Dynamic grooming in IP over optical networks based on the overlay architecture

https://doi.org/10.1016/j.osn.2006.04.005Get rights and content

Abstract

This paper defines a formal framework for the definition of dynamic grooming policies in IP over optical networks. The formal framework is then specialized for the Overlay Architecture, where the control plane of the IP (Internet protocol) and optical levels are separated, and no information is shared between the two.

We define a family of grooming policies for the Overlay Architecture based on constraints on the number of hops and on the bandwidth sharing degree at the IP level, and we analyze the performances as a function of the grooming parameters in regular and irregular topologies.

Results are derived using realistic traffic models that depart from the circuit-like traffic traditionally used in grooming studies.

Introduction

The IP protocol and optical transmission techniques are going to play a fundamental role in the networking scenario of the coming years — if not decades. The presence of these two pillars is an easy bet, and it is a widespread idea that most of the intermediate network management layers will gradually disappear, leaving a scenario where IP packets are carried directly on high-speed wavelength division multiplexing (WDM)-based optical connections [1]. Optical packet switching and optical burst switching are elegant long-term solutions for the natural integration of optical transmission within a packet-based IP network; however, more traditional architectures, where packets are electronically switched by routers connected by circuit-switched optical connections, are going to dominate the commercial market for a long period. GMPLS (Generalized MPLS) is also going to play its part in this scenario, enabling the use of Traffic Engineering (TE) techniques at the IP level, substantially modifying the IP behavior from pure datagram to “virtually connected”. The Label Switched Paths (LSPs) behave as logical connections carrying elastic traffic that can adapt its rate to the network conditions. In this scenario, the interaction between routing and control of the circuit-switched optical network and that of the packet- (or label-) switched IP network is of the utmost importance for the end-to-end performance and the efficient exploitation of network resources. Traffic grooming lies at the heart of this problem.

Traffic grooming is the dynamic multiplexing capability aimed at optimizing the capacity utilization in transport systems by means of the combination of low-speed traffic streams onto high-speed (optical) channels. This problem is a variant of the well-known virtual topology design problem and has received a lot of attention in recent years (see [2] for a review).

In order to fix ideas, let us assume that, due to end-users traffic demands, a service provider dynamically generates LSP requests fsd between sources s and destinations d. The nodes are Label Switched Routers (LSRs) interconnected by Optical Cross Connects (OXCs) that can only switch entire lightpaths from one fiber to another. The action of mixing LSPs onto lightpaths and splitting them from lightpaths is a grooming action. The interaction between the IP and the optical levels, and the policy used in updating the virtual topology (e.g., aggressive or cautious use of optical resources) define the grooming policy within the network. In the rest of the paper, we use the terms (IP) router, LSR, and G-OXC (Grooming-capable OXC) interchangeably, although there might be some differences in specific implementations of the different devices. We do not make any specific assumption on the internal architecture of nodes (e.g., limitations on the number of ports) or on the implementation of the grooming fabric, since the focus of the paper is on the global architecture and not on specific implementations.

There are two main approaches to traffic grooming: static and dynamic. Static grooming refers to some network cost minimization when the traffic matrix is known in advance. The actions undertaken by the network when a request fsd arrives are completely pre-defined, and the network status does not influence decisions. Dynamic grooming is a routing problem in a multi-layer network architecture, since the objective is to find the “best” path to route traffic requests arriving dynamically at grooming nodes. In this case, equivalent fsd requests arriving in different times may be treated differently because of changed network conditions.

The static grooming problem has been studied intensively in the past, mainly on ring and mesh topologies, and it was proved to be NP-hard ([2] provides a review of many proposed heuristics for its solution). However, the rapid increase in Internet traffic and the intrinsic difficulty in forecasting its behavior on both small and large scales [3] make static grooming non-optimal. Even very fast network reconfigurations based on on-line measurements (say 30–60 min) are useless, because the measured traffic is not a good prediction of the future traffic.

In the emerging “IP over Optical” (IPO) network scenario, dynamic grooming becomes a fundamental issue for exploiting network resources. Since the traffic matrix is not known in advance, it is not possible to perform an optimal routing, but only heuristic algorithms can be used to maximize network performance. In order to route sub-wavelength-level IP traffic over a wavelength-routed optical network, the interaction between the two routing layers (IP and optical) must be considered, and the nature of the traffic (best-effort and elastic) loading the network must be taken into account through suitable models.

