A Proposed BSR Heuristic Considering Physical Layer Awareness

Transcription

1 A Proposed BSR Heuristic Considering Physical Layer Awareness 1 st Pedro J. F. C. Souza pedro-freire@hotmail.com 4 th Karcius D. R. Assis Department of Electrical Engineering Federal University of Bahia Salvador, Brazil karcius.assis@ufba.br 2 nd Alex F. Santos Federal University of Bahia Federal University Recôncavo of Bahia Feira de Santana, Brazil alex.ferreira@ufrb.edu.br 5 th Matheus R. Sena matheus-eletro44@yahoo.com 3 rd Raul C. Almeida Jr. ralmeida.ufpe@gmail.com Abstract In Elastic Optical Networks (EON), the problem of routing and spectrum Allocation (RSA) has been solved along the years through the use of optimization techniques or heuristics, with the aim of minimizing the use of network resources or maximizing the capacity for future requests. The effects of the physical layer in the RSA problem are of fundamental importance given that there is a distance limitation with the required bit rate and modulation format used to transmit the signal. On considering the linear and non-linear noise effects, it is possible to include quality of transmission (QoT) in the optimization process. In this paper, we propose a modification to a traditional heuristic for the RSA problem in Elastic Optical networks to consider both effects of the physical layer. The proposed modification takes into account the channel input power and Margin to choose all modulation formats and to assign resources. The main goal is to increase the spectral efficiency and throughput of the channels by mitigating blockages in the physical layer. Simulations were performed for various topologies, and the results suggest advantages in terms of number of blocked channels and spectrum usage of the new proposal in relation to the traditional BSR heuristics. Keywords Elastic Optical Networks; Routing, Modulation and Spectrum Allocation; Non-Linear Effect; IGN Model; Margin; Iterative Algorithm. I. INTRODUCTION The Elastic Optical Networks (EONs) were recently proposed as a solution to maximize optical spectrum efficiency when compared to the traditional WDM Optical Networks. In EONs, the optical spectrum is divided into spectral slices called slots, where an entire number of contiguous slots can be allocated to the connections to provide exactly the required bandwidth. This value is specified by the requested bit rate and the modulation format capable of meeting the quality of transmission (QoT) requirements. Those requirements are ruled by the influence of elements crossed the channel in each link and the distance to be traveled by the signal along the optical path. This paper focus on a static traffic scenario, where the traffic demand is given beforehand and a routing and spectrum assignment planning is performed. In this case, the problem usually comes down to reducing network resources (number of slots) to meet the entire demand, or to meet the maximum of the demand for a given amount of resources [1]. In the face of all these relations between routing and physical layer, an interesting question arises: How the effects of the physical layer interfere on the performance of the algorithms of routing and spectrum allocation? The answer is not trivial, since complex calculations, often with non-linear equations, and dependence on the channel position, bandwidth and launch power need to be performed and considered as restrictive parameters in the RSA algorithms. However, a study that considers some of these effects may serve as a benchmark for the proposal of new RMSA algorithms that consider network and physical layer constraints. The Best Among the Shortest Routes (BSR) is a heuristic algorithm proposed in [2] that considers the spectral usage (utilization) in the cost of each optical link in order to obtain a set of shortest routes that avoids connection blocking due to lack of network resource (network layer blocking). However, just the topology of the optical network is taken into account. This paper extends the BSR by linking the channel bandwidth with its. For a given channel, the algorithm defines its bandwidth by choosing the most spectrally efficient available modulation format that achieves the desired QoT. Also, including the adaptation on the heuristic, the link cost may now feel the physical layer and choose different routes, as well. Finally, in an iterative way, a finite set of values of channel power, SRN margin and modulation format are tested, and the best option that generates the highest spectral efficiency and the lowest number of blocked requests is selected. II. PHYSICAL LAYER BACKGROUND This study has considered both ASE noise generated by the optical amplifiers and the non-linear SCI (Self-Channel Interference) and XCI (Cross-Channel Interference) contributions. Assuming that the channels have an approximately rectangular PSD, the IGN (Incoherent Gaussian-Noise) [3] model has been adopted as a tool to estimate signal quality through the (Signal to Noise Ratio), which is based on an additive White Noise approach. To understand the model, assume that a channel i is transmitted in a route R i with central frequency f ch,i and bandwidth B ch,i. The of that connection can be derived from the following set of equations [3]: /1/$ IEEE

