OTKA-NKTH CNK - Fundamental Research Problems of the Future Internet

Project Data

title:	Fundamental Research Problems of the Future Internet
duration:	2009/06 - 2012/05
leader:	Sándor Molnár
members:	József Bíró, Markosz Maliosz, Trinh Anh Tuan, Balázs Sonkoly, András Gulyás
type:	OTKA-NKTH CNK-77802
budget:	27,000,000 HUF
homepage:

 

Research motivation and main goal

Teletraffic theory originally encompassed all mathematics applicable to the design, control, and management of the public switched telephone networks (PSTN): statistical inference, mathematical modeling, optimization, queueing and performance analysis. Later it was extended to include data networks such as the Internet. Internet engineering, an activity that includes the design, management, control, and operations of the global Internet, would thus become part of teletraffic theory, relying on the mathematical sciences for new insights into and a basic understanding of modern data communications.

The emerging future Internet will include a versatile sets of devices connected in a complex network. This network is intended to offer a broad range of services with guaranteed quality and reliability. An important feature of this network is the extreme size with more billion nodes and links. It is also important that this network will not only be huge in size and complexity but it will also include super high speed links to transfer data traffic.

The traffic design and dimensioning of such a huge complex network will be the challenge of the coming decades. There are well established methods and techniques of the modern teletraffic theory for traffic analysis and modeling, traffic prediction and traffic dimensioning, etc. in smaller networks. Based on these methods a range of traffic management techniques has been developed in the previous decades.

However, the application of these methods for the emerging Internet with the mentioned complexity is challenging and in most cases cannot be performed at all. Moreover, there are traffic engineering problems which have not been addressed so far and there is a lack of techniques to handle them. The tasks of the future teletraffic theory are to provide the tools and methods for an efficient traffic engineering of future Internet.

This basic research program is intended to serve as a framework to establish mathematical tools and methods to set up the teletraffic theory of future Internet.

Research problems

A collection of problems is needed to be addressed to get a good understanding of the traffic characteristics of large complex networks and also to propose new mathematical models and design principles for traffic engineering of future Internet.

The following research questions are addressed:

• what kind of traffic implications of traffic aggregation are there in large, complex networks? (will traffic be smooth or bursty, correlated or non-correlated, fractal or non-fractal, etc.)
• which are those traffic characteristics that are the most influenced by the new expected services of future Internet?
• what are the expected dominant services of future Internet and what are the traffic characteristics of them?
• what are the traffic laws of the future Internet? (is there any simple Erlang formula of future Internet?)
• are there any simple and robust design and dimensioning principles in large, complex networks?
• what are the traffic models for design and management, etc. purposes of large, complex networks?
• what are the expected user behavior characteristics in large networks?
• what are the traffic characteristics implications of service grouping in large networks for traffic design?
• what are the traffic implications of new network architectures and transport protocols for traffic design in large networks?
• what is the queueing behavior of huge traffic aggregates?
• what are the performance implications of network traffic in large, complex networks?

 

Research description

The Internet traffic exhibits a completely different nature from classical telephone traffic and the characterization is not as simple as it was in the case of conventional POTS traffic [1],[2]. The main difference is that in traditional telephony the traffic is highly static in nature. The static nature of telephone traffic resulted in universal laws governing telephone networks like the Poisson nature of call arrivals [1],[2]. This law states that call arrivals are mutually independent and exponentially distributed with the same parameter. The Poisson call arrival model had a general popularity in the last fifty years. The great success of the Poissonian model is due to the parsimonious modeling, which is a highly desirable property in practice.

Another universal law of the POTS traffic is that the call holding times follow more or less an exponential distribution [1],[2]. This model was also preferred due to its simplicity and analytical tractability in spite of the fact that the actual telephone call duration distribution sometimes deviates significantly from the exponential distribution.

The validation of these laws in Internet has failed in several cases [1]. The statistical characteristics of these services are significantly different from voice calls. Especially, the call durations become much longer and more variable compared to classical voice calls. These changes call for reviewing the old laws.

By finding similar universal laws for data traffic and to set up the “Erlang formula of future Internet” is one of the biggest challenge of future teletraffic theory.

One of the remarkable characteristics is the fractal nature of the Internet traffic [3], [4]. In this framework long-range dependence (LRD) and self-similarity have been detected and a group of studies is concentrated on how to detect accurately the LRD property and how to estimate the Hurst parameter [3], [4]. A large group of traffic models (fractional Brownian motion (fBm) models, FARIMA models, Cox’s M/G/1 models, on/off models, etc.) to capture LRD and self-similar properties have also been developed [5], [6]. The performance implications of the fractal property are also addressed in a series of studies, e.g. [7], [8].

In this research we are re-addressing all of these questions concerning the traffic generated by present or predicted future applications. The BME-TMIT is participating in the EU CELTIC TRAMMS research project [9] and this gives opportunity to access measurements on operating networks, among others, in Sweden and Spain. We are also planning to do measurements in the Hungarian academic network.

