| Enterprise systems are growing larger, more distributed, and increasingly complex. They can be composed of hundreds or thousands of heterogeneous workstations and servers, connected via various networking devices, which allow business users to access critical data via multi-tier applications and web services. They can vary in architecture, available bandwidth, computing power, and the amount of black-box resources employed. System administrators are often required to assess the impact on business operations when an enterprise system component fails, which we refer to as assessing the operational impact. Operational impacts can also be caused inadvertently when enterprise system components are reconfigured. Assessing operational impacts accurately is critical to providing business executives with information needed to allocate limited Information Technology resources optimally---for example, maintenance personnel, time, and dollars.;We claim that assessing operational impact requires that administrators relate the component failure to the affected users in a manner that is clear and understandable by business executives. A number of approaches have been presented to calculate these kinds of impact, but many of these approaches have focused on the calculating the dependencies at the application & infrastructure levels. The applications are important only in that they provide a means for the business users to access their critical business data stored in files, databases and other (possibly remote) repositories, or to contact other users directly in a timely manner. Furthermore, the importance of different sets data will vary over time. For example, a certain set of financial data, and the ability to access and modify this data, might be significantly more critical to the business operations near the end of the fiscal year as opposed to other times. Consequently, to more accurately determine the operational impact, an impact assessment system must also monitor the various data sources accessed, the various applications used to access them, and the periods of time for which accessing these files are truly critical to the business users.;This paper presents a framework for monitoring the dependencies between users, applications, and other system components, combined with the actual access times and frequencies. We use operating system commands to extract information from the end-user workstations about the dependencies between system components. We also record the times that system components are accessed, and use data mining tools to detect usage patterns. This information can then be used to predict whether or not the failure of a component will cause an impact during certain time periods. Furthermore, we designed this framework to require minimal installation and management overhead, and to consume minimal system resources, so that it can be employed on a variety of enterprise systems, including those with low-bandwidth and partial-connectivity characteristics. Finally, we implemented this framework in a test environment to demonstrate the feasibility of this approach. This combination of understanding how and when users access various system components allows us to better assess current and future operational impacts. |