2 , = (1) +, =, =, (2),, = 1 ( )/ h (3), = 27,, (2 (4), ), = 1 2 If n i, X is the XCI contribution, so:, = asinh ( (2 ) [,, +, 2 ], ) 4 (2 ) asinh ( (2 ) [,,, 2 ], ) 4 (2 ) (5) () If n = i, X is the SCI contribution, so:, = asinh ( (4 ), ) 2 (2 ) (7) N l is the number of spans in link l, R s,i is the symbol rate of channel i, β 2 is the Chromatic Dispersion Coefficient, α is the Fiber Field Loss Coefficient, G ch,i is the PSD of channel i, γ is the Nonlinear Coefficient, L eff is the effective length of span, L is the length of the span, δ n,i is the Dirac delta function, NF is the amplifier noise figure, A l is the optical amplifier gain, v is the carrier frequency, in this paper we adopted THz, and h is the planck s constant. In this model, it is assumed that all fiber properties are equal, and all losses are compensated with the amplifier gain on the fiber. Under physical impairment awareness, a very important parameter is the minimum required signal-to-noise rate ( thr) [4], which is the that guarantees that the signal BER (Bit Error Rate) will satisfy the established by the SLA (Service Level Agreement) after a Hard Decision Forward Error Correction procedure. This means that a channel must be blocked if its is below the thr for its modulation format. TABLE I. THRESHOLD FOR EACH MODULATION FORMAT TO A PRE FEC-BER OF.4 AND BANDWIDTH FOR A BIT RATE EQUALS TO 3 GBITS/S Modulation Format thr (db) Bandwidth (GHz) Slots QPSK QAM QAM QAM QAM Table I shows the thr of each modulation format calculated for a 3 Gbits/s connection, using an ideal constellation and the AWGN (Additive white Gaussian noise) model to reach a value of pre FEC-BER equal to.4. This value is commonly used in the literature [5], because, after passing through a hard decision decoder, the BER level gets close to The Table I also indicates the connection bandwidth (B ch) and number of slots. The relation = and roll off = were used to calculate the values. III. HEURISTIC The traditional BSR heuristic is a strategy that aims to find the most appropriate routes in terms of minimizing blocking at the network layer. There may exist more than one shortestpath route (refereed in this paper as Candidate Routes - R C) for each source and destination pair in the network. Since there usually are several shortest-path routes for each sourcedestination pair, the number of different solutions, K, for the planning of fixed routes on a given network topology is very high. Each iteration i of BSR is one candidate solution, S i, in the universe of K possible routing alternatives. The steps to choose a solution are: 1) Assign cost 1 for all links and run the routing algorithm to find the initial routing solution (S ) for all source-destination pairs; 2) Estimate the utilization of each link in the network by summing the number of slots requested in each route that passes through the link. 3) Set a new cost ( ) =. ( ) + (1 ). ( ) for each link; where c(l) is the cost of the link l, u(l) is the utilization (number of occupied slots) on link l and x is a constant close to 1. 4) Run the routing algorithm to find a new routing solution; 5) If the maximum number of iterations has not been reached, go back to 2), otherwise end the process. The most appropriate routing solution at the end of the process is chosen. We use in this paper the number of 2 iterations. The traditional BSR can efficiently distribute the load among the links and reduce the number of blocked connections. However, the BSR heuristic as originally proposed assumes that the number of required slots is given beforehand. In addition, the link utilization is given by the sum of the number of slots in each route that passes through the link. If just ASE and SCI are considered, their effects on the chosen route can be estimated, the modulation format can be defined, and the number of slots can be pre-calculated. However, under XCI, the relative position of the neighboring connections (which alter the term f ch,n in ) and their modulation format (which alter the term B ch,n in ) directly affects the XCI intensity on connection i. This makes the choice of the modulation format (and consequently the number of slots) of connection i dependent on the routing, spectrum positioning and assigned modulation format (RMSA) of other (interfering) connections in the network; and these connections also depend on the RMSA choice made to connection i and others. All this makes the RMSA problem severely more difficult. In order to circumvent this problem, we propose an adapted BSR, which, during the iterative process, leads with the ASE and SCI effects on each connection and defines an margin ( ) to each modulation format thr in order to (indirectly) account for the XCI generated by neighboring channels. The margin is set so that, if the of a connection is between thr and thr+ margin of a candidate modulation format, the immediately inferior (i.e.,