Based on these measurements a comprehensive traffic analysis will be performed. Basic statistics of the traffic on different aggregation levels with special scaling, clustering, etc. analysis are planned. An emphasis of this research is to find dominating traffic characteristics and to identify scaling invariant features of the traffic characteristics.

The goal is to construct simple and efficient traffic models and characterizations which well suits both the measurable and/or available parameters on traffic flows and the underlying network performance evaluation models. For example, creating parsimonious traffic models and finding rules for establishing the “Erlang formula of future Internet” is desirable.

The proper design and operation of adaptive network control functions require performance evaluation methodologies and tools, on one hand to make possible the prediction the performance of prospective control functions, and on the other hand to help the provision of quality of services and economical use of network resources. Mainly due to the quickly changing heterogeneous networking environment, there is an ever demanding need of continuous innovation in enhancing and renewing performance evaluation methods. These methods play central role in revealing fundamental characteristics of traffic flows and network nodes, identifying deficiencies of control functions, designing and evaluating new ones and, eventually introducing them into operation.

In this project we also plan to introduce new teletraffic tools and methods for network performance evaluation, with special emphasis on the theory of statistical traffic multiplexers, large deviation techniques and network calculus.

Network calculus is a widely known deterministic performance evaluation tool for packet-based communication networks. Numerous achievements highlight the power of this tool, for example, the re-definition of Expedited Forwarding per hop behavior. Nevertheless, the deterministic nature of this modeling framework does not allow to utilize an important and valuable phenomenon in packet based networks, the statistical multiplexing of flows aggregated. During the past few years mainly this lack motivates the raise of stochastic (probabilistic) network calculus. Stochastic extensions could be used for obtaining tighter performance bounds of QoS measures (end-to-end delay, buffer requirement), and eventually for better resource utilization.

Along with this, we plan to further investigating statistical multiplexing models, especially suitable for large network traffic aggregates, to improve stochastic network calculus,
researching possible applications of large deviation theory in (stochastic) network calculus and eventually arriving at efficient methods for obtaining the network performance with (but not limited to) these modeling frameworks.

Expected results

The answers for the above addressed research problems can help the research community to settle down the fundamental pillars of the teletraffic theory of future Internet. These results
can be summarized as follows:

• traffic aggregation models in large, complex networks
• asymptotic queueing characterization of large network traffic aggregates
• design and performance evaluation of traffic management functions in large, complex networks
• traffic laws of large, complex networks
• models and principles for traffic design and dimensioning of large, complex networks

 

Influence of research

The research goal is in line with the goals of future Internet research in both the USA (GENI) and the EU (FIRE). This research also serves as a part of the NIIF Program for next generation Internet research in Hungary.

This research can also be a continuation of the teletraffic research at the Department of Telecommunications and Media Informatics, Budapest University of Technology and
Economics (BME-TMIT) led by Sándor Molnár. This work has resulted in a well established teletraffic school at BME with a wide range of national and international research cooperation in the previous decade.

The research proposal can also be well included in the PhD research programs of the Department and can serve as a framework for research labour supply.

The expected results will help the teletraffic community to set up the teletraffic theory of the future Internet. This theory is vital to develop efficient designing, management and control methods of the future Internet. The traffic engineering techniques can only be developed for the future Internet based on these expected results.

References

[1] W. Willinger, V. Paxson: Where Mathematics Meets the Internet, Notices of the American
Mathematical Society, vol.45, no.8, Aug. 1998, pp. 961-970.

[2] J. Roberts, Traffic Theory and the Internet, IEEE Communications Magazine, January 2000.

[3] J. Beran, Statistics for Long-Memory Processes, Chapman & Hall, One Penn Plaza, New York, NY 10119, 1995.

[4] P. Abry and D. Veitch, “Wavelet analysis of long range dependent traffic,” IEEE Trans. Inform. Theory, vol. 44, no. 1, pp. 2–15, Jan. 1998.

[5] W. Willinger, M. S. Taqqu, and A. Erramilli, “A bibliographical guide to self-similar traffic and performance modeling for modern high-speed networks,” in Stochastic Networks: Theory and Applications, F. P. Kelly, S. Zachary, and I. Ziedins, Eds., Oxford, 1996, vol. 4 of Royal Statistical Society Lecture Notes Series, pp. 339–366, Oxford University Press.

[6] S. Molnár and I. Maricza eds., “Source characterization in broadband networks,” Interim report, COST 257, Vilamoura, Portugal, Jan. 1999.

[7] A. Erramilli, O. Narayan, and W. Willinger, “Experimental queueing analysis with longrange dependent packet traffic,” in IEEE/ACM Trans. on Networking, New York, NY, Apr. 1996, vol. 4, pp. 209–223, ACM Press.

[8] A. Erramilli, O. Narayan, A. L. Neidhardt, and I. Saniee, “Performance impacts of multiscaling in wide-area TCP/IP traffic,” in Proc., IEEE INFOCOM 2000, Tel Aviv, Israel, 2000, vol. 1, pp. 352–359.

[9] http://www.celtic-initiative.org/Projects/TRAMMS/

[10] P. Thiran, J-Y Le Boudec, “Network Calculus - A theory of deterministic queuing systems for the Internet” Springer, 2002.

---