Evolutionary Design of Hash Functions

Hash tables are fast associative array implementations which became part of modern world of information technology and thanks to its simplicity became very popular among computer programmers. The choice of proper hash function is very important. Improperly selected hash function can result in poor hash table performance and its application. Currently there are many exceptional implementations of general hash functions. Such functions are not constrained to a concrete set of inputs, they perform on any input. On the other hand if we know the input domain we can design a specific hash function for desired application thus reaching better levels of performance compare to a general hash function. However hash function design is not trivial. There are no rules, standards, guides nor automated tools that would help us with such a task. In case of manual design the hash function author has to rely on his/her knowledge, experience, inventiveness and intuition. In case of such complicated tasks there is sometimes advantageous to choose a different path and use techniques such as evolution algorithms. Natural computing is an approach of certain problem solutions that are inspired by the process of species reproduction as defined by Charles Darwin. In this thesis we will design hash functions for the domain of IP addresses, that serve as an unique network device interface identifier in internet protocol networks. The chosen subset of natural computing is the genetic programming, a very specific technique that is an adequate approach to our problem thanks to its properties. Evolutionary designed hash functions offer good properties. They outperform state-of-the-art generic, human-created hash functions in terms of speed and collision resistance