Elsevier

Computer Networks

Volume 42, Issue 6, 21 August 2003, Pages 765-778
Computer Networks

Virtual private network bandwidth management with traffic prediction

https://doi.org/10.1016/S1389-1286(03)00217-2Get rights and content

Abstract

Dynamic link resizing is an attractive approach for resource management in virtual private networks (VPNs) serving modern real-time and multimedia traffic. In this paper, we assess the use of linear traffic predictors to dynamically resize the bandwidth of VPN links. We present the results of performance comparisons of three predictors: Gaussian, auto-regressive moving average (ARMA) and fractional auto-regressive integrated moving average (fARIMA). The comparisons are based on the mean packet delay, the variance of the packet delay, and the buffer requirements. Guided by our performance tests, we propose and evaluate a new predictor for link resizing: linear predictor with dynamic error compensation (L-PREDEC). Our performance tests show that L-PREDEC works better than Gaussian, ARMA and fARIMA in terms of the three metrics listed above. The benefit of L-PREDEC over the Gaussian predictor is demonstrated in two configurations: a common queue with aggregate link resizing and multiple queues with separate link resizing. In both configurations, L-PREDEC has consistently achieved better multiplexing gain and higher bandwidth utilization than its Gaussian counterpart.

Introduction

Virtual private network (VPN) has a bright future in the world of globalization. Corporations and organizations with branches and facilities across the world are facing the problem of interconnecting their subnetworks hundreds or even thousands of miles away. Traditional solutions based on dedicated links are gradually being replaced by VPN. Using public networks such as Internet, VPN service providers can package and ship user data from multiple access points to the corresponding destinations in a safe and efficient way. Various encryption protocols have emerged to enhance the integrity of the confidential information transferred across public networks. The Internet has also been proven to be cheap and effective in delivering information all over the world.

Besides encryption, the VPN providers must also provide QoS guarantees to VPN users. Let us consider a user with two classes of traffic: a QoS traffic stream which has stringent requirements on delay, jitter and packet loss rate; and a best-effort traffic which needs no QoS guarantee. To meet the QoS requirements, the available bandwidth and buffer space is reserved for the first traffic class based on the provisioning policy at the edge of the network (i.e., the access point). The best-effort traffic can take the rest of the resource whenever it is available. Without dynamic bandwidth resizing, each QoS link is reserved with a static bandwidth of b, which is decided by the QoS requirement. To maintain the QoS at an acceptable level, the reserved bandwidth must be significantly larger than the mean of the data rate due to the burstiness of the traffic. A clever priority-queue based packet dispatcher may make the over-reserved bandwidth available to other data streams, but no QoS guarantee could be made.

VPN providers may want to accurately provision the bandwidth and buffer at the start of each user traffic stream. However, the users may not be able to specify their traffic accurately in the service level agreement (SLA). Therefore, the admission control and traffic provisioning must be supplemented with finer levels of dynamic control to achieve the maximum efficiency. To make more bandwidth available to QoS traffic, we may resize the QoS link based on the history and predicted future value of the traffic rate. If we could adjust the reserved bandwidth to cope with the fluctuation of the incoming traffic rate, the mean of reserved bandwidth will be substantially lower than that of the static provisioning. In addition, Link sharing [1] or hose interface [2] allow the VPN providers to share an access link by multiple VPN streams to make a gain from multiplexing. Dynamic link resizing will also help to enhance multiplexing gain. Other works related to traffic engineering, dynamic resource allocation include [3].

In general, dynamic link resizing works by periodically monitoring the traffic rate of a user link and adjusting the reserved bandwidth based on the prediction made from the traffic history. In [2], local maximum predictor and local Gaussian predictor were proposed to be used in VPN resource management. Local maximum predictor was defined as the maximum rate sampled in the observing window. Local Gaussian predictor can be defined as m+αv, where m=E(Xt) is the mean of observed traffic Xt, v=Var(Xt) is the variance and α>0 is a factor to control over-reservation. Though Gaussian predictor is simple and robust, there are more accurate predictors available for network traffic. Among them, auto-regressive moving average (ARMA) [4] and fractional auto-regressive integrated moving average (fARIMA) models [5], [6], [7], [8] can predict network traffic sequences with lower mean square error (MSE) than the Gaussian predictor.

Section snippets

Linear predictors for network traffic

Suppose we are monitoring a VPN link, and calculating the mean of traffic data rate (bits/s) for every time interval [t1,t2), [t2,t3),  , where ti+1ti=dt is the size of the observing window, then we can get a time series of traffic rate X1,X2,…,Xt,… With the assumption of stationarity (non-stationary traffic will be address later), we could apply a linear predictor on that time series.

Based on the correlation structures of the time series, a linear predictor PtXt+h predicts the future value Xt+h

Linear predictor with dynamic error compensation

In this section, we first introduce an enhancement measure for AR and fAR predictors. We verify this method with the BC trace. Then we move on to discuss test results with longer traces and address several other issues.

Multiplexing gain with predictors

In this section, we simulate an access link that is shared by multiple VPN links (Fig. 25) [2]. Assuming a network service provider is providing VPN services to different users. Each user has signed his/her own SLA with the ISP for the connection ordered. Given limited bandwidth available to the service provider, the goal is to satisfy each SLA while reserving minimum bandwidth.

