http://chaos1.la.asu.edu/~yclai/papers/PRE_02_ML_3.pdf

            In most networks, each single node has a minimal effect on the entire network as a whole. The removal of single nodes generally does not endanger the network as a whole. An example of this is the Internet: if a few sites are removed, the entire Internet still stays as a whole and would run as smoothly as it did before the sites were removed. In Adilson E. Motter’s “Cascade-based attacks on Complex Networks,” nodes that do have significant power are focused on. One example of such a network is a power transmission grid. According to Motter, when one node in this grid fails, the balance of the grid fails, and the entire network needs to redistribute the loads of the other nodes so that they can support the links that were broken by the removed node. This can cause a cascade of overload failures. This holds true for routers also. When one goes down, the others must increase the amount of information it transmits to support the downed router. This, in turn, would cause congestion. These two cases are examples of what Motter calls a cascading failure. This is when one failure leads to a redistribution of loads, which in turn causes more failures. The chance that the first failure would cause this cascade is dependent on how much load this one node has. The more important a node is to a network, the greater a chance the cascading failure would occur.

            This article is very similar to our talk on diffusion in networks. We learned that when a certain number of nodes switch to a different product, a cascade would form causing every node to switch. In the case here, when a node with a high load fails, a cascade of failure would occur, causing those around it to fail as well.