An Infrastructure to Support Performance Analysis in Complex Systems
Man-Sze Li, A S. Deshmukh, Albert T. Jones
Systems research efforts have increased in a variety of disciplines. Despite these efforts, it is still difficult to predict long-term performance and to understand the relationship between the performance of the parts and the performance of the whole. The traditional approach to dealing with system performance is based on the philosophy of Descartes, which involves three steps. First, decompose the original global problem into independent, local sub-problems (the parts). Second, find solutions to each local sub-problem, ignoring any interactions. Third, recompose these local solutions to get the global solution (the whole). Researchers in a number of fields have been quite successful at developing approaches to optimize the performance of the parts. They have not been, however, as successful with predicting the performance of whole from those parts. We believe that there are two reasons for this. First, they ignore the underlying organizational structure of the system, which can impact its overall performance. Second, the overall system performance is impacted more by the interactions of the parts than it is by their individual performance. Those interactions are captured inherently in the information that there share. For a variety of reasons, that information is often not available. In this paper, we review some of the modeling approaches that are used to estimate performance. We also review some of the recent network research and its relationship to system performance. We focus on a proposed vision of a new information infrastructure called the Interoperability Service Utility (ISU) discuss how this infrastructure can help address the aforementioned two problems.
, Deshmukh, A.
and Jones, A.
An Infrastructure to Support Performance Analysis in Complex Systems, Proceedings of the PERMIS Conference, Gaithersburg, MD, US, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=822623
(Accessed May 31, 2023)