Before going into detail about recent works on dynamic traffic grooming, let us clarify the networking context in which these proposals can be used. RFC 3717 defines the framework for IPO architectures [4]. In IPO networks, IP routers are attached to an optical core network and connected to other peers via dynamically established lightpaths. The core network is composed of OXCs, thus the optical layer is incapable of processing directly packets, bursts or any sub-wavelength capacity: it provides point-to-point connectivity between IP routers through fixed-bandwidth lightpaths. The set of all the lightpaths established over the physical topology is known as virtual topology.

The virtual topology is used by the IP routers to forward the traffic to the destination. The data plane is thus realized as an overlay network of lightpaths. Considering the control plane, different architectures can be envisioned according to the assumptions about the amount of information exchanged between the IP and optical layers. RFC 3717 states three interconnection models: overlay, augmented and peer. In the peer model, IP routers and OXCs are considered to be peer network elements, thus the topology and other network information are completely shared by a unified control plane. In the overlay model, each layer performs its own routing functions, since no information is exchanged between them. An intermediate architecture between these two is the augmented model, where some aggregated information from one routing instance is passed to the other, in general only from the optical to the IP layer.

The peer and the augmented models are appealing because sharing the knowledge base between the two layers allows the running of an integrated routing function using, for instance, an auxiliary graph, as done in [5]. The integrated management enables better usage of the overall network resources. However, both models seem unfeasible in the near term due to the tight integration between the two levels and scalability issues regarding the amount of exchanged information. The overlay model is instead technically feasible, since it only requires the definition of a clear interface between the IP and optical levels, as well as dynamic lightpath capabilities in the optical level, which are being experimented on in laboratories and research projects [6]. Surprisingly, even if the peer and augmented models do not seem realizable in the near future, most of the dynamic grooming algorithms proposed in the literature implicitly consider such models [5], [7], [8], [9], [10], since the objective is to maximize the utilization of the multi-layer network architecture and these models better support the purpose.

In the overlay model, the absence of information exchange leads inevitably to sub-optimal network utilization [10]. Only a few simple dynamic grooming algorithms based on this model have been proposed so far [11], [12], [13], and the interaction between different routing strategies adopted in the IP and optical levels were never assessed. In [11], constraint-based routing is assumed at the IP level, while [12], [13] take into account also possible limitations in the number of ports. These papers explore the two extreme policies of always privileging the optical level exploitation, or the other way around. In [10], finally, the authors analyze both the augmented and the overlay architecture, assessing the superiority of the first one in terms of performance, and proposing new algorithms that improve performance in the overlay model with respect to [11] and in the augmented model with respect to a grooming algorithm in peer models proposed in [7]. It must be noted, however, that the algorithm proposed for the overlay model, named MLH_OVLY, requires the exploration of the optical connectivity between all the possible LSR pairs in the network to decide whether to accept or block a new connection, which hardly seems feasible in large networks, at least if no information from the optical level is available at the IP level, as in the overlay model.

It must be noted that none of the works on grooming in IPO networks adopted a realistic traffic model. The traffic loading the network is always composed of CBR (Constant Bit Rate) connections characterized by the bit rate and duration only. Any realistic evaluation of algorithms to be deployed within the Internet should instead capture at least some of the basic characteristics of Internet traffic. From the routing point of view, the most important features of Internet applications are the capacity to adapt the rate to changing network conditions (elasticity) and the need to transfer a given amount of data. The holding time of a flow becomes a consequence of the network conditions and not a property of the flow itself. In a previous contribution [14], we discussed in detail the impact of realistic traffic models on dynamic grooming, showing the inherent interaction between the IP and the optical layers and its effect on the overall performance of the network. In this paper, we use only the realistic model, which we called data-based in [14]. Traffic flows in this model share the resources on a virtual topology path following the max–min fairness criterion [15], thus mimicking the ideal behavior of a bundle of TCP connections.

In this paper, we study the impact of different grooming policies in IPO networks based on the overlay interconnection model with realistic traffic. The main contributions of this paper are the following:

  • In Section 2, a formal definition of dynamic grooming in a general interconnection model is given and then specialized to the case of the overlay model.

  • The existing dynamic grooming proposals for overlay networks are assessed as a subset of a more general and parametric grooming policy (Section 3).

  • Performance and trade-offs of the different policies are discussed and explained in both regular (ring and mesh-torus) and irregular topologies, also discussing pros and cons of dynamic versus static grooming and the impact of adopting TE techniques in the IP or optical level.

  • Unfairness issues inherent in dynamic grooming and arising from different physical distances between flow end-points are discussed, and hints on the problem solution are given.

Section snippets

Problem formulation

To the best of our knowledge, a formal description of dynamic grooming has never been defined in the literature. Here we present a formal framework based on graph theory for the definition of dynamic grooming policies.

