Optical Signal-to-Noise Ratio Restoration Algorithm Applied to Optical Network Resilience to Node Failures

Optical Signal-to-Noise Ratio Restoration Algorithm Applied to Optical Network Resilience to Node Failures Rodrigo C. de Freitas, Joaquim F. Martins-Filho, Daniel A. R. Chaves Department of Electronics and Systems Federal University of Pernambuco (UFPE) 50740-550, Recife PE, Brazil e-mail: {rodrigo.choji, jfmf, daniel.rchaves}@ufpe.br Rodrigo C. L. Silva, Carmelo J. A. Bastos-Filho, Helder A. Pereira Polytechnic School of Pernambuco University of Pernambuco (UPE) 52720-001, Recife PE, Brazil e-mail: rcls@ecomp.poli.br, carmelofilho@ieee.org, helder.pereira@poli.br Edluce dos S. Leitao Superior School of Technology Amazonas State University (UEA) 69065-020, Manaus AM, Brazil e-mail: esle.pdd@uea.edu.br Abstract In this paper we propose a novel algorithm to restore calls upon node failure in all-optical networks. Our proposal aims to restore the maximum of lightpaths disrupted by each node failure. The algorithm performs a path restoration looking for the route that presents the higher optical signal-tonoise ratio (OSNR). We compared our proposal to three other well known approaches in the literature. We observed that our proposal obtained lower failure rates in the restoration process for three different studied scenarios considering several physical layer impairments. I. INTRODUCTION A wavelength-routed optical network (WRON) is an effective solution to transmit data at high bit rates for long distances in a mesh network with relatively low cost. In alloptical WRONs, the signal remains in the optical domain along the entire lightpath [1]. Although all-optical networks present lower cost when compared to opaque networks, the physical layer impairments have higher influence on the Quality of Transmission (QoT) of the lightpaths. As a consequence, it is more difficult to find an appropriate route and assign a wavelength to attend to a call request [1], [2]. It also occurs when tackling the survivability problem in all-optical networks. This means that the optical signal degradation has to be taken into account on the search for spare resources in networks impaired by physical layer effects [3]. In general, a failure in a network element (link or node) can cause the interruption of several lightpaths, thereby leading to large data and revenue losses [3]. Since a common requirement for commercial systems is an availability of 99.999 % of the time [4], all-optical networks must be fault tolerant in order to avoid service interruption. The only practical alternative to ensure this availability is to provide mechanisms to guarantee the continuity of services in the case of faults. There are two main survival schemes for optical networks: protection and restoration. In the protection scheme, backup 978-1-4673-0279-1/11/$26.00 c 2011 IEEE routes with free wavelengths are reserved in advance and the resources remain dedicated during the network operation. In the restoration scheme, the system has to discover dynamically another route with a free wavelength for each disrupted connection [3]. In dynamic restoration scheme, the spare capacity available within the network is utilized for restoring services affected by a failure. In general, dynamic restoration schemes are more efficient in utilizing capacity due to the multiplexing of the spare-capacity requirements and provide resilience against different kinds of failures, while protection schemes are faster and guarantee the re-establishment of the lightpaths [5]. The dynamic restoration algorithms can be classified in two approaches [5]: i) link restoration, in which the new lightpath is calculated between the nodes that are connected to the broken link or the broken node and; ii) path restoration, in which a new lightpath is calculated from the source to the destination node, considering a modified topology, without the link or the node failed. Since link failures occurs more frequently than node failures in optical networks, most of the research efforts on survivability in optical networks have been focused on situations regarding link failures. For example, Yueping et al. [6] aimed to optimize the search for alternative routes in the case of single failure; Eiger et al. [7] designed restoration mechanisms intended to be independent of the type of failure. However, research dedicated to all-optical networks survivability, taking into account physical layer impairments, are still relatively scarce. Georgakilas et al. [8] evaluate the performance of an impairment-aware routing (IAR) scheme in the presence of single and dual-link failures; the problem of survivable lightpath provisioning in WDM networks considering the optical degradation of the signal triggered by the physical impairments was also investigated by Markidis and Tzanakaki in [9]. A path restoration algorithm which restores connections, under physical layer impairments, that

