Managing bandwidth allocations between competing recreational and non-recreational traffic on campus networks

https://doi.org/10.1016/j.dss.2008.01.005Get rights and content

Abstract

Network performance is a serious concern faced by many campus network managers across the country. As demand for entertainment-based Peer-to-Peer (P2P) applications that involve the transfer of large audio and video files continues to grow, managers are faced with the increasingly difficult task of determining how much bandwidth should be allocated to these applications. Unrestricted P2P traffic has the potential to monopolize bandwidth and severely degrade network performance. University IT managers are placed in a particularly difficult position, because they must juggle demands for non-recreational traffic without severely restricting recreational use of the network. This paper discusses a solution for optimizing bandwidth allocations on a campus gateway Internet link.

Introduction

The introduction of new applications, changes in user behavior, and increasingly faster access technologies all have contributed to an increased demand for bandwidth — the maximum amount of data transferred over a communications link during a fixed time interval [27]. Network management involves the allocation of finite resources, such as bandwidth, to many different services and applications [16]. Managing bandwidth is critical in terms of traffic engineering, and can have a dramatic impact on both local and end-to-end network performance. In particular, the provisioning of bandwidth associated with recreational peer-to-peer (P2P) file sharing applications has become a major concern for Internet Access Providers (IAP)s in recent years, as applications like Napster, KaZaA, Gnutella, and WinMX impose enormous bandwidth demands on networks and can cause serious performance problems [10], [31], [32], [35].

This paper demonstrates a decision support methodology to set optimal bandwidth allocations for competing recreational P2P traffic and non-recreational traffic (here referred to as non-P2P, or NP2P) with respect to minimizing the total cost of network operations. Total costs include the explicit costs to the IAP associated with network management, as well as the implicit costs resulting from unsatisfied users under various bandwidth allocation scenarios. Management decisions include bandwidth allocations for both P2P and NP2P traffic during specific time periods, as well as the number of allocation changes made during the day. A goal program (GP) is used to estimate both P2P capacity demand and P2P user demand at different time periods and bandwidth allocations. Capacity demand refers to the amount of bandwidth (Mbps) demanded by P2P users, and user demand refers to the number of users actively involved in P2P downloads. A Markov Decision Process (MDP) is used to solve the cost minimization problem.

A real-world example for optimizing bandwidth allocations between competing P2P and NP2P interests is provided using empirical data from a large university. While underlying network topologies, service quality architectures, and data may differ from provider to provider, the example illustrates how the methodology can be applied in practice. Contributions include: the introduction of a relative cost approach for managing bandwidth allocations between competing traffic types, the demonstration of a technique for estimating P2P demand for different time periods and bandwidth allocations that cannot be observed in practice under existing policy allocations, and the demonstration of a practical solution methodology for a real-world network management problem using empirical data from an actual university IAP.

Internet applications typically follow a client/server model, where a dedicated host, the server, functions as the service provider for numerous clients who use the offered services [11], [34]. In the P2P model, there is no semantic difference between client and server. P2P allows users to exchange files directly, without using a centralized server. While P2P applications are not “bad”, many popular applications are recreational in nature. Recreational P2P exchanges tend to involve the transfer of large files that may be hundreds of megabytes (MB) or even many gigabytes (GB) in size [35]. P2P file requests often search a large number of hosts, and transfers may occur from multiple hosts in separate locations simultaneously, generating large amounts of overhead traffic [34]. Many recreational P2P applications also bypass or alter TCP's congestion control mechanism, creating bandwidth “hungry” applications that monopolize the available bandwidth and cannot be effectively managed via TCP congestion and flow control.

Although network management problems attributed to recreational P2P file sharing are not restricted to university networks, the severity of the P2P problem and the wide spread use of traffic shaping solutions is well documented on campuses across the nation [14], [23], [31], [32]. In response to external pressure from the recording and film industries, and due to the network management problems resulting from unrestricted P2P traffic, many universities are actively exploring management solutions to mitigate the negative effects associated with recreational P2P traffic [14], [23]. A simple management solution would be to completely eliminate P2P traffic, however, the use of university and overlay networks is not restricted to non-recreational activities. While a few universities have imposed an outright ban on certain P2P applications, many choose to impose bandwidth restrictions on P2P traffic using some type of traffic shaping solution. This approach is less likely to anger students and allows administrators to minimize the copyright debate [14]. In many cases, universities have no desire to impose restrictive policies on recreational P2P activities, as P2P file sharing applications are very popular among students and such policies may be viewed as oppressive and result in unsatisfied users [14], [26]. It is important to note that not all recreational P2P file sharing involves the exchange of copyrighted material, and all exchanges of copyrighted material are not necessarily illegal [14]. The use of networking protocols associated with P2P file sharing is not inherently illegal and IT managers often view demands to restrict P2P file sharing as excessive and intrusive. A more detailed discussion of copyright and legal issues associated with P2P file exchanges can be found in Refs. [1], [22].

