Elsevier

Performance Evaluation

Volume 66, Issues 9–10, September 2009, Pages 505-523
Performance Evaluation

Measurement-based optimal resource allocation for network services with pricing differentiation

https://doi.org/10.1016/j.peva.2009.03.003Get rights and content

Abstract

In this paper, we introduce a model for allocating available resources in service-oriented network, with particular focus on delay sensitive services. The model is based on a pricing scheme for the offered services and also takes into consideration the quality of service requirements of each service class through a probabilistic delay-bound constraint. The proposed policy is dynamic in nature and relies on online measurements of the incoming traffic for adjusting the class allocations. We illustrate its performance and its robustness to various tuning parameters through an extensive simulation study that considers various simulation scenarios including experiments based on real network traces.

Introduction

In this paper, we introduce a model that ensures efficient resource allocation, while maximizing the provider’s utility in service-oriented networks. Our model considers a pricing scheme for the offered services and the quality of service (QoS) requirements of each service class, which operates under a probabilistic delay-bound constraint.

Our work is motivated by the immense advances in the information technology sector which have led telecommunication providers to enrich their service portfolio with value-added services so as to drive the success of their businesses. Plain transport services are not profitable and, hence, services such as triple/quadruple play, multimedia messaging and presence are emerging via the service-oriented architecture paradigm, which is the foundation for next-generation telecommunication networks. Furthermore, services offered through virtual environments constitute another trend; specifically, people get the opportunity to utilize applications that demand significant computing resources and expensive licensing using remote high performance servers [1] or even attend lectures and presentations being offered via virtual worlds [2].

The above services require careful management of the available network (e.g., bandwidth, network storage) and computing resources (e.g., memory, CPU capacity). Allocation of resources should be done optimally and dynamically, since static sharing can lead to under-utilization of the network infrastructure. In this paper, we concentrate on the optimal allocation of a single network resource.

Our focus is on delay sensitive web services in service-oriented networks (SONs) [3]. Nevertheless, the proposed framework is rather general and applicable to other types of traffic flows. The emphasis is on services facilitated by the advances in SON appliances [4], [5] that enable application layer functionalities such as XML-aware networking and content-based routing. In [6], it is noted that as networks carry more XML and web service traffic, networks will need hardware-accelerated, content-aware tools to process the contents of application message payloads and take various policy-driven actions in response to what they find. The idea of computations on messages passing through network elements was first introduced in active networks and a survey can be found in [7]. Application-level awareness is also discussed in [8] and is considered to be of key importance in future Internet design. Such awareness can be employed on web services like e-commerce, web auctions, stock quotes and banking transactions that exhibit delay sensitive characteristics. The delay sensitive nature of such services may play a critical role on whether the transaction will be eventually completed or not; users experiencing long delays are eager to postpone or even desist the procedure. Therefore, it is in the provider’s best interest to enhance the quality of experience of such activities.

The aforementioned service quality improvement can be achieved by adopting an optimal resource allocation scheme like the one proposed next. In our scheme, pricing is used to attain service differentiation based on the priority and the value of each service. We also employ online feedback control that evaluates the performance of each service class based on the current traffic characteristics and the desired QoS specifications and reacts accordingly. Moreover, we leverage an exponentially weighted moving average (EWMA) control scheme to predict imminent arrivals and capture significant traffic changes. We assume a fractional Brownian motion (fBm) traffic model because of its ability to adequately capture characteristics of Web traffic, such as self-similarity and the presence of heavy tailed marginal distributions (see Crovella et al. [9] for the ubiquitous presence of self-similarity in Web traffic). Further, evidence of self-similar phenomena in e-commerce HTTP traffic is established in [10].

Our flow control methodology follows along the lines of the seminal papers by Kelly et al. [11], [12] in which the network utility maximization (NUM) problem is analyzed. Their major contribution is the proposal of a distributed algorithm, that yields a fair allocation of the network resources. In addition, the algorithm is stable in the sense that it will always converge to a solution despite any possible perturbations on the information that the algorithm collects from the network. The model presented in this paper differs from Kelly’s work; we study a single network resource and the optimization is solved in a centralized fashion. A distributed allocation of multiple network resources is part of ongoing work.

