A distributed impairment aware QoS framework for all-optical networks

https://doi.org/10.1016/j.osn.2010.05.003Get rights and content

Abstract

Different physical impairments can occur in optical transmission systems. Impairments such as fiber nonlinear effects are dependent on network state and vary with traffic and topology changes. In all-optical networks, impairments can accumulate along a lightpath and cause significant signal degradations. It is important to consider these impairments and the corresponding degradations in the routing algorithm design to provide quality of service (QoS). We propose a distributed QoS framework to achieve traffic engineering and QoS assurance for all-optical networks. Analytical models and new algorithms are designed in the framework to predict lightpath signal quality in dynamic network environments. The framework has also been used to compare performance of several routing and wavelength assignment algorithms with impairments taken into account.

Introduction

The long range view of optical networks includes end-to-end all-optical paths that are dynamically established in response to user needs. New routing and wavelength assignment (RWA) strategies will be required to enable these goals. Most RWA algorithms that have been described in the literature do not take into account physical impairments that can occur either in network components (e.g. optical switches, multiplexers, demultiplexers, optical amplifiers, etc.) or in optical fibers. In currently deployed commercial optical networks, optical signals may be regenerated at each repeating site. The influences of physical impairments are minimized by designs that incorporate dispersion compensation fiber in the regeneration process. As optical networking evolves towards all-optical networks, where signals are not regenerated, consideration of impairments which may affect quality of service (QoS) in RWA algorithm design becomes necessary and important. In an all-optical network, optical signals are transmitted from source to destination totally in the optical domain. Physical impairments can accumulate along a lightpath and cause significant signal degradations. It is possible that the signal quality at the destination node may be so poor that the bit error rate (BER) reaches an unacceptably high value.

In an all-optical network, a lightpath (a path plus a wavelength channel for data transmission) is set up through a cascade of independent fiber links. These links usually have different physical characteristics and thus different influences on signal quality. Moreover, the physical impairments on a lightpath also change with network state. For example, the influences of the fiber nonlinear effects XPM (Cross-Phase Modulation) and FWM (Four-Wave Mixing) are both dependent on the number of co-propagating channels and the spacing between wavelengths of co-propagating channels and the channel wavelength under consideration. The dynamic assignment of lightpaths in response to end user service request results in time-dependence in the network state and in the extent of impairments. To provide quality of service in an all-optical network, RWA algorithms need to intelligently select lightpaths satisfying user QoS requirements under the constraints of network physical characteristics and network state.

To decrease complexity, state of the art impairment aware RWA algorithms usually have three steps: route computation, lightpath signal quality predication, and wavelength assignment. The routing algorithm is used to find a path (i.e. route) from source to destination, while the wavelength assignment algorithm is used to select a free QoS satisfying wavelength on the chosen path. A number of solutions have been proposed in the literature to extend classic RWA algorithms and take physical impairments into account.