pass through a failed link, was proposed in our previous work [10]. On the other hand, node failures are another possibility to be reckoned with. Entire central offices can fail, usually because of catastrophic events such as fires, flooding, earthquakes, tsunamis [4]. These events are rare, but they cause widespread disruption and when they occur the failure is generally more serious than a link failure, isolating a region that is connected through the node from communicating with other places. Because of this, many efforts to develop efficient survival strategies to handle with this type of failure in optical networks have been made recently. In this paper, we focus on path restoration scheme and we propose a novel algorithm to restore calls upon a node failure in all-optical networks considering physical layer impairments. The proposed algorithm searches for an alternative route with the higher OSNR for each disrupted lightpath. The analytical model used to evaluate the OSNR of the lightpaths was proposed by Pereira et al. [11]. The rest of the paper is organized as follows: Section II presents some previous works tackling the node failure problem in optical networks. In Section III, we describe the OSNR model and we detail the physical layer effects considered in this paper. In Section IV, we present the methodology used to validate our proposal and the novel node failure restoration algorithm based in OSNR. In Section V we describe the simulation setup. In Section VI we show some simulation results. In Section VII we give our conclusions. II. LITERATURE REVIEW: NODE FAILURE RESTORATION APPROACHES The design of resilient optical networks to single or doublelink failures and proposals of efficients algorithms to restore a disrupted connection from a broken link, has been extensively researched. However, the study of nodes failures has been rarely discussed. Kim and Lumetta [12] researched the effectiveness of link-protection schemes in terms of their ability to handle node failures in all-optical networks. Cheng et al. [13] proposed a path-based joint working and backup path selection scheme and a greedy algorithm to reserve backup bandwidth for survivability against random multiple-link. In the work of Cheng et al., we can note an important feature: although have been considered explicitly fail in several links (multiple-link failures), the solution can be applied for a node fault scenario too, since a single node failure can cause the disrupting of several connections that pass through links connected to the failed node. Eiger et al. [7] present an algorithm for optical networks with a restoration mechanism that provides an end-to-end path protection to a set of given demands under a single link or node failure with a focus on optical networks. Bao et al. [14] propose a hybrid protection strategy based on node-disjointness, named segment and path shared protection, against arbitrary simultaneous double-node failures in optical networks. III. OSNR DEGRADATION All-optical networks generally operate with high transmission capacities and the signal remains in the optical domain between the edge nodes, i.e. the signal propagates along the core of the optical network without any optical-electrical-optical conversion. Because of the linear and nonlinear physical layer effects in optical fibers and the additive noise inserted by the network elements along the lightpaths, the OSNR of the transmitted signal can be degraded, which have directly impact on the QoT [11]. In this paper, we used an analytical model, proposed by Pereira et al. [11], based on OSNR degradation to take into account the effect of gain saturation and amplified spontaneous emission (ASE) noise depletion in amplifiers, the residual chromatic dispersion effect, the coherent crosstalk in optical switches and the polarization mode dispersion (PMD) in optical fibers. The model considers these effects all together and it uses simple analytical equations obtained from well known behaviors of network devices [11]. We neglected the effect of four-wave mixing (FWM) since it takes a long time to be evaluated and is not crucial to demonstrate that our proposal is suitable to solve the survivability problem in alloptical networks. IV. OSNR-R NODE FAILURE RESTORATION ALGORITHM Dynamic restoration schemes must discover dynamically an alternative lightpath for each interrupted connection [15]. However, to our knowledge, all the previous approaches use the shortest path or the minimum number of hops metrics to search these alternatives lightpaths and most of them only considers link failure. We propose in this paper to search dynamically for alternatives lightpaths with higher OSNR at the destination node. We believe this can allow to simultaneously finding lightpaths with acceptable QoT and distributing the load along the network. Simple algorithms to balance the load such as least resistance weight (LRW) could be applied, but these approaches tend to often find routes that are too