QoS aware green routing and wavelength assignment in core WDM networks

https://doi.org/10.1016/j.jnca.2015.01.007Get rights and content

Abstract

In order to reduce the power consumption of data centers, green routing has been touched upon and noticed by research community. From optical WDM network layer point of view, energy aware routing and wavelength assignment refers to establish connections on a subset of network topology while maintaining connectivity and QoS. Idle network elements can be put to sleep or hibernate mode in order to save energy. In this paper, a multi-objective ILP formulation for RWA problem is developed. Then, five novel energy aware routing approaches are proposed in which we aim at simultaneously minimizing energy consumption in WDM network as well as the number of transitions from sleep to active mode while guaranteeing a QoS constraint. By applying the proposed algorithms, noticeable and significant reduction in total number of state transitions from sleep to active mode is experienced. Besides, energy efficiency and blocking rate as two main QoS parameters can be flexibly traded off by adjusting proper values for input parameters of the proposed algorithms.

Introduction

Number of Internet subscribers has been growing rapidly day to day and Optical WDM networks have been identified as the only solution in order to provide a large amount of bandwidth for this huge traffic demands. Data centers around the world as main sources of data are connected via high speed optical links that constitute optical WDM core networks. Worldwide data centers alone consume about 26 GW of electrical power corresponding to about 1.4% of worldwide energy consumption (Katz, 2009). This amount of power consumption is growing approximately 12% each year due to the increase in Internet expansion. It is predicted that in the next coming years, the large companies operating data centers will have to spend more money on energy than on equipments. Based on the Environmental Protection Agency (EPA) report to congress, data centers in United States have consumed 1.5% of total U.S. electricity with the cost of about 4.5 billion dollars (United States Environmental Protection Agency, 2007). This amount of energy consumption of ICT, especially data centers, has raised attention in research and manufacturing community in order to accomplish a more energy efficient structure to save energy and money (Zhang and Ansari, 2013).

Design of energy efficient data centers, or simply GREEN data centers, can be realized based on two main approaches; using the green elements or greening the operation process of data centers performance. From the engineering point of view, greening a data center means to reduce the energy consumption while keeping the same level of performance. In most cases, various parts of data centers especially network equipments are designed to operate at peak load which results in extra unnecessary power consumption when the traffic is low. Therefore, in order to achieve a more energy efficient structure, it is indispensable to make the performance of equipments proportional to the traffic (Bianzino et al., 2012). This approach can be realized from the network layer point of view. We begin in the following, with a brief background review, in order to provide some insights into the concepts and issues that we develop in more detail in the sections that follow.

Routing and wavelength assignment (RWA) is known as the common problem for optical networks for which various number of solutions have been proposed and evaluated (Zang et al., 2000, Rahbar, 2012). Energy awareness is the issue which has been considered recently in routing and traffic engineering. Energy aware routing aims at aggregating traffic flow over a subset of network elements while other elements are switched off or put to sleep mode. The key idea behind the energy aware routing approaches is to switch off or put to sleep maximum number of network devices within low traffic periods, while maintaining connectivity and quality of service (QoS).

An RWA problem can be decoupled into two separate sub-problems namely routing sub-problem and wavelength assignment sub-problem. The wavelength assignment sub-problem can be solved using one of common approaches namely First-Fit, Most-Used, Random-Pick and Least-Used (Zang et al., 2000). The First-Fit algorithm, known as the most common approach, checks for idle wavelengths on all the links of a route, and then selects the wavelength with the lowest index number as the assigned wavelength. The routing sub-problem, generally, is solved using shortest path (or minimum weight) algorithm on a given weighted graph (see Cormen et al. (2001) for example). Various link weight/cost assignment procedures are applied in order to achieve particular objectives. If the link weight/costs are set to unity, then applying the shortest path algorithm will result in routes with minimum number of hops. The Minimum-Hop routing scheme followed by the well known First-Fit wavelength assignment is referred to minimum hop routing and wavelength assignment (MH-RWA) algorithm from now on. When the links weights are set to link length, the approach is called shortest path RWA (denoted by SP-RWA in this paper). Both MH-RWA and SP-RWA schemes obviously lack energy efficiency because of the fact that almost all of the network elements are turned on and the network energy consumption is not proportional to traffic load.

It should be noted that WDM has some drawbacks that is worth investigating from various perspectives. Current WDM systems require full allocation of wavelength capacity to a light-path, and hence, this issue leads to rigid large granularity for such systems. In order to provide finer granularities and enhance the scalability and spectral efficiency of WDM systems, spectrum sliced elastic optical networks (i.e., SLICE) have been presented (Jinno et al., 2009). Elastic optical networks can provide efficient accommodation of multiple data rates which result in supporting various services. Unique features of SLICE have recently made it a hot topic in research community and significant research efforts have focused on optimizing the performance of such networks from network level and device level perspectives. However, enabling technologies to deploy elastic optical networks include bandwidth variable transponders at the network edge and bandwidth variable wavelength cross connects (WXC) in the network core. From networking point of view, flexible routing and spectrum assignment (RSA) algorithms need to be carefully designed instead of existing RWA techniques (Yang et al., 2011).

