How Computer Networks Can Become Smart
Ricardo Bagnasco, Joan Serrat · InTech eBooks · 2011
In the past, computers had a simpler life, few applications had network capabilities and users only wanted to connect to Internet in few specific situations as to check emails, view some mostly plain-text webpage or to play few games with their friends using a “not very wide” bandwidth. But several aspects have changed nowadays, we have better connections, more content available on Internet (the biggest computer network ever known), there are new devices able to connect allowing the presence of more network elements than before and, in addition, the rapid advances in real-time supported applications that are expected, make it difficult to manage them as has been done. Today’s network applications require extensive human involvement in management but we need a more independent (smart) network so research community has been developing lot of work in the last few years trying to ease this issue and to introduce autonomic behaviour in computer and telecommunication networks1. Manage a network is a complex task where the purpose is to control the behaviour of their elements in order to fulfil some high level goals. The automation of networks and systems thus far has been applied to simplified problems. Typically, we can find some works about improving routing2; configuring a network element to better support some application; or controlling the compliance of contracts between peers. Among the different enabling technologies for automate network communications the policy based management is one of the most representatives. This paradigm allows segregating the rules that govern the behaviour of the managed system from the functionality provided by the system itself (Boutaba & Xiao 2002). This is particularly useful in environments where it is necessary to dynamically change goals and add new services or resources. The most well developed implementations of this paradigm rely on pre-programmed rules based on logic. Nevertheless, to be called autonomic, a system must show a degree of flexibility to self adapts to changes and this is hardly achievable by means of static policies. Then, it emerges the need of learning and to figure out the mechanism that will lead to new policies. Another important aspect of a network is that it has high and low level goals so it would be important to understand the whole vertical structure from sending/receiving bits till more complex application tasks in order to combine and enforce the best action choice. One