Elsevier

Computer Networks

Volume 40, Issue 1, September 2002, Pages 73-87
Computer Networks

Per-flow QoS support over a stateless Differentiated Services IP domain

https://doi.org/10.1016/S1389-1286(02)00267-0Get rights and content

Abstract

This paper consists of two parts. In the first part, we propose an admission control paradigm, called Gauge & Gate Reservation with Independent Probing (GRIP), devised to operate over a stateless Differentiated Services IP domain. GRIP admits a new flow upon the successful and timely delivery, through the domain, of probing packets independently generated by the end-points. Failed reception of probing packets is interpreted as congestion in the network. Our solution is fully distributed and scalable, as admission control decisions are taken at the edge nodes, and requires no coordination between routers, which are stateless and remain oblivious to individual flows. An interesting feature of the GRIP operation is its backward compatibility (at the expense of experienced performance) with existing routers. The performance of GRIP is related to the capability of routers to locally take decisions about the degree of congestion, and suitably block probing packets when congestion conditions are detected. In the second part of the paper we describe a specific GRIP implementation, characterized by the capability of providing strict Quality of Service guarantees, thanks to suitable assumptions made on the supported traffic and on the traffic control mechanisms, in a specific domain.

Introduction

It is known that the Integrated Services (IntServ) approach, while allowing hard Quality of Service (QoS) guarantees, suffers from scalability problems in the core network. To overcome this and other limits of IntServ, the Differentiated Services (DiffServ) paradigm has been proposed [15]. By leaving untouched the basic Internet principles, DiffServ provides supplementary tools to further move the problem of Internet traffic control up to the definition of suitable pricing/service level agreements (SLAs) between peers.

However, DiffServ lacks a standardized admission control scheme, and does not intrinsically solve the problem of controlling congestion in the Internet. Upon overload in a given service class, all flows in that class suffer a potentially harsh degradation of service. RFC2998 recognizes this problem and points out that “further refinement of the QoS architecture is required to integrate DiffServ network services into an end-to-end service delivery model with the associated task of resource reservation” [18]. It is thus suggested in RFC2990 to define an “admission control function which can determine whether to admit a service differentiated flow along the nominated network path” [17].

The aim of this paper is to propose a scalable per-flow admission control function, called Gauge & Gate Reservation with Independent Probing (GRIP), devised to operate over a stateless DiffServ domain. GRIP combines an admission control procedure based on end-point operation with localized measurements and decisions taken by core routers. Unlike reservation protocols such as RSVP, GRIP does not require routers to support an explicit signaling protocol. Instead, routers implicitly convey congestion information at the network edges by means of data forwarding plane operations, and specifically by dropping probing packets, transmitted by end-points upon call set-up, and labeled with a suitable Differentiated Services Code Point (DSCP). Since the role of detecting congestion is delegated to each router, different administrative domains can select the level of QoS provided, in full agreement with the DiffServ design guidelines. In fact, as shown in the following, the specific performance achievable by our admission control operation depends on the associated specific implementation and each administrative entity may arbitrarily tune the operational point of GRIP.

This paper is not meant to present a fully-fledged solution for the Internet nor a full protocol specification. Indeed, a number of detailed issues are not addressed in this paper, such as security, sensitivity with respect to variation of system parameters, interworking between different domains and policies, assumptions on traffic source behavior, relationship between the proposed solution, which operates at the network layer, and upper layers, implementation details, etc. Rather, our goal is to present ideas which may contribute to the on-going discussions in the international arena, and may be integrated into the IETF vision. In our opinion, GRIP could operate end-to-end over the whole Internet, but, more realistically, it could also be one of many possible different control mechanisms adopted to control traffic in different specific domains. In this vision, the degree of QoS support provided within each domain would depend on the tightness of control that the edge-to-edge mechanism will be capable to support. Schemes ranging from explicit per-flow resource reservation mechanisms (such as RSVP), down to aggregate forms of traffic control (e.g., via measurement based mechanisms, such as the one of GRIP) should be allowed to exist in different domains and interoperate [18]. The ultimate goal is that each domain should be placed in the ideal conditions of determining the suitable throughput/QoS support tradeoff within that domain. Note that, in this vision, GRIP could be an important building block to define and enforce a specific Per-Domain Behavior (PDB1). Also, the implicit signaling idea at the basis of GRIP could be deployed as a cross-domain signaling procedure. (The IETF Working Group “Next Steps in Signaling”, NSIS is perhaps the WG more involved in these issues.)

