NODE REACHABILITY ESTIMATION BY EAVESDROPPING

Traditionally, management stations check the reachability of network n odes by polling them at regular intervals. This method is rather costl y regarding network load and CPU usage on the management station, espe cially when the number of nodes to check gets large. The measurements may also have been influenced by delays in intermediate communication paths. We propose a method that is based on special modules in network monitors that estimate reachabilities by looking at the traffic of no des on the same segment as the monitor. The problem of relating networ k traffic to reachability estimates is solved by the use of self-learn ing neural networks. After training, the modules only need to send an alarm signal to a management station if they conclude a node is no lon ger reachable, thereby alleviating a management station of doing tests on machines that are perfectly reachable.