Traffic shaping based on an exponential token bucket for quantitative QoS: implementation and experiments on DiffServ routers
Introduction
Defining Per-Hop Behaviours that can guarantee quantitative QoS (even in a statistical sense) in a Differentiated Services architecture still remains an open issue. As a result, sensitive traffic that requires specific (even probabilistic) loss and/or delay bounds is usually treated by the Expedited Forwarding (EF) PHB in a DiffServ environment, which merely gives to the aggregated stream of sensitive traffic absolute priority over any other class. This way it is not possible to support more than one class that demands quantitative and/or stringent QoS guarantees. The research community is still investigating the development of traffic control mechanisms enabling quantitative QoS control [1].
Developing such mechanisms is not a trivial task, mainly due to the complexity and multi-level heterogeneity of the Internet (i.e. various link layers, routing protocols and algorithms, protocol versions and implementations, etc.). Furthermore, it is rather hard to find a global and representative model for Internet traffic, which at the same time remains simple enough to enable effective and easy to implement traffic control mechanisms. The problem of Internet traffic modelling is further aggravated due to the constantly changing nature of Internet traffic [2], [3].
In order to cope with the complexities of packet-level traffic control and yet provide mechanisms for controllable and quantitative QoS, a framework for packet level traffic control that relies on the M/G/1 queuing model has been proposed and verified. The core of this framework is an algorithm for shaping of traffic streams through packet spacing [4]. This algorithm is quite general, since it can be applied to virtually any kind of packet-based traffic (non-stationary, non-Markovian, heterogeneous, long-range dependent, etc.) in order to produce streams with specific multiplexing behaviour at network nodes. Packet spacing is based on a non-linear law, which leads in a tight upper bound to the queue-length distribution at a multiplexer fed by many independent shaped streams. This upper bound provides a foundation for statistical yet quantitative QoS guarantees, which is one of the virtues of the framework. It should also be noted that the packet spacing law enforces a specific effective rate to the traffic stream. The enforced effective rate features the convenient additivity property, allowing easy derivation of traffic control and accounting functions, such as policing, admission control, bandwidth allocation and charging.
The theoretical background and the traffic control laws of this framework have been presented in previous works [5], [6], [7]. They have also been studied experimentally and verified by a rich set of simulation and experimental results based on artificially generated traffic following diverse models (exponential, uniform, heavy-tailed, heterogeneous traffic mix, etc.) [7]. Furthermore, experiments and simulations with real traffic traces have investigated the implications of applying this kind of traffic shaping in operational networks. In particular, the effect of the delay introduced by the packet spacing law, as well as strategies for assigning effective rates to the various traffic streams have been studied in [8].
Having this set of sound theoretical and experimental results at hand, this paper focuses on the implementation of the proposed shaping law and its accompanying ‘M/G/1′-based traffic control functions in actual network devices (routers) and specifies how these results can be incorporated into a DiffServ architecture. These are complemented by a set of experiments conducted in a testbed with routers, in which these traffic control schemes were implemented. The purpose of these experiments is not to exhaustively analyse and validate the proposed traffic control scheme (this has already been done in [6], [7], [8]), but rather to demonstrate that the scheme can be incorporated in a real implementation without sacrificing its properties. In terms of implementation platform, Linux was selected mainly due to the fact that it provides a rich and extensible traffic control architecture giving ample room for embedding new traffic control mechanisms. Thanks to their traffic control capabilities, Linux routers can be appropriately configured to act as either edge or core routers in a DiffServ framework.
The rest of the paper is structured as follows. Section 2 briefly reviews the theoretical underpinnings of our framework and illustrates the traffic control scheme for quantitative QoS. Section 3 elaborates on the application of this ‘M/G/1′-based traffic control scheme in the scope of a DiffServ network. This application is particularly important given that it drives the real implementation described in this paper. Section 4 presents the main principles of traffic control in Linux routers and emphasizes on the embedded implementation of the traffic control scheme. Section 5 reports experimental results obtained on a DiffServ network testbed with Linux routers incorporating this scheme. Finally, Section 6 draws our main conclusions regarding the implementation of the traffic control laws, as well as their operation in a DiffServ network domain. Note that the presentation of our framework in Section 2 is done for completeness reasons. Readers interested in an in-depth coverage and experimental analysis and validation should consult [5], [6], [7], [8].