long to maintain the QoT. We used the following methodology to evaluate our approach: (1) we studied the network stability in relation to the use of its resources in order to define when the network achieves the steady state; (2) we stored the network state in order to realize the node failures simulations in a realistic scenario. Given this stored network state, we removed one node at a time and we attempted to recover the calls that were passing through it. A call is considered restored when there is a free wavelength in an available path to restore the disrupted call and the OSNR of this lightpath is above a threshold value (OSNR QoT ). We did these simulations for every node in the network. Algorithm 1 presents the pseudocode for the used methodology to evaluate the failure rate for Path Restoration using OSNR-R with node failure in all-optical networks. In line 1 we simulate a failure in each node of the network topology. In line 2, we create a data structure to mantain the information of the network links. In line 5,

we identify all links connected to the failed node. Another data structure is created in line 6, but now it mantains the information of the network nodes. In lines 8 and 9 the resources used by the link connected to the failed node are set free. In line 10, the calls are marked to indicate that they already participated of the restoration process. In lines 12 and 13, the node and the link are indeed removed of the topology. Between lines 14 and 21 the restoration process is done. Algorithm 1 Methodology to evaluate the failure rate for Path Restoration in a Node Failure Scenario 1: for every node-failed of the network do 2: Create an auxiliary topology called net-link; 3: Create a list called recovered-calls; 4: Create a list called link-calls; 5: for every link connected to the node-failed do 6: Create an auxiliary topology called net-node; 7: for every call that passes through the current link do 8: Set free the occupied lightpaths which pass by the link, using net-link; 9: Set free the occupied lightpaths which pass by the link, using net-node; 10: Save the calls of current link in a list called linkcalls; 11: end for 12: Simulate a node failure using the auxiliary topology net-node; 13: Remove the current link of the auxiliary topology net-link; 14: for every call of link-calls list do 15: if the call is in the recovered-calls list then 16: Treat the next call; 17: else 18: Search for an alternative lightpath in the auxiliary topology net-node, considering higher OSNR-R at the destination node; 19: Evaluate physical layer impairments; 20: end if 21: end for 22: Count the quantity of recovered calls; 23: end for 24: end for V. SIMULATION SETUP We have considered the following physical impairments in the simulations: the effect of gain saturation and the ASE noise depletion in amplifiers, the coherent crosstalk in optical switches, the residual chromatic dispersion effect and the PMD. We tested and compared our proposal (OSNR-R Node Failure Restoration Algorithm) to three well known metrics to find a backup route: shortest path (SP), minimum number of hops (MH) and least resistance weight (LRW). The procedure for the other algorithms is the same used in Algorithm 1, except for the line 18. The line 18 is substituted by: Search for an alternative lightpath considering shortest path, for the SP algorithm; Search for an alternative lightpath considering minimum number of hops, for the MH algorithm; and Search for an alternative lightpath considering the load distribution, for the LRW algorithm. The topology used to perform the simulations is shown in Fig. 1 and Table I describes the network parameters used in our simulations. We used the Simulator for Transparent Optical Networks (SIMTON), developed by Chaves et al. [16]. Fig. 1. Modified Pacific Bell Topology. TABLE I DEFAULT SIMULATION PARAMETERS. Parameter Value Definition P sat 26 dbm Amplifier output saturation power. OSNR in 40 db Input optical signal-to-noise ratio. OSNR QoT 23 db Optical signal-to-noise ratio for QoT criterion. B 40 Gbps Transmission bit rate. B o 100 GHz Optical filter bandwidth. f 100 GHz Channel spacing. λ i 1529.56 nm The lower wavelength of the grid. λ 0 1557 nm Zero dispersion wavelength. α 0.2 db/km Fiber loss coefficient. L Mux 2 db Multiplexer loss. L Demux 2 db Demultiplexer loss. L Switch 2 db Optical switch loss. δ 10% Maximum pulse broadening. W 16,21,40 Number of wavelengths per link, considered in each simulation scenario. F 0 (NF) 3.162 (5 db) Amplifier noise factor (Noise figure). λ tx 0.05 nm Transmitter laser linewidth. Initially, we analyzed the network occupation in order to identify the necessary number of calls to reach the steady state regarding the average number of wavelengths used in each link. This investigation is important because the faults should be generated considering the steady state of the network. We