The solution methodology described in this paper is based on a widely used management technique/tool called traffic shaping. Traffic shaping refers to a range of technologies used to promote quality-of-service (QoS) by classifying and prioritizing traffic to control bandwidth utilization [7]. Traffic shaping is not the only QoS management approach available, and IAPs may employ multiple techniques, practices, and/or tools simultaneously. As QoS is a broad topic that can be defined differently and can encompass various techniques, protocols, and services, QoS is explicitly defined here as the ability to guarantee and limit bandwidth for certain users and services [5]. Traffic shaping products, such as Packeteer's PacketShaper™ facilitate network management and improve performance by allowing managers to identify and allocate bandwidth to specific traffic types like Web traffic or recreational P2P file sharing applications. University providers must therefore manage the trade-off of bandwidth between competing recreational P2P traffic, and non-recreational traffic keeping in mind that as the bandwidth allocated to a particular service or traffic type decreases, user satisfaction also tends to decrease [3], [14].

Section snippets

Literature

This research builds directly on a theoretical methodology for optimizing bandwidth allocations between competing traffic presented in Ref. [21], where a generalized utility maximizing framework is presented. Additional published research on bandwidth management and network resource allocation is also applied. Prasad et al. [27] provide an overview of bandwidth estimation metrics, measurement techniques, and tools that are commonly used in end-to-end network performance assessment. Firoiu et

Data

The data used in the research are empirical TCP stream data obtained from a large university. Data collection occurs on the edge router to campus, and captures only those TCP streams destined for off-campus sources, or incoming from off-campus sources. The data are collected and aggregated into hourly time intervals for analysis. The raw TCP session log data for each hour comprise many gigabytes (GB) of data. Table 1 provides an example of TCP stream data. The data are converted from their

Methodology

For convenience, Table 2 summarizes the methology and shows the inputs and outputs associated with each step. Observed demand refers to the empirical NP2P and P2P traffic data collected from the test IAP under the existing bandwidth allocation policy. Since the observed data are biased by the bandwidth policy currently used, P2P demand estimates are required for both the bandwidth and number of users for all alternative policy allocations (those not observed in the empirical data). Unconstrained

Sensitivity results and management implications

Sensitivity analyses are performed to provide insight into how bandwidth allocation decisions may be impacted by changes in relative costs. In practice, IAPs may elect to assign different relative cost values to different traffic types. For example, NP2P traffic could be weighted much more heavily than P2P traffic. This approach would yield solutions more favorable to NP2P users. Likewise, the relative change cost coefficient can be adjusted to favor more static or dynamic policies. Change

Conclusions

A practical, relative cost-based approach for optimizing bandwidth allocations between competing NP2P and P2P traffic is discussed. A GP is presented to estimate P2P demand in terms of both the number of P2P users as well as the bandwidth demanded by those users for different time periods and bandwidth allocations. Sensitivity analyses are performed for both changes in the relative user costs where the value of each traffic type is expressed as a ratio, and for the relative cost of making

David C. Novak is an Assistant Professor in School of Business Administration at the University of Vermont. He received a Ph.D. in Management Science and Information Technology from Virginia Tech. He also holds a M.S. in Applied Economics and a B.A. in Economics from Virginia Tech. Dr. Novak's research interests include traffic management, routing, and quality-of-service (QoS) issues on both telecommunications and transportation networks, and applying operations research methodologies to real

References (35)

  • L.A. DaSilva

    Pricing for QoS-enabled networks: a survey

    IEEE Communications Surveys

    (2000)
  • T. Dawson

    Traffic Shaping. O'Reilly LinuxDevCenter.com

    (August 24 2000)
  • V. Firoiu et al.

    Theories and models for internet quality of service

    Proceedings of the IEEE

    (2002)
  • K.P. Gummandi et al.

    Measurement, modeling, and analysis of a peer-to-peer file-sharing workload

  • W.J. Hopp et al.

    Factory Physics: Foundations of Manufacturing Management

    (2003)
  • D. Joachim

    University Gets Tough on P2P. Internetweek.com

    (February 18 2004)
  • M.O. Junginger et al.

    A self-organizing publish/subscribe middleware for dynamic peer-to-peer networks

    IEEE Network

    (January/February 2004)
  • Cited by (0)

    David C. Novak is an Assistant Professor in School of Business Administration at the University of Vermont. He received a Ph.D. in Management Science and Information Technology from Virginia Tech. He also holds a M.S. in Applied Economics and a B.A. in Economics from Virginia Tech. Dr. Novak's research interests include traffic management, routing, and quality-of-service (QoS) issues on both telecommunications and transportation networks, and applying operations research methodologies to real world problems. He has published in journals such as the European Journal of Operational Research, Decision Support Systems, the Journal of Transport Geography, and Computers and Operations Research.

    View full text