As for the organization, the paper is divided into two logical parts. The first part is composed by 2 Related work, 3 The GRIP paradigm, which are dedicated, respectively, to set the framework, by discussing previous and related work, and to describe the mode of operation of GRIP. In the second part of the paper, we describe a specific GRIP implementation (Section 4) within a particular domain, and we evaluate the relevant performance (Section 5). This domain is the one proposed in the framework of a R&D project sponsored by the European Union (project SUITED). The integrated communication infrastructure of this domain consists of multiple system components with mutually complementary characteristics: (i) a Ka-band regenerative satellite system; (ii) the General Packet Radio Service, representing the near-term version of the Universal Mobile Telecommunication System (UMTS) and the UMTS itself; (iii) a wireless local area network (based on 802.11); (iv) a subset of the currently available “best effort” Internet, upgraded with QoS support features. These four components provide a global “coverage” in a specific area and constitute an IP domain, managed by a single operator who intends to offer high quality services to the users who subscribe to the domain itself, by relying on suitable assumptions. The basic mechanisms of our proposal have been implemented in a test-bed (demonstrator) developed in the framework of the above R&D project, where GRIP is currently running in agreement with our theoretical findings. Finally, Section 6 is dedicated to our conclusions.

Section snippets

Related work

Recent literature (see [6] and therein contained references) has shown that an admission control function can be provided over stateless networks by means of the so-called End-point Admission Control (EAC). EAC builds upon the idea that admission control can be managed by pure end-to-end operation, involving only the source and destination hosts. At connection set-up, each sender–receiver pair starts a probing phase whose goal is to determine whether the considered connection can be admitted to

The GRIP paradigm

The GRIP mechanism, proposed in this paper, combines an admission control operation, driven by end-points, with run-time traffic measurements, performed within each router to detect congestion. GRIP does not employ a signaling protocol to set-up a new call, but relies on “implicit signaling”, i.e., a call is blocked if the end-to-end admission control operation is not completed. In the following two sections we describe, respectively, the GRIP end-nodes operation, and the GRIP router operation.

The Decision Criterion

In defining the Decision Criterion, our aim is to provide strict QoS guarantees under all operational conditions. To this purpose, information about QoS traffic flow characteristics is needed, to correctly reach ACCEPT/REJECT decisions. Thus, we assume that traffic sources are regulated at network edges by standard Dual Leaky Buckets (DLBs) (see [21] for additional details).

In any case, GRIP does not require signaling explicit information about the traffic mix composition, that is how many

Performance evaluation

A performance evaluation that includes the effect of the stack variable requires consideration of a dynamic scenario, accounting for flow departures and arrivals. In this section, we derive the utilization coefficient of a generic router's output link.

Assume that the duration of offered flows is exponentially distributed, with mean value 1/μ. To analyze high load conditions (which are the most critical ones), we assume a so-called continuous load model [12], in which a new flow is accepted to

Conclusions

GRIP is a reservation paradigm that allows an evolution from the actual best-effort Internet to a future QoS capable infrastructure. In conformance with DiffServ principles, GRIP does not rely on explicit signaling protocols to provide an admission control function. Such a function is achieved by requiring each router to be capable of distinguishing probing packets from information packets and properly enforcing a suitable dropping discipline. In addition, we have proposed procedures that allow

Acknowledgements

The authors sincerely thank the anonymous reviewers for many helpful suggestions. This work is co-supported by the European Union in the framework of the IST program (projects SUITED and WHYLESS.COM) and in the framework of the ITEA project POLLENS.

Giuseppe Bianchi received the Laurea degree in Electronic Engineering from Polytechnic of Milano, Italy, in 1990, and a specialization degree in Information Technology from Cefriel, Milano, in 1991. He spent 1992 as Visiting Researcher at the Washington University of St. Louis, MO, and 1997 as Visiting Professor at Columbia University, NY. He has been Assistant Professor at Polytechnic of Milano from 1993 to 1998. He is currently Associate Professor at the University of Palermo. His research

References (21)