3 less spectrally efficient) modulation format is assigned to the connection. On the other hand, a candidate modulation format is assigned only if the connection is above thr+ margin of the modulation format. This process is performed along the BSR iterative process using to calculate the required number of slots of each connection the most spectrally efficient modulation format that fulfills the above conditions. With the best set of routes at the end of the BSR process, a second phase stars to establish the requests in the network and account for the XCI effect. The First-Fit (FF) spectrum assignment algorithm is used to establish the demands in the network. After that, the real XCI effect on the connections is quantified. This is performed starting from the longest routes to the shortest ones. If the thr of a connection is not satisfied, it is called in this paper as blocked channel. Also, we assumed that the modulation formats are indexed as m=1, 2,, M, where M is the highest spectral efficiency modulation format available in the network. In this work, we have assumed the same launched power and margin to all channels. This does not take into account the fact that shorter paths should require lower margins and XCI is more pronounced in inner channels so that a higher launched power could be allocated to external channels. This is under investigation. In order to estimate the best modulation format to be used, we run the modified BSR varying the channel input power from -7dBm to 7dBm in steps of 1dB, and margin from db to 5dB in steps of 1dB. The proposed heuristic pseudocode is described in Algorithm I. In addition, for the 32 QAM case the physical layer calculation considers a bandwidth of 2.4 slots, but in the network layer it is reserved 3 slots to allocate this channel. Algorithm 1 Proposed Heuristic to choose the Modulation Format INPUT: Channel Power, Requested Bit Rate and Margin OUTPUT: Modulation Formats 1: for Channel = 1 to Number of channels do 2: MF = M (Highest available modulation format index) 3: ok=; 4: while = do 5: Calculate considering ASE and SCI; : if ( ) + then 7: Modulation(Channel)=MF; : NSlots(Channel)=.. 9: ok=1; 1: else 11: - 1; (next available modulation format) : end if 13: end while : end for Algorithm II describes the pseudocode of the second part of the proposed BSR heuristic. It is used to calculate the number of blocked channels for each possible combination of channel input power and margin. After allocating all requested channels in the spectrum, the XCI contribution is effectively found. The is calculated for each channel and it is registered the number of channel that are below the Thr of its modulation format. Algorithm 2 Proposed Heuristic to calculate the number of blocked channels INPUT: Channel Power and Modulation Formats OUTPUT: Total Blocked Channels and Utilization 1: 2: Calculate Spectral Occupancy of the network (Utilization) 3: Block =; 4: for Ch = 1 to Number of channels do 5: Calculate considering ASE, SCI and XCI; : if ( ( )) then 7: Block =; : else 9: = + 1; 1 end if : 1 end for 1: In the end of the Algorithm the Power channel input, margin and modulation formats are chosen considering the option with less blocked channel and spectrum utilization. This paper focus on a static traffic scenario, but it is important to highlight that under dynamic traffic conditions, although XCI may be calculated to the request and all active connections, an margin may be applied to prevent the establishment of the request with a very narrow margin, which could imply in blocking of future requests. In the literature []-[], the usual strategy is distanceadaptive heuristics, which determines the modulation format and the consequent number of slots based on the light path distance. In a network with very different link lengths, such assumption may not be effective, since a specific modulation format may reach different number of hops depending on the routing solution and a not adjusted modulation format may be set up for the request. In this work, we do not assume such simplification, but effectively calculate the noise from the amplifiers, SCI and XCI on the choice of the modulation format for each candidate light path. IV. SIMULATION AND NUMERICAL RESULTS To analyze the effectiveness of the new strategy, simulations were made with networks of different sizes. The smallest is Abilene Net with 11 nodes, the intermediary is NFS NET with nodes and the largest is the EON (European optical network) Net with 19 nodes. In this article, the simulation parameters are listed in the Table II. The first analysis to be made is the heuristics performance in the best point found in terms of utilization considering less blocking channels. The set of formats used for the test were 4QAM, QAM and QAM. Figure 1 shows the results of blocked channels and utilization in points of best found channel power for different margin and The results showed that increasing the margin up to 2dB we reduce the blocked channel (%) using same spectrum space. Also, with the margin equals to 3, utilization was improved, and the number of blocked channels was zero for