We can classify the routing algorithms proposed in the literature into two categories according to whether physical impairments are considered in route computation or not. The first category incorporates the physical impairments into a link cost function to compute a shortest path from source to destination. In [1], the cost function was suggested to be defined as 1Q or 1Q2 where Q is the link Q factor. In [2], the cost of a wavelength j on link i is defined as W(i,j)=δXT2(i,j)+δNL2(i,j)+δASE2(i,j)+αi where δXT2, δNL2, δASE2 are the crosstalk noise, nonlinear noise, and ASE noise respectively. The stabilizing factor αi is used to turn off impairment consideration when the influence of impairments is negligible (In [2], the solution first chooses a free wavelength and then decides the shortest path.). In [3], [4], [5], [6], the link cost function is defined to be impairments’ Q-factor penalties. In [7], the link cost function considers both link usage and FWM noise power. An advantage of this category is the comparatively high efficiency of finding a lightpath which can possibly satisfy QoS requirements. But the used link cost function inherently requires the additive property of the described physical impairments. The fact that a physical impairment has the additive property means that the noises generated by the impairment in different network components, fibers, or links are independent, and the total noise power at the end of a lightpath is the mathematical summation of the powers of constituent noise components. The linear impairments (e.g. Chromatic Dispersion CD, Polarization-Mode Dispersion PMD, and crosstalk) usually have this property, but the nonlinear impairments (e.g. Self-Phase Modulation SPM, XPM, and FWM) do not. As an example, the link cost function in [7] is defined as C=αF×Wj=1j=Fi=1i=Wwij+βP¯FWMPth where F and W are the number of fibers per link and number of wavelengths per fiber respectively, wij represents the usage of a wavelength (set to 1 for idle, and 0 for occupied), P¯FWM is the average FWM noise power in idle wavelengths, and Pth is the FWM threshold power level. In (2), the first term on the right hand side (RHS) considers the wavelength usage on a link and the second term considers the FWM noise power, with the two adjustable parameters α and β representing the comparative importance of the two factors. However, in our view, it is not appropriate to consider these two inherently different factors together by a simple linear combination (it may be difficult to tune α and β). Besides, the FWM noise power does not have the additive property, as will be illustrated in this paper. The lightpath FWM analytical model applied in [7] is an oversimplification.

The algorithm proposed in [8] assumes a centralized network management system to perform call admission control, routing and wavelength assignment, call establishment, and termination. For a given connection request, the algorithm pre-computes a set of candidate lightpaths considering the lightpath QoS and the influence on existing lightpaths, then a final lightpath is selected according to different principles. Among them, the protection threshold is used to improve algorithm fairness (i.e. reserve resource for lightpaths with large numbers of hops); the highest Q factor is used to decrease the average BER in the network; the max–min Q factor is used to decrease the worst case network BER. The time complexity seems to be high, especially with respect to the number of wavelengths. Besides, the proposed solution also inherits the drawbacks of centralized routing schemes (e.g. The centralized management system has to know the physical characteristics of the whole network and keep track of the network status. The management system is difficult to scale and itself can be a performance bottleneck.). Due to these reasons, in this paper we will focus on the distributed routing schemes.

The second category ignores physical impairments in the route computation step in order to improve computation speed. The used link metric can be physical distance [9], [10], link congestion [11], their combination [12], etc. For example, the routing algorithm in [9] used a layered network graph, with each layer corresponding to one wavelength. The link cost function is defined to be the physical distance, and is set to if one wavelength is already used. A disadvantage of this category is its lower efficiency of finding a QoS satisfying lightpath when compared to the first category.

With the link cost function being defined, both categories invoke Dijkstra’s algorithm, the Bellman–Ford algorithm, or the A*algorithm [13] to compute a shortest path from source to destination. To decrease connection blocking probability, sometimes k-shortest paths are tried [3], [6], [9], [11]. We note that for link cost functions which ignore network status, the application of shortest path tends to cause congested links and increase blocking probability (this is especially a problem for link cost functions which only consider physical distance).

In impairment aware RWA algorithms, the signal quality of a candidate lightpath is predicted before wavelength assignment. Either close-form analytical models or heuristic models based on simulations or experiments [11], [14] can be applied to do the computation. In [7], [9], [12], [15], [16], the analytical models for computing signal degradations caused by Amplified Spontaneous Emission (ASE), thermal noise, shot noise, PMD, Distributed Raman Amplifier (DRA) noise, crosstalk, XPM, and FWM were provided. Considering the effect of different impairments after photo-detection, the end-to-end signal quality of a lightpath can be measured by the Q factor, defined as [10]Q=10log(RPs,mσ1+σ0) where R is the photo-detector responsivity, Ps,m is the peak power of a signal channel, and σ0=σth2+σshot2+σasease2σ1=σth2+σshot2+σasease2+σsig-ase2+σxpm2+σfwm2. In (3), the photo-current at the space state (when sending bit ‘0’) is assumed to be 0. In (4), (5), σth2 is the power of the thermal noise, σshot2 is the power of the shot noise, σsig-ase2 is the power of the signal beating ASE noise, σasease2 is the power of the ASE beating ASE noise, σxpm2 is the power of the noise generated by XPM, and σfwm2 is the power of the noise generated by FWM. The analytical method has the disadvantage of being computation intensive, while the heuristic method has the disadvantage of being inflexible because the derived heuristic equations are usually based on some specific link traffic (e.g. fully loaded). But in an optical network the traffic can dynamically change. We also note that to our knowledge the signal quality prediction models proposed in the literature almost all use the independency assumption (i.e. different types of impairments are independent and impairments on different links are also independent).