  • W. Almesberger, T. Ferrari, J.Y. Le Boudec, SRP: A scalable resource reservation protocol for the Internet, IWQoS'98,...
  • G. Bianchi, N. Blefari-Melazzi, Admission control over assured forwarding PHBs: a way to provide service accuracy in a...
  • G. Bianchi, N. Blefari-Melazzi, M. Femminella, F. Pugini, Joint support of QoS and mobility in a stateless IP...
  • L. Breslau, S. Jamin, S. Schenker, Comments on the performance of measurement-based admission control algorithms, IEEE...
  • F. Borgonovo, A. Capone, L. Fratta, M. Marchese, C. Petrioli, PCP: a bandwidth guaranteed transport service for IP...
  • L. Breslau, E.W. Knightly, S. Schenker, I. Stoica, H. Zhang, Endpoint admission control: architectural issues and...
  • C. Cetinkaya, E. Knightly, Egress admission control, in: Proceedings of IEEE Infocom 2000, Tel-Aviv, Israel, March...
  • V. Elek, G. Karlsson, Admission control based on end-to-end measurements, in: Proceedings of IEEE Infocom 2000,...
  • A. Elwalid, D. Mitra, Traffic shaping at a network node: theory, optimum design, admission control, IEEE Infocom 97,...
  • A. Elwalid et al.

    A new approach for allocating buffers and bandwidth to heterogeneous, regulated traffic in an ATM node

    IEEE J. Select. Areas Commun.

    (1995)
There are more references available in the full text version of this article.

Cited by (22)

  • Efficient and stateless deployment of VoIP services

    2009, Computer Networks
    Citation Excerpt :

    Now, if we exploit the characterization provided in the previous section of this paper, we can remove the assumption that the DLB has to assure also that traffic does not underflow the sustainable rate specification with dummy packets, useful only for a generic, blind approach. In this way, we can increase the average system throughput in terms of carried voice calls, i.e. the average number of voice calls that can be admitted beyond the worst case derived in [21]. This is because we are now able to evaluate the actual rate of packets entering the system, as follows.

  • An end-to-end QoS framework with on-demand bandwidth reconfiguration

    2005, Computer Communications
    Citation Excerpt :

    In addition, since probing packets are actually transmitted through a network during the measurement period, the probing packets increase the traffic load and contribute to network congestion. In [6], a new QoS framework, named DiffServ-aware MPLS Traffic Engineering (DS_TE), is proposed. In this framework DiffServ is complemented by MPLS Traffic Engineering mechanisms that operate on an aggregate basis across all DiffServ traffic classes.

  • A game theoretic model and a distributed admission control scheme for quality of service routing

    2018, International Journal of Systems, Control and Communications
  • A mechanism of dynamic meter sharing on openflow switch

    2016, Jisuanji Xuebao/Chinese Journal of Computers
View all citing articles on Scopus

Giuseppe Bianchi received the Laurea degree in Electronic Engineering from Polytechnic of Milano, Italy, in 1990, and a specialization degree in Information Technology from Cefriel, Milano, in 1991. He spent 1992 as Visiting Researcher at the Washington University of St. Louis, MO, and 1997 as Visiting Professor at Columbia University, NY. He has been Assistant Professor at Polytechnic of Milano from 1993 to 1998. He is currently Associate Professor at the University of Palermo. His research interests include wireless access protocols and network architectures, QoS support in both wireless and wired IP networks, and performance evaluation.

Nicola Blefari-Melazzi received his “Laurea” degree in Electronic Engineering in 1989, magna cum laude, and earned the “Dottore di Ricerca” (PhD) degree in Information and Communication Engineering in 1994, both at the University of Roma La Sapienza, Italy. Since 1998 he is an Associate Professor at the University of Perugia. In January 2002 he won an Italian national call for Full Professorship. Dr. Blefari-Melazzi has been involved in various consulting activities and research projects, including standardization and performance evaluation work. His research projects have been funded or co-funded by the Italian Public Education Ministry, the Italian National Research Council, by industries and European organizations and programs. Dr. Blefari-Melazzi served as referee, TPC member, session chair and guest-editor to IEEE conferences and journals. His research interests focus on modelling and control of broadband integrated networks, multimedia traffic modelling, architectures and protocols for wireless LANs, satellite networks, queuing systems, mobile and personal communications, quality of service guarantees and real time services support in the Internet.

Mauro Femminella is a Ph.D. student of the Department of Information and Electronic Engineering of the University of Perugia since 1999. He received his “Laurea” degree in Electronic Engineering, magna cum laude with publication of his thesis, from the University of Perugia in 1999. His research interests focus on satellite networks, quality of service and mobility in IP networks. Currently, he is involved in IST (SUITED, WHYLESS.COM), MURST (RAMON) and ESA/CNIT projects. He is co-author of a number of papers and technical reports.

View full text