believe most of failures in a realistic scenario occur when the network is already in the steady state operation. For this analysis, we simulated 200, 000 calls generated by a Poisson process. The source-destination pairs were generated by an uniform distribution. We analyzed three scenarios: (i) 16 wavelengths per link, as shown in Fig. 2, (ii) 21 wavelengths per link, as shown in Fig. 3 and (iii) 40 wavelengths per link, as shown in Fig. 4. Table II shows the network blocking probabilities due to the lack of wavelengths (BP λ ) and insufficient quality of transmission (BP QoT ), for network load of 60 Erlang. One should observe that the steady state is reached for the three studied scenarios after 200 and before 600 simulated calls. TABLE II BP QoT AND BP λ FOR 16, 21 AND 40 WAVELENGTHS PER LINK, CONSIDERING SP, MH, LRW AND OSNR-R ALGORITHMS. Qty.λ SP MH LRW OSNRR BP QoT BP λ BP QoT BP λ BP QoT BP λ BP QoT BP λ 16 0.0068 0.076 0.018 0.169 0.049 0.012 0.0044 0.056 21 0.01 0.012 0.03 0.084 0.0526 0 0.0069 0.0074 40 0.011 0 0.04 0 0.052 0 0.0079 0 Fig. 2. Average number of used wavelengths per link as a function of the number of simulated calls for 16 wavelengths per link. Fig. 3. Average number of used wavelengths per link as a function of the number of simulated calls for 21 wavelengths per link. For each scenario we chose an arbitrary point after stabilization to stop the simulation and we stored the network state, including the description of the active lightpaths. After that, one node is removed from the network topology, simulating Fig. 4. Average number of used wavelengths per link as a function of the number of simulated calls for 40 wavelengths per link. a node failure. We tested the SP, MH, LRW and OSNR-R algorithms to recover each of the lightpaths that pass through the failed node. This process of removing nodes and searching for alternative routes is done for every node in the network. VI. RESULTS We tested and compared our proposal (OSNR-R Node Failure Restoration Algorithm) with SP, MH and LRW algorithms for three scenarios (16, 21 and 40 wavelengths), as we mentioned in Section V. Fig. 5 shows the restoration failure rate as a function of network load, considering 16 wavelengths per link. In this case, we have a total blocking probability around 7%, due mostly to lack wavelength. We observe that OSNR-R node failure restoration algorithm presented a slight advantage over LRW and SP. Our proposal far outperformed the MH restoration algorithm. Also we can observe that until 60 Erlang, LRW algorithm has a worse performance than SP, but from 60 Erlang onwards, LRW presents a better performance than SP due to the increased load. Instead, the OSNR-R node failure restoration algorithm reachs better performance for all network loads. This behavior was expected since the failure rate and the network blocking probability are mainly due the lack of available wavelength in this scenario, as can be seen from Table II. Fig. 6 presents the failure rate in the scenario with 21 wavelengths per link. In this case, we have a total blocking probability around 2%, distributed as 1% of call blocking due to the lack wavelength and 1% of call blocking due to physical layer impairments. In this situation, we observe that OSNR- R node failure restoration algorithm remains having a better performance than SP, MH and LRW restoration algorithms. One can observe that the SP algorithm obtained the second better performance for this scenario. This is expected because the network has a reasonable amount of wavelengths, therefore the intense network traffic is not a decisive factor, as in the scenario 1. Fig. 7 shows the same analysis for 40 wavelengths per link. One should notice that the blocking due to the lack of available wavelengths is zero (see Table II), i.e. the calls are blocked solely due to the physical layer impairments. In this scenario, the OSNR-R node failure restoration algorithm is the approach that recovers more disrupted calls again. One

0,74 0,72 0,70 0,68 0,66 0,64 0,62 0,58 0,56 0,54 0,52 0,48 0,46 0,44 0,42 SP MH LRW OSNR-R Scenario 1 6 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Network Load (Erlang) 0,58 0,56 0,54 0,52 0,48 0,46 0,44 0,42 SP MH LRW OSNR-R Scenario 3 = 40 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Network Load (Erlang) Fig. 5. Restoration failure rate as a function of network load considering 16 wavelengths per link. Fig. 7. Restoration failure rate as a function of network load considering 40 wavelengths per link. 0,74 0,72 0,70 0,68 0,66 0,64 0,62 0,58 0,56 0,54 0,52 0,48 0,46 0,44 0,42 SP MH LRW OSNR-R Scenario 2 = 21 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Network Load (Erlang) proposal achieved failure rate as good as the LRW approach. In the other cases, OSNR-R node failure restoration algorithm presented lower failure rate than the other three approaches. 