After determining the signal quality of the candidate lightpath, the extension of the classic wavelength assignment algorithms is straightforward. In the First-Fit algorithm [7], [17], the QoS satisfying wavelength with the lowest index is selected. In the Random-Pick algorithm, a QoS satisfying wavelength is randomly selected from the available wavelengths [2], [15]. In the Best-Fit algorithm, the wavelength with the best signal quality is selected [11], [12], [18]. Among these algorithms, the First-Fit has the effect of compacting the used and unused wavelengths (in terms of wavelength index) so large hop-number connection requests have a better chance of being accepted [19]. In the Random-Pick algorithm, since the used wavelengths are randomly spread in the frequency domain, the crosstalk effects might be limited [15]. For a complete review of impairment aware RWA algorithms, the interested reader can refer to [20].

In the design of distributed impairment aware RWA algorithms, the concerns of scalability and flexibility are of critical importance. The signaling overhead should be carefully controlled by considering the large amount of parameters used to characterize different physical impairments. The algorithms also need to be flexible enough such that new impairments can be easily incorporated. The important impairments of optical networks can change as optical transmission technology evolves. For example, the influence of PMD is negligible for transmission rates less than and equal to 10 Gbps, but needs to be considered when the transmission rate is 40 Gbps or higher.

In this paper, we propose a new distributed framework for QoS routing in all-optical networks. As in [9], [10], the framework also has two steps: route computation and lightpath probing to verify signal quality and reserve resources for a connection. We have considered physical impairments including attenuation, dispersion, ASE noise, and the fiber nonlinear effects XPM and FWM. Major physical impairments (e.g. ASE noise) are taken into account at the route computation step to improve algorithm efficiency compared to [9], [10]. New algorithms have also been designed to predict the noise powers caused by fiber nonlinear effects in complex network environments at the lightpath probing step. The proposed framework is scalable, flexible, and can be easily extended to consider new physical impairments.

The paper is organized as follows. In Section 2, the network model of all-optical networks is explained. In Section 3, the analytical models and algorithms used to assess physical impairments (including ASE, XPM and FWM) on lightpaths are investigated in detail. These models and algorithms will be used to predict signal quality in dynamic all-optical networks and be incorporated into the proposed distributed framework. Then in Section 4, we discuss the design of the proposed distributed framework. In Section 5, we show by numerical simulations that the application of signal quality information provided by analytical models can be used to improve the routing algorithm performance. The discussion and conclusions are given in Section 6.

Section snippets

Network model

The network under study is an all-optical network, which consists of multiple optical switch nodes connected by optical links in an arbitrary topology, as illustrated in Fig. 1.

In Fig. 1, the network is logically separated into the data plane and the control plane according to their different functionalities. The major functionality of the data plane is to transmit optical pulse streams from source to destination under the control of the control plane, and consists of transmission media

Analytical models for the assessment of physical impairments

Both the source node and the destination node of a connection request will use analytical models to estimate the signal quality of a lightpath. The source node only considers ASE noise when choosing candidate paths, because it is the dominant impairment in current-generation and near-future next-generation optical networks. The destination node will consider additional physical impairments when estimating lightpath signal quality to provide QoS assurance. The signal quality of a lightpath is

The design of a distributed QoS framework (DQF)

The proposed QoS framework has three parts: the Physically Aware Routing algorithm (PAR), the Physically Aware Backward Reservation protocol (PABR), and a Quality-aware Fit–Fit wavelength assignment algorithm (QFF). The PAR algorithm takes major physical impairments into account when computing paths for a connection request. This can improve the algorithm efficiency compared to [9], [10]. We assume a separate control network is available for transmitting signaling messages.