Grooming policies

In the scenario depicted above, independently of the definition of Λ and Ω, the set of rules Δ defines how and when to invoke Λ. Although many criteria can be envisaged, based on any network measure available, the simplest rule is based on the number of IP hops between s and d. In other words, Δ defines as a rule the invocation of Λ only if there is no path πtν available that does not exceed K hops from s to d. We call this policy Hop Constrained Grooming, HC(). Additionally, HC can include

Results and discussion

A theoretical performance analysis is not feasible due to the complexity of the system, thus we resort to simulations for the performance evaluation, exploring both regular and irregular topologies.

After describing the simulation tool that we use in Section 4.1, defining the relevant performance parameters in Section 4.2, and introducing the networking scenarios that we consider in Section 4.3, we devote the rest of the section to results. Results are organized in three sections, each one

Conclusions

This paper has introduced a formal description of dynamic grooming policies, clearly defining the limits between grooming in overlay architectures and grooming in peer or augmented architectures, where there is total or partial integration of the optical and IP control planes.

A family of grooming policies, based on constraints on the number of hops and bandwidth available at the virtual topology level, has been defined and analyzed in different regular and irregular topologies, discussing

References (26)

  • A. Jajszczyk

    Optical networks — the electro-optic reality

    Optical Switching and Networking

    (2005)
  • C. Casetti et al.

    A new class of QoS routing strategies based on network graph reduction

    Computer Networks

    (2003)
  • R. Dutta et al.

    Traffic grooming in WDM networks: past and future

    IEEE Network Magazine

    (2002)
  • M.E. Crovella et al.

    Self-similarity in world wide web traffic: Evidence and possible causes

    IEEE/ACM Transactions on Networking

    (1997)
  • B. Rajagopalan, J. Luciani, D. Awduche, IP over optical networks: A framework, IETF RFC 3717,...
  • H. Zhu et al.

    A novel generic graph model for traffic grooming in heterogeneous WDM mesh networks

    IEEE/ACM Transactions on Networking

    (2003)
  • C. Cavazzoni et al.

    The IP/MPLS over ASON/GMPLS test bed of the IST project LION

    IEEE/OSA Journal of Lightwave Technology

    (2003)
  • M. Kodialam, T.V. Lakshman, Integrated Dynamic IP and Wavelength Routing in IP over WDM Networks, in: Proc. of INFOCOM...
  • R. Srinivasan et al.

    Dynamic routing in WDM grooming networks

    Photonic Network Communications

    (2003)
  • W. Yao, B. Ramamurthy, Dynamic traffic grooming using fixed-alternate routing in WDM mesh optical networks, in:...
  • S. Koo, G. Sahin, S. Subramaniam, Dynamic LSP provisioning in overlay, augmented, and peer architectures for IP/MPLS...
  • C. Assi et al.

    Integrated routing algorithms for provisioning sub-wavelength connections in IP-Over-WDM networks

    Photonic Network Communications

    (2002)
  • X. Niu et al.

    Connection establishment of label switched paths in IP/MPLS over optical networks

    Photonic Network Communications

    (2003)
  • Cited by (8)

    • Ant Colony Optimization-based distributed multilayer routing and restoration in IP/MPLS over optical networks

      2021, Computer Networks
      Citation Excerpt :

      Therefore, we have two different and isolated ant colonies with their respective data structures: one for the optical layer and another for the IP layer. The interaction between the optical and IP layers defines the multilayer routing policy of the network, which can also be called the grooming policy of the network[28]. In the rest of this paper, we adopt the formal framework definitions for multilayer routing policies that were presented in[28].

    • Routing and restoration in IP/MPLS over optical networks by means of ant colony optimization

      2019, Proceedings - IEEE Global Communications Conference, GLOBECOM
    • Planning data transfers in grids: A multi-service queueing approach

      2012, Concurrency and Computation: Practice and Experience
    • Dynamic grooming, is it really dynamic?

      2011, International Congress on Ultra Modern Telecommunications and Control Systems and Workshops
    • Augmented grooming in IP over optical networks with elastic traffic

      2008, International Symposium on Performance Evaluation of Computer and Telecommunication Systems 2008, SPECTS 2008, Part of the 2008 Summer Simulation Multiconference, SummerSim 2008
    View all citing articles on Scopus

    This work was supported by the Italian Ministry of Education and Research (MIUR) through the FIRB GRID.IT project, and by the Hungarian Italian Intergovernmental S&T Cooperation Programme for 2004–2007, project reference number I-17/03. The project is developed under the European Union E-NEXT Network of Excellence umbrella.

    View full text