A multiscaling traffic model for UDP streams
Introduction
The Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP), the two transport protocols most widely used in the Internet, provide, respectively, connection and connectionless services at the transport layer. They differ in various aspects. TCP furnishes reliable delivery services, with a transmission window governed by both flow control and congestion control mechanisms which introduce processing overhead as well as limit the potential bandwidth a connection can use. UDP, on the other hand, neither furnishes reliable data delivery nor has the mentioned overhead and limitation, so it is more appropriate for real-time applications such as voice over IP and video on demand.
Although the deployment of real-time applications on the Internet has increased, roughly 80% of the traffic still uses TCP as the transport protocol which justifies the large number of traffic models for TCP streams that has been proposed [1], [2], [3]. The proportion of UDP traffic in the Internet does not diminish the need for accurate models for such type of traffic given that UDP does not reduce its transmission rate under congestion situations and consequently it can jeopardize the bandwidth acquired by TCP streams under congestion. Understanding and being able to reproduce the behavior of UDP streams at the packet level is a key for the assessment of the efficacy of congestion control mechanisms. Nonetheless, not too much attention has been paid to such type of model, with the most popular models focusing on the behavior at the flow level [4], [5].
Moreover, the scaling nature of Internet traffic has been subject to debate in recent years. Some authors advocate the use of monoscaling models (monofractal, self-similar) while others prefer the use of multiscaling models (multifractal). The authors of this paper analyzed a large number of IP, TCP and UDP traces to shed light on scaling nature of these types of traffic [6]. Results suggest that TCP streams determine the scaling nature of IP flows as well as that UDP flows are multifractal [6]. Such multifractal nature can be understood by the diversity of packet size transported by the UDP protocol ranging from small packet such as those of VoIP to large ones generated by media streaming applications.
This paper proposes a traffic model for UDP streams, firstly introduced in [7], which reproduces the traffic at the packet level as well as its scaling characteristics. The model is not oriented to the traffic generated by specific applications but rather to the aggregation of the packets generated by applications which use UDP protocols at the transportation layer. This model was based on the identification of marginal distributions existing in UDP flows, as well as the identification of those distributions needing to be considered in greater detail. The model is a 4-state one and its accuracy is compared to that of other multiscaling models proposed in the literature. It can be used to generate synthetic data for simulation experiments at low computational cost. Besides providing an extensive evaluation of the proposed model and expanded comparison with related models, this paper differs from [7] by the precise characterization of the multifractal nature of UDP traffic as well as by the multifractal analysis of the traffic generated by the proposed model.
The following section reviews the concept of scaling in network traffic. Section 3 describes various distributions of UDP traffic streams. Section 4 presents the construction of the model. Section 5 overviews two multifractal models commonly used in the literature. Section 6 provides an analysis of its performance. Section 8 concludes the paper.
Section snippets
Scaling nature of network traffic
Since the seminal work of Leland et al. [8], several studies have shown that network traffic presents scale invariance, or scaling, which is the absence of any specific time scale for which the “burstiness” of a traffic stream can be characterized. Hence, an accurate description of traffic must account for a variety of time scales. Such traffic presents long range dependency which implies that the auto-correlation of the traffic decays very slowly, or hyperbolically; moreover this
Characterization of marginal distributions
The approach adopted for derivation of UDP traffic model presented here involves an initial identification of the distributions which best characterize the relevant marginal distributions of the UDP traffic streams. Some of these distributions are bimodal such as the distribution of packet size. Rather than defining a special distribution to describe the bimodal shape of some of the UDP stream distributions, we modeled the two distinct regions of the UDP stream distribution using two different
Proposed traffic model
It was clear from the data obtained that both modes of packet size distribution need to be considered in the modeling process, with the precise modeling of packet size constituted the initial step in the derivation of the model. States were defined for the two regions of the packet size distribution (packets smaller and greater than 750 bytes). To refine the model, various combinations of potential distributions for packet interarrival time and burst duration were then investigated, and the
Multifractal traffic models
This section describes two multiractal traffic models: the Markovian arrival process and the Multifractal wavelet model which are used in this paper for comparison with the proposed model.
Numerical results
To evaluate the effectiveness of the proposed model, simulation experiments were conducted using both synthetic traces generated by the analytical models and real network traces (Table 1). The aim was to compare results produced using a specific model with those resulting by the use of real network trace.
For the models proposed in this paper, packets are generated according to the interarrival time and packet size distribution associated with each state, with the Weibull distribution used for
Related work
There has been great interest in developing simple models for aggregated UDP traffic that can be used for traffic generation in simulations. Some of these works are briefly surveyed in this section.
In [18], a Hidden Markov model for Internet traffic sources at packet level was introduced. Experimental results show that the model is able to estimate statistical parameters and produce synthetic traces. By exploiting temporal dependencies, the model is able to perform short-term prediction. In [19]
Conclusions
Although there has been an increasing use of the UDP protocols by real-time applications over the Internet, the great majority of Internet traffic is carried by the TCP protocol. Nonetheless, it is highly important to have an accurate model for UDP streams since the UDP protocol does not react to network congestion. This paper introduced a simple model composed of four states that accurately reproduces the patterns existing in real UDP flows. The approach used was to characterize the marginal
Acknowledgements
This work was partially sponsored by FAPESP and by CNPq.
References (23)
- et al.
Multifractality in TCP/IP traffic: the case against
Comput. Networks
(2005) - et al.
Internet traffic modeling by means of hidden markov models
Comput. Networks
(2008) - et al.
Envelope process and computation of the equivalent bandwidth of multifractal flows
Comput. Networks
(2005) - et al.
A multifractal wavelet model with application to network traffic
IEEE Trans. Inform. Theory
(1999) - J.L. Vhel, B. Sikdar, A multiplicative multifractal model for TCP traffic, Computers and Communications, IEEE Symposium...
- V. Ribeiro, R. Riedi, M. Crouse, R. Baraniuk, Multiscale queuing analysis of long-range-dependent network traffic, in:...
- C. Hollot, Y. Liu, V. Misra, D. Towsley, Unresponsive flows and aqm performance, in: INFOCOM 2003, Twenty-Second Annual...
- Y. Liu, F. Lo Presti, V. Misra, D. Towsley, Y. Gu, Fluid models and solutions for large-scale ip networks, in:...
- L. Ostrowsky, Modelos de tráfego para fluxos gerados pelo protocolo UDP, Master’s thesis, Campinas State University,...
- L. Ostrowsky, N. da Fonseca, C. Melo, A traffic model for udp flows, in: Communications, 2007, ICC ’07 IEEE...