent processes into an existing tool framework. It is more desirable to be able to contribute independent processes that interoperate easily with others without utilizing traditional software coupling mechanisms such as subroutine calls and interfaces.