GIG network and application system performance analysis using extreme value theory

Abstract

The performance of the Global Information Grid (GIG) and its heterogeneous network systems is a difficult problem facing GIG engineers in terms of characterizing end-to-end performance and defining service level agreements (SLA) between numbers of organizations. The Defense Information Systems Agency (DISA) Enterprise Wide Systems Engineering (EWSE) program management office, within Office of the Secretary of Defense - Network Networks and Information Integration (OSD-NII), is chartered to confront these issues. As the GIG continues to evolve in design and complexity, engineers will need to design robust end-to-end services and network/communications architectures to support the DoD and Intelligence communities. In order to attack these problems, the use of a modeling and simulation methodology (M&S) becomes a cost effective way to determine performance of complex system such as the GIG. The use of tools such as the Joint Communications Simulation System (JCSS) and OPNET Modeler are a few methods for studying the GIG from an end to end perspective to obtain performance metrics such that SLApsilas can be negotiated by DISA Enterprise Wide systems Engineering (EWSE) programs and its customers. Along with the M&S methodology, certain mathematical theories are employed to support these analyses. In this paper, Extreme Value Theory (EVT) is introduced and used in our analysis. EVT is commonly used in the IT industry to analyze the systempsilas performance, in particular the systempsilas robustness and scalability. The analysis employing the EVT is very helpful in identifying the problems generated by the system in extreme cases, therefore reducing the variable space required for discrete event simulations (DES).