Threshold Attendance under Soft-Crash Model: TAG Protocol and Markovian Analysis

Abstract

A realistic and systematic network evaluation should subsume an availability model and a failure model. We combine a "hard availability" model we call threshold attendance, whereby a certain minimum number of network elements must be present at any time, with a soft-crash failure model, whereby after experiencing a failure, a network element is still able to function correctly for a while in an emergency mode at a risk of a major crash. A Threshold Attendance Guarantee (TAG) protocol, earlier studied from a security viewpoint, is deployed to ensure threshold attendance while controlling the duration of the emergency mode. We study the network under TAG using an "isolated" node-type Markovian analysis, offering insights into the tradeoffs between some relevant availability and reliability characteristics, and showing a simple model-free way to account for a positive correlation between the network elements’ behavior that can reflect mass disasters.