|
Navigation
 This wiki This page |
Human Error Fault InjectionPODS course / EPFL
Erik DeLamarter, Conail Stewart Problem Description, Motivation, Proposed Solution, Proposed Implementation, Milestones Problem DescriptionFew methods exist for testing the impact of human error on system dependability. Of those, they are mostly developed inhouse by enterprises with specific host applications in mind. Unfortunately, this approach tends to produce confidential methods that would be difficult to apply across a broad range of applications.MotivationHuman errors constitute a significant percentage of the recorded causes of the system failure. Currently, few tools are available to test the impact of human errors on host systems. Primarily, this is due to the difficulty in precisely defining what constitutes a human error, however there is also the problem of adequately simulating human behavior.Proposed SolutionTo solve this problem we will design a Framework to recreate human errors. This framework will be designed to interact with the host system in the same way a human does, be it in an administrative sens (configuration files for example) or a client view (requests to the system).Proposed ImplementationThe implementation of our solution will be a Tool capable of recreating human errors in the form of erroneous data input to the system or misconfiguration. The tool will be capable of running the host system and interacting with it at a “human” level, thus it will be capable of rewriting configuration files, starting and stopping the service as an administrator could, and issuing requests to the system, like SQL or HTTP request in the event of a database or web server. This tool will therefore run at a higher level than fault injectors which usually interact with the code or the program's execution.Milestones
|
Share |