Multiple linear regression based parameter influencer model of a Self-healing network

Autonomic features can be built into systems by including Self-* properties. Amongst the various Self-* properties the Self-healing feature is important because it ensures increased uptime and no or less downtime by healing the system and/or its components. System adminstrators desperately need help as complexity of networks and their management systems continue to grow unabatedly. In this paper, we describe how a multiple linear regression based parameter-influencer model can be used to build in Self-healing features in a network. We describe a scenario in a system where each parameter in the system is affected distinctly by many influencers. Statistical techniques are used to identify extent of dependency of parameters on more than one influencer. Empirical data and smaller ranges of values of data from case studies encouraged us to assume a linear relation as a starting point. After identifying the relationship amongst them the appropriate influencers are modified so as to affect the values of parameters. They are thus brought away from their threshold values and near their median values and in turn stability is brought into the system. Our model is programmed as a daemon on the network and a case study is described where historical data is used to tune the network parameters more accurately.