This paper provides a comparative evaluation of a number of known classification algorithms that have been considered for both software and hardware implementation. Differently from other sources, the comparison has been carried out on implementations based on the same principles and design choices. Performance measurements are obtained by feeding the implemented classifiers with various traffic traces in the same test scenario. The comparison also takes into account implementation feasibility of the considered algorithms in resource constrained systems (e.g. embedded processors on special purpose network platforms). In particular, the comparison focuses on achieving a good compromise between performance, memory usage, flexibility and code portability to different target platforms.
展开▼