On feasibility of predicting system crash based on local observations : a simulation study

• Main Author: Lyu, Junyi
• Other Authors: XIAO Gaoxi
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/150488
• Institution: Nanyang Technological University

This dissertation mainly introduces the feasibility of predicting system wide crash based on gathered information of local observation. First, the system is abstracted as a scale-free network which is a reasonable abstraction of the most real network. Then, based on the network model, the processes of system crash are observed by using the KQ-cascade crash model. On the basis of KQ-cascade crash model, two kinds of local observation methods are carried out. The first one focuses on the big hubs of the network, while the second one focuses on the general situation of the network by selecting a few samples randomly. A large number of simulations are carried out to prove the feasibility of local observation methods. In the local observation method on big hubs, the simulation results show that the trends of the change rate of the proportion of the critical nodes in both one-hop and two-hop neighbors begin to increase almost at the same time as that of the remaining nodes proportion, when the pseudo-steady state system is about to have a sudden crash. Similarly, in the local observation method on randomly selected nodes, the change rates of the proportion of the critical nodes in the one-hop and two-hop neighbors of the samples and the change rate of the remaining nodes also begin to raise up almost simultaneously. Therefore, the local observation based on the critical nodes is effective to estimate the global situation, especially at the pseudo-steady state and the sudden crash. To sum up, local observation based on the critical nodes of their neighbors is a feasible way of monitoring the system crash.