Fig. 26 shows the first configuration where each VPN link has an independent packet queue and links are resized

Conclusion and future work

In this paper, we evaluated the linear predictors (AR and fARIMA) for VPN dynamic link resizing. Based on the experiments with linear and Gaussian predictors, We proposed a new predictor, L-PREDEC, to improve the response time of the linear predictor to the burstiness of the traffic. Our simulation showed that L-PREDEC can improve the response time of linear predictors. In a VPN environment, prediction based bandwidth resizing will provide multiplexing gain without the degradation of QoS.

Our

Acknowledgements

This work has been partially supported by NSF under grant 0086251. The possibility of using traffic prediction in improving intrusion detection systems has been motivated by research supported by ARO under grant DAAD19-01-1-0502. The views and conclusions herein are those of the authors and do not represent the official policies of the funding agencies or the University of Central Florida.

The authors are grateful to Dr. Zhi-Li Zhang (Editorial Board member) and the three anonymous reviewers for

Wei Cui is a Ph.D. candidate in Computer Science at the University of Central Florida, Orlando. He has been performing research on computer networks and wireless communications and has published several refereed articles in those areas.

References (15)

  • T. Anjali et al.

    Optimal policy for label switched path setup in MPLS networks

    Computer Networks

    (2002)
  • R. Garg, H. Saran, Fair bandwidth sharing among virtual networks: a capacity resizing approach, in: IEEE Infcom’2000,...
  • N.G. Duffield, P. Goyal, A. Greenberg, P. Mishra, K.K. Ramakrishnan, J.E. van der Merwe, A flexible model for resource...
  • P.J. Brockwell et al.

    Introduction to Time Series and Forecasting

    (1996)
  • J. Beran

    Statistics for Long-memory Processes

    (1994)
  • Y. Shu, Z. Jin, L. Zhang, L. Wang, Traffic prediction using FARIMA models, in: ICC’99, 1999, pp....
  • J. Ilow, Forecasting network traffic using FARIMA models with heavy tailed innovations, in: Proc. 2000 IEEE Int’l Conf....
There are more references available in the full text version of this article.

Cited by (19)

  • Structure optimization of BiLinear Recurrent Neural Networks and its application to Ethernet network traffic prediction

    2013, Information Sciences
    Citation Excerpt :

    Leland et al. made the first thorough analysis on the characteristics of the traffic data set [14]. Experiments on this data set have been reported for various purposes including bandwidth management and Quality of Service [25,13,9]. Sang and Li reported an upper bound for the performance that online prediction can achieve with a specified error constraint or expected prediction interval [24].

  • A distributed resource management model for Virtual Private Networks: Tit-for-Tat strategies

    2012, Computer Networks
    Citation Excerpt :

    Traffic prediction is an important and very difficult topic for the provisioning of a network having to support many dynamic VPNs. Numerous models have been proposed in the literature [8,31]. These models have to predict the traffic load behavior of a VPN according to its recent past.

  • Hybrid optimization for QoS control in IP Virtual Private Networks

    2008, Computer Networks
    Citation Excerpt :

    Such huge number is spread across many administrative domains, organized in a hierarchical structure, to cope with complexity and scale. Within this scenario, an additional interesting possibility is that of creating Virtual Private Networks (VPNs) for larger users, thereby further delegating control actions to more limited domains (see, among others [2–6]). A wide range of network architectures and technologies has been proposed (and used) to support this kind of service, based on both “layer 2” (Frame Relay or, in a broad interpretation of layer 2, ATM and MPLS) and “layer 3” technologies (IP over IP, GRE, IPSEC) [7].

  • Adaptive bandwidth provisioning with explicit respect to QoS requirements

    2005, Computer Communications
    Citation Excerpt :

    Consequently, in case packet level QoS requirements are given as the target for provisioning, the scheme of [13] can neither be directly applied. Ref. [14] works out a provisioning scheme for VPN links involving a specific traffic prediction called Linear Predictor with Dynamic Error Compensation (L-PREDEC). The provisioning scheme uses the so called over-reservation factor α to set the bandwidth to ασerr+m, where σerr is the mean square error of the predictor, m is the measured mean rate.

View all citing articles on Scopus

Wei Cui is a Ph.D. candidate in Computer Science at the University of Central Florida, Orlando. He has been performing research on computer networks and wireless communications and has published several refereed articles in those areas.

Mostafa A. Bassiouni received his B.Sc. and M.Sc. degrees in Computer Science from Alexandria University and received the Ph.D. degree in Computer Science from the Pennsylvania State University in 1982. He is currently a professor of Computer Science at the University of Central Florida, Orlando. His research interests include distributed systems, computer networks, real-time protocols and concurrency control. He has authored some 140 papers published in various computer journals, book chapters and conference proceedings. His research has been supported by grants from ARO, ARPA, NSF, STRICOM, PM-TRADE, CBIS, Harris, and the State of Florida. He has served as member of the program committee of several conferences, as the program committee chair of CSMA’98 and CSMA’2000 and as the guest co-editor of the special issue of the Journal of Simulation Practice and Theory, vol. 9, April 2002.

View full text