This study aims at providing an energy efficient approach for existing WDM systems, where energy efficiency is still of high importance for such networks. The network model and assumptions are based on fixed bandwidth optical cross connects and network elements are considered as in WDM systems.

The work in Aksanli et al. (2012) has considered the network of data centers, where the availability of green energy resources in each data center has been taken into account. A green energy aware routing (GEAR) algorithm has been proposed with the objective of minimizing the total brown energy consumption. Note that brown energy resources are those resources that produce carbon dioxide. It means that the selected routes for each connection request are not necessarily the shortest length routes, nor the least energy consuming ones, but they are the routes with the lowest brown energy consumption and most of their energy is provided by green energy resources. Wu et al. (2009) and Wu (2009) have proposed several energy aware routing heuristics for static traffic. The key idea is to route the requests from already activated links. The proposed most used (MU-RWA) algorithm begins by assigning link energy consumption as the link costs and tries to route the requests from the least cost paths. While routes are established, the links are switched on and the costs of activated links are set to zero so that for the next coming requests, these links are selected with high probability in order to avoid waking up the sleeping links. This approach yields to low energy consumption but there exists two main drawbacks. First, since the algorithm is trying to pack the requests on some particular links, the blocking rate is extremely high resulting in QoS degradation. Second, some links with high energy consumption would never be activated in MU-RWA because of its cost assignment procedure.

The research in Wiatr et al. (2010) has considered blocking rate as the QoS parameter which is being traded off with energy efficiency. In order to avoid creation of bottleneck links, weighted power aware routing and wavelength assignment algorithm (W-PA-RWA) has been proposed in which the link cost assignment is similar to MU-RWA with a slight difference. The links which are already activated are assigned less costs compared to the ones which are sleeping. By choosing proper algorithm parameters, one can achieve energy efficiency and QoS provisioning. Same idea has been adopted in Jirattigalachote et al. (2011), Cavdar (2011), and Muhammad et al. (2010). Energy efficient path protection is an issue which is taken into account in Jirattigalachote et al. (2011) and Muhammad et al. (2010). It is worth mentioning that green grooming has recently become the subject of many researches in which light-path establishment in IP-over-WDM networks is tackled (Schöndienst and Vokkarane, 2014, Naas et al., 2014, Guo et al., 2013). In Buzzi et al. (2009) and Fisher et al. (2010), authors have formulated the problem in the form of integer linear programming which falls in subclass of capacitated multi commodity cost flow problem (CMCF) (i.e., the problem in which multiple commodities have to be routed over a graph with some capacity constraints). The CMCF problem is known as NP-hard thus some heuristics have been proposed. Moreover, Chiaraviglio et al. (2013) have evaluated the energy efficiency obtained by applying either energy proportional or sleep mode approaches to optical core networks. The latter work considers a general routing problem without focusing on the RWA issue and the related constraints. Besides, a smart link wake up policy is not presented and the work solely demonstrates how much energy can be saved using sleep mode schemes. It is proved that each network element׳s energy consumption consists of a constant term which is added to a load proportional term. When the load proportional term is greater than the constant term, then sleep mode schemes will yield better results compared to energy proportional approaches, in terms of energy efficiency. Note that the state of the art optical network elements energy consumption profile is not completely proportional to the corresponding load, meaning that sleep mode schemes are preferred to energy proportional approaches.

The key idea in all energy-aware routing or simply EA-RWA algorithms is to switch off idle links and nodes in order to save energy in low traffic periods. However, the links which are switched off or put to sleep are ready to be activated when necessary. As a matter of fact, we need some time in transition from the sleep mode to the active mode. This time period which falls in the range of a few milliseconds is neglected in all the proposed approaches. However, it surely is an important parameter especially in practical cases. This is our motivation to propose an energy efficient RWA algorithm which saves energy with low number of transitions from sleep mode to active mode for network links and nodes.

Link transition time from sleep mode to active mode becomes a limiting factor in practical light-path provisioning, especially when we deal with dynamic traffic scenarios. The importance of transition time from the sleep to active mode is discussed in the following from network provider as well as end user points of view. Under dynamic traffic case, call requests arrive at random time instants. Time duration between the request arrival time and the time it is established is referred to as setup delay. Each call expects to be established as soon as it arrives. Thus, link transition time from sleep to active mode, if not carefully taken into account, may significantly increase the overall average setup delay. In addition, link utilization is a major concern for network designers, and therefore, bringing the network to the QoS-satisfied state, with the lowest number of transitions, may increase the average utilization. This is because state of the art WDM systems are capable of carrying tens of high bit rate wavelength channels on a single fiber. Moreover, identifying the most strategic links, which are capable of increasing the connectivity of the graph, would result in reaching the desired QoS threshold using the lowest number of switchings. This, in turn, may increase the energy efficiency of the network since lower number of line amplifiers are put to operation mode. In particular, the efficient utilization of the deployed network capacity is one of the network operators’ ultimate goals that should be carefully considered, especially from routing point of view.

The objective of this paper is to develop an EA-RWA algorithm with QoS constraint which aims at simultaneously reducing total power consumption and total number of links and node transitions from sleep mode to active mode.

