Grimm C., Schlüchtermann G. IP Traffic Theory and Performance

Springer, 2008. — 498 p.Nobody can deny that IP-based traffic has invaded our daily life in many ways and no one can escape from its different forms of appearance. However, most people are not aware of this fact. From the usage of mobile phones – either as simple telephone or for data transmissions – over the new form of telephone service Voice over IP (VoIP), up to the widely used Internet at the users own PC, in all instances the transmission of the information, encoded in a digital form, relies on the Internet Protocol (IP). So, we should take a brief glimpse at this protocol and its constant companions such as TCP and UDP, which have revolutionized the communication system over the past 20 years. The communication network has experienced a fundamental change, which was dominated up to end of the eighties of the last century by voice application. But from the middle of the nineties we have observed a decisive migration in the data transmission.
If the devoted reader of this monograph reads the title ‘IP traffic theory and performance’, she/he may ask, why do we have to be concerned with modeling IP traffic, and why do we have to consider and get to know new concepts. She/he may argue that on the one hand, since the early days of Erlang and his fundamental view on the traffic description in the emerging communication world, formulas and tables have contributed to the building of powerful communication networks. On the other hand she/he may be guided by the argument that, even if we do not meet the correct model, i.e. if our standard knowledge does not suffice and fails, there is enough technical potential in the classical telecommunication network, in terms of equipment, to overcome any bottleneck.
In some respect, we will disprove this misleading attitude. But before going into details, we can already argue that on the one side, and this is done in several parts of the monograph, IP-based traffic does not fit into the classical framework of the Erlang theory. Since the network connections is no longer end-to end built, but it is chaotic at the first glance, and that is the strength of the IP-based networks, since the Internet is self-organized, i.e. deciding more less at each router or node, which route it will take. This introduces the stochastic aspect, which runs through the IP modeling as well through our book as a dominant feature.
On the other side making server or router as powerful as possible so as to make any modeling superfluous, has its decisive drawback. The network is not clearly structured, so that at each possible node, the capacity and service rate is large enough to encounter any traffic load. Only a few bottlenecks diminish the performance and would especially influence extensively the time sensitive traffic, as Voice over IP or video streaming, with its strong quality of service (QoS) requirement. The customer would avoid any of these services as a consequence. This in fact has occurred already in reality and is not a fiction. Incorrect design of networks according to IP traffic requirements lead to a rejection or at least an unexpected delay of new services like VoIP or video on demand.
In addition, a variety of technical possibilities raises more expectations: if we
have the traffic capacity, we want, and we will use it without restriction – and
in turn the network will meet its limits soon. Avoiding these difficulties and
being prepared for future challenges, we present models, indicate consequences
and outline major key aspects, like queueing for judging performances of the
network.
After the discovery of the Bellcore group in the early 1990’s it was Ilkka Norros who developed a first approach in describing the IP-based traffic. He used the fractional Brownian motion as stochastic perturbation to incorporate the basic phenomena of self-similarity and long-range dependence of the connectionless traffic. As already mentioned this is in contrast to the classical circuit switched traffic, where averaging over all scales is leveling the bursty character – this bursty character does not change over all time scales in the IP case. In fact, this approach was not new, since in the sixties Benoît Mandelbrot and John Winslow Van Ness used the fractional Brownian motion and its self-similarity for the description of stock pricing in financial markets. The race for the most appropriate model was opened! A significant variety of models was introduced, which is a consequence of the fast growing number of applications accompanied by different protocols, especially with the TCP/IP. Some tried to describe the large scales and the more Gaussian traffic, other the more bursty traffic and again others the small scales influenced by the control cycle triggered by the TCP/IP. Here, especially the multifractal models and cascades entered the scene, with some of course relatively complicated and with the lack of a suitable chance for application. Up to now a unified theory is missing, which of course may not be near to fulfillment, since the variety of modern networks counteracts these efforts.
Hence, we will present the major models and indicate their advantages and limitation. It is clear that it would be beyond the scope of the book to give a full list of all approaches. It is our aim on the one hand to give a certain feeling, why the IP traffic differs from the classical Erlang description, to give insight into the different approaches, where of course we will to some extent only introduce them and will not go into details. This is left to the reader for further study. We will show, how one can map the traffic to the models using standard statistical methods and we will finally try to answer the question, as to what are the decisive key values, like queueing, waiting time and QoS thresholds. These factors enter finally into the major question of optimization – from the network point of view as well as from the economical standpoint. Summarizing we do not pursue completeness or even a full and profound description of all existing models for IP traffic. In fact, we would like the monograph to be considered as a ‘window not mirror’ as Rainer Maria Rilke in his ‘testamony’ once put it [79]. Writing the book was for us like the metaphor of Rilke, with only a small ‘window hole’ compared to the huge building of nature.Introduction to IP Traffic
Classical Traffic Theory
Mathematical Modeling of IP-based Traffic
Statistical Estimators
Performance of IP: Waiting Queues and Optimization