The route computation

Numerical simulations

In the current implementation of PAR, we have used the link cost function of LORA [34]cost(l)=βusage(l) where usage(l) is the number of used wavelengths on the link l and β is a parameter that needs to be tuned according to network topology and traffic load.

We use OPNETTM simulations to study the performance of the proposed QoS framework. Because OPNETTM is a network layer simulator, the physical characteristics of a simulated network are provided in an XML configuration file. When a simulation

Discussion and conclusion

We have developed new algorithms to assess the influence of ASE noise and fiber nonlinear effects on lightpaths in all-optical networks. These algorithms are then incorporated into the proposed distributed QoS framework to provide on-line estimation of lightpath signal quality.

The distributed QoS framework includes three parts. In PAR, major physical impairments (e.g. ASE noise) are taken into account to choose the candidate paths in the route computation step. The optimization goal of PAR is

References (36)

  • S. Azodolmolky et al.

    A survey on physical layer impairments aware routing and wavelength assignment algorithms in optical networks

    Computer Networks

    (2009)
  • S. Zsigmond, G. Németh, T. Cinkler, Mutual impact of physical impairments and grooming in multilayer networks, in:...
  • J. He, M. Brandt-Pearce, Y. Pointurier, S. Subramaniam, QoT-aware routing in impairment-constrained optical networks,...
  • S. Pachnicke et al.

    Assessment of a constraint based routing algorithm for translucent 10 Gbits/s DWDM networks considering fiber nonlinearities

    Journal of Optical Networking

    (2008)
  • T. Deng, S. Subramaniam, Adaptive QoS routing in dynamic wavelengthrouted optical networks, in: 2nd International...
  • P. Kulkarni, A. Tzanakaki, C. Mas Machuca, I. Tomkos, Benefits of Qfactor based routing in WDM metro networks, in:...
  • G. Markidis et al.

    Impairment-constraint-based routing in ultralong-haul optical networks with 2R regeneration

    IEEE Photonics Technology Letters

    (2007)
  • A. Marsden, A. Maruta, K. Kitayama, Routing and wavelength assignment encompassing FWM in WDM lightpath networks, in:...
  • Y. Pointurier et al.

    Cross-layer routing and wavelength assignment in all-optical networks

    IEEE JSAC

    (2008)
  • H. Yurong et al.

    Connection provisioning with transmission impairment consideration in optical WDM networks with high-speed channels

    Journal of Lightwave Technology

    (2005)
  • C. Politi, V. Anagnostopoulos, Physical layer impairment aware routing algorithms based on analytically calculated...
  • R. Cardillo, V. Curri, M. Mellia, Considering transmission impairments in wavelength routed networks, in: Proc. of ONDM...
  • Y. Zhai, Y. Pointurier, S. Subramaniam, M. Brandt-Pearce, QoS-aware RWA algorithms for path-protected DWDM Networks,...
  • A. Morea et al.

    QoT function and A* routing an optimized combination for connection search in translucent networks

    Journal of Optical Networking

    (2008)
  • A. Marsden, A. Maruta, K. Kitayama, Routing and wavelength assignment encompassing FWM in WDM lightpath networks, in:...
  • B. Ramamurthy et al.

    Impact of transmission impairments on the teletraffic performance of wavelengthrouted optical networks

    Journal of Lightwave Technology

    (1999)
  • G. Markidis, S. Sygletos, A. Tzanakaki, I. Tomkos, Impairment aware based routing and wavelength assignment in...
  • H.A. Pereira, D.A.R. Chaves, C.J.A. Bastos-Filho, J.F. Martins-Filho, Impact of physical layer impairments in...
  • Cited by (7)

    View all citing articles on Scopus

    This work was supported by the NSF under the NeTS Grant 0624874.

    1

    Present address: Epic Company, Madison, WI, United States.

    View full text