0,8 0,7 0,6 0,5 0,4 0,3 Fig. 6. Restoration failure rate as a function of network load considering 21 wavelengths per link. should observe that the LRW algorithm is very inefficient in this case, probably because it tends to distribute load along the network and does not take into account the impairments from the physical layer. We can note that LRW and MH algorithms have exactly the same performance until 60 Erlang. Fig. 8, Fig. 9 and Fig. 10 show the boxplot of the failure rate for the restoration algorithms considering 16, 21 and 40 wavelengths per link, respectively. For these simulations we considered a network load of 60 Erlang. Fifty simulations were performed for each situation. These results indicate the confidence level of the results presented in Fig. 5, Fig. 6 and Fig. 7. These figures confirm that in the worst case, our SP MH LRW OSNR-R Routing Algorithm Fig. 8. Boxplot of the restoration failure rate for the restoration algorithms considering 16 wavelengths per link. VII. CONCLUSIONS We showed that the OSNR-R can be used to tackle the survivability problem in all-optical networks. We presented an implementation for this restoration algorithm, which takes into account the physical layer impairments to search dinamically for the alternatives lightpaths. It uses an analytical model which takes into account the main linear and nonlinear effects

0,70 0,65 0,55 0,45 SP MH LRW OSNR-R Routing Algorithm Fig. 9. Boxplot of the restoration failure rate for the restoration algorithms considering 21 wavelengths per link. 0,65 0,55 0,45 SP MH LRW OSNR-R Routing Algorithm Fig. 10. Boxplot of the restoration failure rate for the restoration algorithms considering 40 wavelengths per link. REFERENCES [1] B. Mukherjee, WDM Optical Communication Networks: Progress and Challenges, IEEE J. Sel. Areas Commun., vol. 18, no. 10, pp. 1810 1824, Oct 2000. [2] M. O Mahony, C. Politi, D. Klonidis, R. Nejabati, and D. Simeonidou, Future Optical Networks, IEEE/OSA Journal of Lightwave Technology, vol. 24, no. 12, pp. 4684 4696, 2006. [3] C. Ou and B. Mukherjee, Survivable Optical WDM Networks, 1st ed., ser. Optical Networks Series. Springer, 2005. [4] R. Ramaswami and K. Sivarajan, Optical Networks - A Practical Perspective, 2nd ed., ser. Academic Press. Morgan Kaufmann Publishers, 2002. [5] S. Ramamurthy, L. Sahasrabuddhe, and B. Mukherjee, Survivable WDM Mesh Networks, Journal of Lightwave Technology, vol. 21, no. 4, pp. 870 883, apr 2003. [6] L. Yueping, Z. Nie, and X. Zhou, Finding the Optimal Shortest Path Tree with Respect to Single Link Failure Recovery, in 4th International Conference on Network Computing and Advanced Information Management, 2008, pp. 412-415. [7] M. Eiger, H. Luss, and D. Shallcross, Network Restoration Under a Single Link or Node Failure using Preconfigured Virtual Cycles, Telecommunication Systems, vol. 46, pp. 17 30, 2011. [8] K. Georgakilas, K. Katrinis, A. Tzanakaki, and O. Madsen, Performance Evaluation of Impairment-Aware Routing Under Single- and Double-link Failures, IEEE - 4th International Conference on Networked Computing and Advanced Information Management, vol. 2, no. 8, pp. 633 641, 2010. [9] G. Markidis and A. Tzanakaki, Routing and Wavelength Assignment Algorithms in Survivable WDM Networks under Physical Layer Constraints, in 5th International Conference on Broadband Communications, Networks and Systems, sept. 2008, pp. 191 196. [10] R. C. Freitas, R. C. L. Silva, H. A. Pereira, D. A. R. Chaves, C. J. A. Bastos-Filho, and J. F. Martins-Filho, A Novel Restoration Algorithm Based on Optical Signal to Noise Ratio for Transparent Optical Networks, in Telecommunications Brazilian Symposium (SBrT 2011), oct 2011. [11] H. A. Pereira, D. A. R. Chaves, C. J. A. Bastos-Filho, and J. F. Martins-Filho, OSNR Model to Consider Physical Layer Impairments in Transparent Optical Networks, Photonics Network Communications, vol. 18, no. 2, pp. 137 148, 2009. [12] S. Kim and S. S. Lumetta, Addressing Node Failures in All-Optical Networks, Journal of Optical Networking, vol. 1, no. 4, pp. 154 163, Apr 2002. [13] X. Cheng, X. Shao, and Y. Wang, Multiple Link Failure Recovery in Survivable Optical Networks, Photonic Network Communications, vol. 14, pp. 159 164, 2007. [14] N. H. Bao, Z. Z. Zhang, L. M. Li, H. F. Yu, and H. B. Luo, A Hybrid Protection Strategy Based on Node-Disjointness Against Double Failures in Optical Mesh Networks, Photonic Network Communications, pp. 1 10, 2011. [15] B. Mukherjee, Optical WDM Networks, 1st ed., ser. Optical Networks Series. Springer, 2006. [16] D. A. R. Chaves, H. A. Pereira, C. J. A. Bastos-Filho, and J. F. Martins- Filho, SIMTON: A Simulator for Transparent Optical Networks, Journal of Communication and Information Systems, vol. 25, 2010. [Online]. Available: http://iecom.dee.ufcg.edu.br/ jcis/ that affect the QoT of the signals. We performed simulations in three different scenarios and our proposal achieved an equal or superior performance when compared to other well known approaches in all cases. ACKNOWLEDGMENT The authors acknowledge the financial support from FACEPE, CNPq, CAPES, UFPE, UEA, UPE and TJAM for scholarships and grants.