The contribution of this research is to develop a novel multi-objective ILP formulation for the energy aware static RWA which aims at simultaneously minimizing total power consumption, total cost of RWA, and total blocking rate. The developed three-objective ILP model is then solved with five efficient heuristics in order to achieve a satisfactory solution. Furthermore, to the best of our knowledge, no attempt has been carried out in order to take into account sleep to active transition time in designing EA-RWA. In this paper, we apply minimum spanning tree (MST) idea in designing a light-provisioning approach for all-optical networks in order to save energy while maintaining connectivity and satisfying a QoS constraint in the network. Besides, we aim at reducing the number of link/node transitions from sleep mode to active mode, at the same time.

This paper is organized as follows. In Section 2, network power consumption model, assumptions and problem definition are brought up. In Section 3, five novel EA-RWA heuristics are proposed. Section 4 evaluates the performances of the proposed heuristics and compares them with existing solutions. Finally, in Section 5, we conclude the article with an outline of topics which need to be investigated in direction with current research.

Section snippets

Network model, assumptions, and problem definition

In this section, we would like to develop a multi-objective integer linear programming (MO-ILP) formulation for energy aware static light-path establishment, considering QoS constraints. The developed formulation aims at minimizing: (1) total power consumption in the network, (2) total cost of establishing connections, and (3) total blocking rate.

First, we model the power consumption in the network by carefully separating the consumed power in different parts of the network elements. When a

Proposed energy aware RWA algorithms

Recall that the main idea behind EA-RWA approaches is to initially route the requests from low power links, and then assign low costs to such links in order to establish the remaining requests on the already switched on elements rather than turning on new active links. In order to reduce the number of sleep to active transitions, we develop a minimum spanning tree (MST) based RWA approaches to solve the routing and wavelength assignment problem for static traffic case.

In this research, five

Performance evaluation

Proposed algorithms are applied to two well known PacNet and COST239 network topologies, (e.g., Shen et al. (2009) and O’Mahony (1996)), properties of which are illustrated in Table 2 (see Fig. 8). Note that a unit of distance indicates 10 km in Fig. 8. Network power consumption parameters in our simulations are shown in Table 3 which are obtained from available data sheets (Shen and Tucker, 2009, Aleksic, 2009). It is also assumed that there exist eight available wavelength channels on each

Conclusion and future works

This research aims at reducing power consumption of the optical WDM internetwork of data centers in which a huge amount of data is being transmitted through high speed optical links. In order to do so, our objective is to make network load and resource utilization proportional to each other. Five energy aware RWA algorithms called MST-RWA and MST-LWUP1, MST-LWUP2, MST-LWUP3, and MST-LWUP4 was proposed in order to save energy. Simulation results indicate the fact that the proposed algorithms

References (31)

  • M.L. Fredman et al.

    Fibonacci heaps and their uses in improved network optimization algorithms

    J ACM

    (1987)
  • L. Guo et al.

    Green provisioning of many-to-many sessions over WDM optical networks

    J Lightwave Technol

    (2013)
  • M. Jinno et al.

    Spectrum-efficient and scalable elastic optical path network: architecture, benefits, and enabling technologies

    IEEE Commun Mag

    (2009)
  • R.H. Katz

    Tech titans building boom

    IEEE Spectr

    (2009)
  • J.B. Kruskal

    On the shortest spanning subtree of a graph and the traveling salesman problem

    Proc Am Math Soc

    (1956)
  • Cited by (6)

    • Energy-Aware Provisioning in Optical Cloud Networks

      2017, Computer Networks
      Citation Excerpt :

      An Exact ILP and an auxiliary matrix-based heuristic are proposed to solve this problem in each case when the Maximum Transparent Reach (MTR) is considered as distance and hops. Ebrahimzadeh et al. [39] consider amplifiers, switches, and transceivers as the main energy consuming devices in optical WDM networks. Suppose that the traffic matrix is known in advance, they propose both an exact ILP and five heuristics to accommodate the requests in an energy-efficient way.

    • Multiple many-to-many multicast routing scheme in green multi-granularity transport networks

      2015, Computer Networks
      Citation Excerpt :

      In [16], the auxiliary graph based heuristics and energy-awareness are introduced into the green optical networks so that the currently active components are used as many as possible and thus substantial energy is saved in network operation. In [17], five energy aware routing approaches are proposed to simultaneously minimize energy consumption in WDM networks and the number of transitions from sleep to active mode while considering blocking rate. However, the above researches on energy saving routing focus only on one single layer.

    • Electrical optical network RWA algorithm based on prediction and layered graph model

      2016, 2016 12th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery, ICNC-FSKD 2016
    • A survey on physical layer impairments aware routing and wavelength assignment algorithms in transparent wavelength routed optical networks

      2016, 2016 39th International Convention on Information and Communication Technology, Electronics and Microelectronics, MIPRO 2016 - Proceedings
    • Energy-aware traffic routing and resource allocation in WDM optical networks

      2015, SBMO/IEEE MTT-S International Microwave and Optoelectronics Conference Proceedings
    View full text