Scalable Fault Tree Construction and Analysis

Currently, the best way for a cloud-computing service such as Amazon Web Services or Microsoft Azure to identify likely points of failure in its network is to sit and wait for something to break. In a network the size of Amazon’s, for example, the only way the company might be made aware of the fact that five servers all depend on the same switch to reach the internet would be if that switch failed. Thus, companies are forced to passively react to the consequences of hidden dependencies, rather than fix them before they become a problem. Such companies could benefit from a tool that would allow them to assemble dependency graphs of their network components and use them to identify risk groups, or sets of components which could take down the network if they failed simultaneously. Companies could use this information to strengthen their networks, and end users of cloud services could use it to analyze several cloud services and compare the relative risk in using them.A vital step in the construction of such a tool is the assembly and analysis of the fault tree, a data structure that aims to represent the hierarchy of dependencies upon which a given service runs. The fault tree generator is Python code I wrote which generates a fault tree in the form of a Boolean satisfiability formula from a list of dependency data, and then identifies the most prevalent risk groups in the tree.