Section snippets
Overview of M/G/1 modeling results
In [5], [6], it was demonstrated that the M/G/1 queuing model is governed by equations that can be handy for traffic control of packet flows. The most important of these results are briefly reviewed. A multiplexer with service rate C and a buffer size large enough to be considered infinite is considered. It is assumed that this multiplexer serves a packet stream of M/G type, i.e. with exponentially distributed packet arrivals with mean arrival rate λ and generally distributed packet sizes with V
Definition of Per-Hop-Behaviours
As explained in Section 2.1, the asymptotic slope q of the queue length log-scale CPDF in a M/G/1 multiplexer can express a probabilistic quantitative QoS requirement for either loss or delay. Therefore, in a Differentiated Services architecture, a Per-Hop Behaviour for quantitative QoS can be defined by a specific value of q provided that the aggregate traffic stream offered at a multiplexer is of the ‘M/G’-type. This can be achieved by applying the shaping law of Section 2.2 on the individual
Linux routing and traffic control functionality
The Linux operating system has the capability to route IP packets between the network interfaces of a Linux system and features a fairly rich set of traffic control functionality. The Routing and Traffic Control subsystem within Linux is easily configurable as well as extendable and lends itself to the implementation of a QoS architecture [13].
Packet processing by the Linux kernel is highlighted in Fig. 6. It consists of three main blocks. The input classification and de-multiplexing block
Experimental results and evaluation
Using tc to configure the etbf and other Linux queuing disciplines, we may implement Per-Hop-Behaviors (PHB) targeting particular values of the QoS parameter q. The configuration of core routers is straightforward, since it is a matter of activating a scheduling discipline (subject to requirements set in Section 3.1). Configuring leaf routers is, however, much more interesting. Leaf routers apply the etbf shaping algorithm to micro and aggregate flows, so that a particular QoS point q and
Conclusions
Several traffic control results targeting IP-based QoS in a DiffServ context have been proposed in the research literature. Most of these efforts emphasize on the qualitative differentiation of packet flows into different traffic classes. In this work, we have emphasized on a traffic control framework that can deliver probabilistic, yet quantitative QoS in IP packet flows. The theoretical foundations of this framework have been thoroughly illustrated in our earlier works and are only briefly
Evangelos K. Vayias ([email protected]) received the Dipl.-Eng. and PhD degrees, both from the department of electrical and computer engineering, National Technical University of Athens, Greece, in 1996 and 2000, respectively. He is now a project consultant with Intracom SA, network professional services department, specializing in next-generation IP-based network technologies. Prior to this, he was a research associate with the Telecommunications Laboratory of the National Technical University
References (22)
Traffic theory and the internet
IEEE Communications Magazine
(2001)- et al.
Wide-area traffic: the failure of poisson modeling
IEEE/ACM Transactions on Networking
(1995) - et al.
Where mathematics meets the internet
Notices of the American Mathematical Society
(1998) Shaping of traffic streams through data spacing
IEEE Communications Letters
(1999)- et al.
Traffic engineering using a class of M/G/1 models
Journal of Network and Computer Applications
(1998) - et al.
CAC and traffic shaping for end-to-end performance control in ATM: the two-class paradigm
Computer Networks and ISDN Systems
(2000) - et al.
Packet-level traffic control for the internet: a framework and its experimental validation
Performance Evaluation
(2002) - et al.
Enforcing effective rates for packet-level QoS control in IP networks: theory and validation based on real traffic data
Telecommunication Systems Journal
(2004) - J.M. Pitts, J.A. Schormans, Introduction to IP and ATM Design and Performance, 2000, ISBN 0471 49187...