Flow control has been studied extensively in the context of wired networks [13], [14], [15], [16], [17], [18], [19], [20]. In [13], an optimization approach to flow control is presented, where the goal is to maximize the total utility of all sources over their transmission rates. The basic algorithm solves the dual problem and involves links calculating bandwidth prices and sources selecting rates based on current link prices. In [14], the authors investigate the problem of allocating transmission data rates to users in the Internet in a distributed fashion. In addition to the customary concave utility functions, they propose the use of sigmoidal (non-concave) ones as appropriate for capturing the elasticity of delay sensitive services, like video and audio. Therefore, a non-convex optimization problem needs to be addressed. In [15], the authors study the utility maximization problem in networks where flows arrive and depart dynamically (as opposed to Kelly’s work where flows appear to have infinite backlog to transfer). Their objective is to maximize the long-term expected system utility, under the link capacity constraints. In [16], a game-theoretic framework for bandwidth allocation for elastic services (an elastic service is defined as a service that can modify its data rate according to the available bandwidth within the network) is proposed and a distributed algorithm that yields the optimal and fair allocation is provided. In [17], the authors propose to maximize a utility function specified by the network subscribers and resources are shared based on the solution of that optimization problem. The above papers are mostly concerned about congestion control on the Internet (e.g., a typical application of rate control appears in the TCP traffic context). Nevertheless, studies based on the NUM framework have emerged for power control and rate adaptation in wireless networks [21], [22], MAC protocol [23], etc. A nice tutorial on cross-layer optimization can be found in [24].

Our choice of incorporating the delay requirements of flows into the objective function closely resembles the work in [18], [19]. Specifically, Xu et al. [18] propose a measurement-based resource allocation scheme based on a linear pricing model and average queue delay guarantees. The main disadvantage of that work was the lack of scalability to large number of traffic classes. Moreover, the proposed QoS metric of average queue delays may provide inferior performance compared to the one obtained through stochastic delays. In [19], network utility maximization is achieved through utility functions that incorporate the delay requirements of incoming traffic classes and distributed algorithms are proposed for the solution of the optimization problem. However, the work does not address the frequency and the conditions for solving the resulting optimization problem. In addition, they rely on the not-so-reliable average delay metric. Congestion management in a network where users have different delay requirements is also studied in [20].

The online traffic control part of this work utilizes the EWMA control scheme [25]. EWMA has been considered in [26] where the authors monitor traffic intensities so as to optimally allocate the resources of a Switched Processing System (SPS — a SPS represents a canonical model of systems characterized by the flexible service requirements of incoming traffic flows). Traffic measurements play also a key-role in [27] for setting the optimal pricing scheme that maximizes social welfare using traffic monitoring. Similarly, an optimal measurement-based pricing scheme for M/M/1 queues, where the total charge depends on both the mean delay at the queue and arrival rate of each customer is presented in [28].

The main contributions of the paper are twofold: first, the introduction of a utility function that incorporates a cost component which relies on a probabilistic delay metric, which to the best of our knowledge, is the first attempt of integrating a stochastic delay metric into the NUM formulation. Second, our optimization is done in a dynamic and efficient manner by employing the EWMA control scheme. The paper is organized as follows: the modeling framework of our measurement-based optimal resource allocation (MBORA) system and its components are introduced in Section 2. In Section 3, we present the optimization problem to be solved, while in Section 4 we illustrate the MBORA module responsible for online traffic measurements and monitoring. Finally, we provide extensive experimental evaluation results in order to provide a more thorough understanding of the performance of our system and then we draw our conclusions.

Section snippets

Modeling framework

The employed modeling framework was introduced in [18] and is depicted in Fig. 1. In its present form it represents a single network element, which may correspond to either a traditional network component, such as a switch or a router, or a modern network “service center”, like IBM’s Datapower service-oriented network appliances [5] or Cisco’s application-oriented network message routing systems [4].

It is assumed that the network element serves two categories of traffic classes; deterministic

Pricing model and problem formulation

In the first part of this section, we introduce our utility function that incorporates ideas from economics and pricing of communication networks [33]. Further, we elaborate on our probabilistic delay metric. In the second subsection, we discuss the solution of our convex optimization problem.

Online traffic measurement and traffic monitoring

Next, we discuss the measurement module, whose main responsibilities include online traffic measurements and traffic monitoring. Traffic changes are monitored using the EWMA control chart and when traffic changes of any service class are detected, an out-of-control signal is triggered. Then, a process for online traffic estimation is initiated and when it is completed the new traffic parameters are passed to the optimization module.

As previously mentioned, it is assumed that traffic is governed

Performance evaluation

In this section, we conduct a comprehensive evaluation of the proposed framework in the over-provisioned case through several numerical case studies and simulations. First, we examine the sensitivity of our system with respect to various parameters, such as the mean arrival rate, the pricing parameters and the delay bounds. We examine scenarios involving two types, as well as six types of service classes to investigate the robustness of the system. Then, we compare the performance of our scheme

Conclusion

In this paper, an integrated framework for optimal allocation of resources of an application-aware network intermediary that serves next generation network services was proposed. Services are priced based on their priorities and their desired QoS policies. Traffic estimation and monitoring tools were used so as to supply an autonomic, measurement-based framework in which the optimization problem is solved only when traffic shifts appear. Extensive simulation results that include real life data

Acknowledgments

The authors would like to thank Professor Stilian Stoev for making the code of the LASS tool available to them. Further, we would like to thank the anonymous referees for their useful and constructive comments.

