Resilience of Networked Systems under Connectivity-Based and Load-Based Failures

A general modeling and simulation framework is developed in this work to quantitatively evaluate the resilience of networked systems under two types of failures: connectivity- and load-based. Two newly designed dynamic healing mechanisms are demonstrated. The model considers concurrent cascading failure and healing processes on networks. The discrete-time simulations generate system trajectories, i.e., number of failed nodes at each time step. The 95% recovery time is used as the resilience metric to evaluate and compare the healing performance. Based on two real-world networks, the dependence of system trajectories and resilience metric on various model parameters is explored. If the triggering level (fraction of inactive nodes when healing starts) is too high, the system would either undergo a very slow recovery or never recover to a satisfactory level at all. However, this work provides a counter example to the intuition that the smaller the triggering level, the shorter the recovery time. While low budgets (number of nodes allowed to recover at each time step) lead to prolonged or unsuccessful recovery, it appears that the resilience metric converges to a limit when budget is raised to high enough. This may have practical implications, as node recovery requires resources and a budget too high or too low would be wasteful. This works lays the foundation for subsequent studies on more complex mechanisms and processes on the networks, optimization of model parameters for maximum resilience, as well as applications to more real-world scenarios.