- et al.
Worst-case performance of a buffer with independent shaped arrival processes
IEEE Communications Letters
(2000)
Limits and approximations for the M/G/1 LIFO waiting-time distribution
Operations Research Letters
Cited by (7)
A linguistic decision support model for QoS priorities in networking
2012, Knowledge-Based SystemsCitation Excerpt :They are applicable at local, global, end-to-end networks [1,2] and often related to physical transmission media. Recently, different researches [27–29] have been conducted to describe the strengths and weaknesses related to each of these mechanisms and point out the benefits of QoS for organizations, allowing network administrators to take control of the use of these resources. Due to the large number of QoS techniques and their different application targets, the area of interest and application that we consider in this paper is focused on the traffic shaping technique that plans and prioritizes the network traffic by reducing or eliminating less important packages, leaving large bandwidth for important packages reach their destination as quickly as possible.
Time based traffic policing and shaping algorithms on campus network internet traffic
2017, Journal of Telecommunication, Electronic and Computer EngineeringModeling and performance analysis of QoS data
2016, Proceedings of SPIE - The International Society for Optical EngineeringA new adaptive throughput policy algorithm on campus ip-based network internet traffic
2015, Journal of Theoretical and Applied Information TechnologyAdaptive throughput policy algorithm with weibull traffic model for campus IP-based network
2014, Information Technology JournalTraffic management based on token bucket mechanism for WiMAX networks
2012, Cluster Computing
Evangelos K. Vayias ([email protected]) received the Dipl.-Eng. and PhD degrees, both from the department of electrical and computer engineering, National Technical University of Athens, Greece, in 1996 and 2000, respectively. He is now a project consultant with Intracom SA, network professional services department, specializing in next-generation IP-based network technologies. Prior to this, he was a research associate with the Telecommunications Laboratory of the National Technical University of Athens, where he worked in numerous research projects in the areas of resource management and performance evaluation of broadband networks, distributed object-oriented technologies for network management, interactive multimedia services, and grid computing. He also consulted for several public and private organizations in Greece. He is a member of the Technical Chamber of Greece and a Cisco Certified Internetwork Expert.
John K. Soldatos ([email protected]) was born in Athens, Greece in 1973. He obtained his Dipl-Eng. degree in 1996 and his PhD in 2000, both from the electrical and computer engineering department of the National Technical University of Athens (NTUA). He has had an active role in several EU co-funded research projects. Specifically, he has actively participated in EXPERT AC-094, WATT AC-235, IMPACT AC-324, Chameleon EP 20597, CATCH-2004 IST-1999-11103, and LION IST-19990-11387 projects, and is now with Athens Information Technology (AIT), involved in the CHIL-FP6-506909 project. He has also consulted in many ICT projects for leading Greek enterprises (INTRACOM S.A, PEGASUS S.A, IBM Hellas S.A, OTE S.A, TEMAGON S.A). Dr. Soldatos has extensively lectured in NTUA and AIT, and he has given numerous invited lectures. As a result of his activities he has co-authored more than 45 papers published in international journals and conference proceedings. Since March 2003 he has been with AIT, where he is currently an assistant professor. His current research interests are in broadband traffic control, network management, grid computing, and ubiquitous and autonomic computing.
George Kormentzas ([email protected]) is currently a lecturer at the University of the Aegean, department of information and communication systems engineering. He received the Diploma in electrical and computer engineering and the PhD in computer science, both from the National Technical University of Athens (NTUA), Greece, in 1995 and 2000, respectively. From 2000 to 2002, he was a research associate with the Institute of Informatics & Telecommunications of the Greek National Center for Scientific Research ‘Demokritos,’ working in a number of EU funded research projects in the area of high-speed wired and wireless networks. His research interests are in the fields of traffic analysis, network control, resource management, and quality of service in broadband networks. He has more than 40 publications in books, journals, and international conference proceedings in the above areas. He is a member of IEEE, IFIP, and of the Technical Chamber of Greece.