4 all To explain the achievement of finding more appropriate point of power and Margin, it was necessary to study the behavior of the network utilization, blocked channels and the average of all channels of the network for each input (Power, Margin) [9]. Results of Abilene Network are presented in Figure 2. Blocked Channels (%) TABLE II. PARAMETERS USED IN THE SIMULATOR Parameters Values Span Length 1 Km Attenuation Coefficient.57 Km -1 Chromatic Dispersion Coefficient -21.3ps 2 Km -1 Non-linear Coefficient 1.3W -1 Km -1 Carrier Frequency THz Symbol Rate 3 Bd Spectral Roll Off Bandwidth 1 slot.5 GHz Noise Figure 4.4 db Number of polarization Abilene Utilization (Number of Slots) NSFNet Utilization (Number of Slots) EON Utilization (Number of Slots) Abilene Blocked Channels (%) NSFNet Blocked Channels (%) EON Blocked Channels (%) Margin Fig. 1. Results of utilization and blocking channel (%) in the best power input situation. As can be observed, initially with the margin equals to zero the power of maximum does not coincide with the minimum of blocked channels. As the and the utilization are tied in the heuristic, this makes this point of minimum of blocking assumes lower modulation formats. Beginning to increase the margin, the results showed that the two powers began to make one from another. With margin equal to 3dB, the two powers coincide and consequently the spectral usage decreases. Continuing to increase the margin, it is observed that the utilization does not continue to decline because it is already at the point of highest, but a higher margin restricts the choice of modulation formats that use less spectrum. The same trend was observed in the other This behavior can be explained by the fact that without the margin, the heuristic does not consider the XCI effect. Therefore, there is a possibility of choosing modulation formats with high threshold and after the entry of neighbor channels, they end up being blocked. In the case of zero margin the chosen power in the heuristic was -2dBm. For lower power intensity, the is lower, and the ASE noise is more representative which made better prediction occurs in the choice of the format. Adding the margin in the method of choice, the estimate of XCI contribution began to make part of the choice of modulation formats. This fact can be observed Utilization (Number of Slots) in the curve for Block (%) that shows abrupt variations vs Channel Power each time an increase of channel power changes the modulation format of the channel. This strategy brought an average gain in utilization and blockage, due to the fact of making possible to choose channel power with higher, as can be seen in Figure 2 highlighted with red circles Block (%) for Margin = Block (%) for Margin = Average for Margin = Average for Margin =1 (a) Margin = db and 1dB Block (%) for Margin =2 Block (%) for Margin =3 Average for Margin =2 Average for Margin =3 (b) Margin = 2dB and 3dB Block (%) for Margin =4 Block (%) for Margin = Average for Margin =2 Average for Margin =5 (c) Margin = 4dB and 5dB Fig. 2. Results of Average and blocking channel (%) for different Margin in Abilene Network. To finish the first analysis, Table III summarizes the results of blocked channels (%) and utilization for the traditional BSR with all channels assuming 4QAM, QAM or QAM, to represent its physical layer unawareness, and the proposal heuristic, both at the best point of channel power founded in the algorithm

5 The results show that the adapted BSR proved to be more advantageous, in most of the cases, when compared with the traditional BSR. In many cases the traditional heuristic did not reached zero blocked channels. For those which have reached zero blocked channels as the adapted heuristic, an improvement in the spectral utilization was shown with the adapted strategy. For instance, it was achieved 31% less slots used than the 4QAM traditional case in the NSF NET. TABLE III. RESULTS OF BLOCKED CHANNELS (%) AND UTILIZATION IN DIFFERENT SCENERIES Heuristic Blocked Channels Utilization (Slots) Abilene Adapted BSR 5 Abilene BSR 4QAM Abilene BSR QAM 2 74 Abilene BSR QAM NFS Adapted BSR 3 NFS BSR 4QAM 91 NFS BSR QAM 4 NFS BSR QAM EON Adapted BSR 92 EON BSR 4QAM EON BSR QAM 9 EON BSR QAM 7 71 However, a slightly different was found when comparing the adapted heuristic with the QAM case in a large topology (EON). The cause of a 3% higher spectral utilization can be explained by the fact that we are considering the same margin for all channels. In a larger network, there is a higher probability of overestimation on the XCI effect for a set of channels. In turn, this overestimation causes an error in Algorithm 1 making it chooses a format of modulation less than needed to not have blocked. This makes greater use of spectrum than necessary. The second analysis is related to the behavior of the heuristic in choosing modulation format when it is available only high-power intensities, in our case 7dBm. How much greater is the channel power, the greater is the contribution of the XCI effect on the channel. So, this test will aim to show how the assignment of the margin has helped in the prediction of this term avoiding blocking and reducing the use of spectrum, even with higher XCI contribution. Number of Blocked Channels Adapt. Heuristic N Block QAM Heuristic N Block QAM Heuristic N Block 4 QAM Heuristic N Block Adapt. Heuristic Utilization QAM Heuristic Utilization QAM Heuristic Utilization 4 QAM Heuristic Utilization Margin Fig. 3. Comparison of Blocked Channels and utilization between the modified heuristic and a heuristic that does not consider the physical layer Figure 3 has shown, for Abilene topology, that without the margin, the number of blocked channels went high. Also, Utilization (Number of Slots) when you increase the margin in the heuristic, the values of blockade were reduced to zero. Already in the first attempt with a margin equal to 1dB, we had better results of blocked channels that the proposal of all channels assuming QAM or QAM, in all As for the case of all channels being 4QAM, despite having obtained the result of zero blocking channels, the adapted heuristic presented a better use of spectrum. The same behavior was observed in the other Thus, we conclude that not even calculating effectively the XCI effect in the parameter of the, our adapted heuristic predicts satisfactorily modulation formats that will prevent the blocking in the network, even in cases where the XCI greatly influences the QoT of the network. V. CONCLUSION In this article, it was presented an adaptation for bandwidth allocation and route assessment to support the planning of optical networks considering the effects on the physical layer. When using a methodology of choosing the best power input channels and the best format for modulation, it is shown, through three distinct networks, that there is an improvement in the throughput network, since less channels are blocked. In addition, it was observed that exists a margin which achieves the maximum average of the channel and the minimum number of blocked channels with same input Power. This information has helped the heuristic BSR to increase the spectrum efficiency, transmitting same information with less slots. Finally, we showed that the strategy selected the correct modulation format, in most of the cases, even when situations where XCI was very intense. REFERENCES [1] M. Aibin and K. Walkowiak. Adaptive modulation and regeneratoraware dynamic routing algorithm in elastic optical networks.ieee International Conference on Communications (ICC), pp , 215. [2] K. D. R.Assis, A. F. dos Santos and I. M. Queiroz. Routing in EON networks under mixed static and dynamic traffic. In Optical Metro Networks and Short-Haul Systems IX. International Society for Optics and Photonics, 217. [3] P. Poggiolini, G. Bosco, A. Carena, V. Curri, Y. Jiang and F. Forghieri. The GN-model of fiber non-linear propagation and its applications. Journal of lightwave technology, v. 32, n. 4, p , 2. [4] L. Yan, E. Agrell, H. Wymeersch and M. Brandt-Pearce. Resource allocation for flexible-grid optical networks with nonlinear channel model. IEEE/OSA Journal of Optical Communications and Networking, v. 7, n. 11, 215. [5] D. J. Ives, P. Bayvel and S. J. Savory. Physical layer transmitter and routing optimization to maximize the traffic throughput of a nonlinear optical mesh network. IEEE International Conference on Optical Network Design and Modeling, p. -173, 2. [] X. Chen, Y. Zhong and A. Jukan. Multipath routing in elastic optical networks with distance-adaptive modulation formats. International Conference on Communications (ICC), p , 213. [7] L. Zhang, W. Lu, X. Zhou and Z. Zhu. Dynamic RMSA in spectrumsliced elastic optical networks for high-throughput service provisioning. IEEE International Conference on Computing, Networking and Communications (ICNC), p. 3-34, 213. [] M. Jinno, et. al. Distance-adaptive spectrum resource allocation in spectrum sliced elastic optical path network [topics in optical communications]. IEEE Communications Magazine, v. 4, n., 21. [9] R. Dar, M. Feder, A. Mecozzi and M. Shtaif. Inter-channel nonlinear interference noise in WDM systems: Modeling and mitigation. Journal of Lightwave Technology, v. 33, n. 5, p , 215.