Parts of second author’s work have appeared in the 2007 Sarnoff symposium and the 1st IEEE Workshop on Enabling the Future Service-Oriented Internet. GM was supported in part by NSF grants CCR-0325571 and DMS-0505535. MD was supported in part by IBM and by the Center for Advanced

M. G. Kallitsis was born in Nicosia, Cyprus, on July 1980. He received the Diploma degree in Electrical and Computer Engineering from the National Technical University of Athens (NTUA), Athens, Greece, in 2005 and his M.Sc. in Computer Engineering from North Carolina State University, Raleigh, USA in 2007. Currently, he is working toward the Ph.D. degree at the same university. For his academic progress, he has been awarded the National Scholarship Foundation (IKY) of Cyprus and Korgialenion

References (38)

  • T. Wolf, Service-centric end-to-end abstractions in next-generation networks, in: Proceedings of 15th IEEE...
  • M.E. Crovella et al.

    Self-similarity in World Wide Web traffic: Evidence and possible causes

    IEEE/ACM Transactions on Networking

    (1997)
  • F. Kelly et al.

    Rate control in communication networks: Shadow prices, proportional fairness and stability

    Journal of the Operational Research Society

    (1998)
  • F. Kelly

    Charging and rate control for elastic traffic

    European Transactions on Telecommunications

    (1997)
  • S.H. Low et al.

    Optimization flow control, I: basic algorithm and convergence

    IEEE/ACM Transactions on Networking

    (1999)
  • J.-W. Lee et al.

    Non-convex optimization and rate control for multi-class services in the Internet

    IEEE/ACM Transactions on Networking

    (2005)
  • K. Ma, M. Ravi, J. Luo, On the performance of primal/dual schemes for congestion control in networks with dynamic...
  • H. Yaïche et al.

    A game theoretic framework for bandwidth allocation and pricing in broadband networks

    IEEE/ACM Transactions on Networking

    (2000)
  • P. Xu et al.

    Profit-oriented resource allocation using online scheduling in flexible heterogeneous networks

    Telecommunication Systems

    (2006)
  • Cited by (9)

    • Dynamic green self-configuration of 3G base stations using fuzzy cognitive maps

      2013, Computer Networks
      Citation Excerpt :

      This procedure ensures that a minimum number of collected samples, so as to reduce the chance of spurious variations. To distinguish significant changes from random variations, an approach based on exponential weighted moving average (EWMA) control charts is used [23]. The resulting value is ultimately controlled against a threshold, in order to convert the continuous value to either element of the discrete set {−1, 1}.

    • QoS-constrained sensing task assignment for mobile crowd sensing

      2014, Proceedings - IEEE Global Communications Conference, GLOBECOM
    • Demand response control for PHEV charging stations by dynamic price adjustments

      2012, 2012 IEEE PES Innovative Smart Grid Technologies, ISGT 2012
    View all citing articles on Scopus

    M. G. Kallitsis was born in Nicosia, Cyprus, on July 1980. He received the Diploma degree in Electrical and Computer Engineering from the National Technical University of Athens (NTUA), Athens, Greece, in 2005 and his M.Sc. in Computer Engineering from North Carolina State University, Raleigh, USA in 2007. Currently, he is working toward the Ph.D. degree at the same university. For his academic progress, he has been awarded the National Scholarship Foundation (IKY) of Cyprus and Korgialenion foundation scholarships. His academic interests include optimization and resource allocation in communication networks, QoS provisioning and traffic characterization of network services. Michael is a student member of the IEEE and of the Cyprus Scientific and Technical Chamber (ETEK).

    G. Michailidis received his Ph.D. in Mathematics from UCLA in 1996. He was a post-doctoral fellow in the Department of Operations Research at Stanford University from 1996 to 1998. He joined The University of Michigan in 1998, where he is currently a Professor of Statistics, Electrical Engineering & Computer Science. His research interests are in the areas of stochastic network modeling and performance evaluation, queuing analysis and congestion control and statistical modeling and analysis of Internet traffic.

    M. Devetsikiotis [S’ 1985, M’ 1993, SM’ 2004] received his Diploma in Electrical Engineering from the Aristotle University of Thessaloniki Greece, in 1988, and the M.Sc. and Ph.D. degrees in Electrical Engineering from North Carolina State University, Raleigh, in 1990 and 1993, respectively. He joined the faculty of the Department of Systems and Computer Engineering at Carleton University in April 1995, and later joined the Department of Electrical and Computer Engineering at NC State where he is now a full Professor. He has served as Chairman of the IEEE Communications Society Technical Committee on Communication Systems Integration and Modeling. He is an Area Editor of the ACM Transactions on Modeling and Computer Simulation, a member of the editorial board of the Journal of Internet Engineering, the IEEE Communication Surveys and Tutorials and the International Journal of Simulation and Process Modeling, and an IEEE Communications Society Distinguished Lecturer.

    View full text