Computer Performance Engineering

Survivability is a critical attribute of modern computer and communication systems. The assessment of survivability is mostly performed in a qualitative manner and thus cannot meet the need for more precise and solid evaluation of service loss or degradation in presence of failure/attack/disaster. This talk addresses the current research status of quantification of survivability. First, we carefully define survivability and contrast it with traditional measures such as reliability, availability and performability [2, 8, 7]. We use “survivability” as defined by the ANSI T1A1.2 committee – that is, the transient performance from the instant an undesirable event occurs until steady state with an acceptable performance level is attained [1]. Thus survivability can be seen as a generalization of recovery after a failure or any undesired event [3]. We then discuss probabilistic models for the quantification of survivability based on our chosen definition. Next, three case studies are presented to illustrate our approach. One case study is about the quantitative evaluation of several survivable architectures for the plain old telephone system (POTS) [5]. The second case study deals with the survivability quantification of communication networks [4] while the third is that of smart grid distribution automation networks [6]. In each case hierarchical models are developed to derive various survivability measures. Numerical results are provided to show how a comprehensive understanding of the system behavior after failure can be achieved through such models.