ABSTRACT

In this paper, we contrive a model that underpins the offline Physical Layer Impairment-Routing and Wavelength Assignment (PLI-RWA) issue in translucent networks. We introduce an innovative PLI-Signal Quality Aware RWA (PLI-SQARWA) algorithm that a) guarantees zero blocking due to signal degradation and wavelength contention and b) aims at minimizing the total required number of network components i.e. regenerators and all-optical wavelength converters (AOWCs). Further, in view of reducing the time delay due to optical-electrical-optical (OEO) conversions, we propose a novel electro-optical hybrid translucent node architecture. We show that PLI-SQARWA outperforms a recent heuristic for RWA and regenerator placement (RP) in terms of capital expenditure (CapEx) and time delay; while demonstrating superior blocking performance at all traffic loads. In addition, at high traffic loads, PLI-SQARWA also starts to provision savings on operational expenditure (OpEx). We proceed to the performance comparison of network equipped with the proposed hybrid node and existing translucent and transparent node architectures. The results clearly show that use of the hybrid node incurs less time delay at a similar blocking performance shown by nodes which use OEO conversion for both, regeneration and/or wavelength conversion. The results presented also highlight the significance of equipping the PLI-RWA routing phase with signal quality awareness in order to reduce the network component count and the use of AOWCs to minimize time delay due to OEO conversions.

1. Introduction

Routing and wavelength assignment (RWA) is vital for efficient wavelength division multiplexed (WDM) network design [1]. In recent years, realizing that, de-facto, signal transmission is significantly affected by physical layer impairments (PLIs) inherent in the fiber and that a cross-layer design is requisite in both, a transparent and a translucent WDM network, extensive research has been conducted on the issue of physical layer impairment-RWA (PLI-RWA) [2].

The PLI-RWA approaches delineate rules and strategies for lightpath establishment and primarily aim at minimizing the number of rejected requests due to either capacity or QoT limitations. Existing PLI-RWA algorithms are faced with two cases wherein; a) traffic demands are known a-priori and hence, decisions are taken offline using static PLI-RWA or b) traffic demands arrive in a dynamic fashion and decisions are made online using dynamic PLI-RWA [3]. In literature, investigations on PLI-RWA mainly focus on a) selecting regeneration sites and number of regenerators to be deployed on these sites (regenerator placement (RP) problem) and/or b) given sparse placement of regenerators, selecting which of these regenerators to use (regenerator allocation (RA) problem) [4].

The authors in [5] proposed a RP and constraint-based routing (RP-CBR) approach which aims at limiting regeneration to some network nodes while considering the impact of PLIs on quality of transmission (QoT). The proposed algorithm relies on a topology driven strategy and the result clearly show that RP-CBR is suitable for on-the-fly network operations and significantly decreases the blocking probability.

Authors in [6] proposed an improved version of the RP-CBR algorithm [5]. In the amended algorithm, termed as RP-CBR⁺, different RP combinations are investigated and the first QoT admissible combination providing least number of regenerators is retained as the solution. Results show that for all traffic load values, RP-CBR⁺ presents non-zero blocking while using the same number of regenerators as RP-CBR algorithm.

The Lightpath Establishment with RP (LERP) algorithm [7] finds k-alternate shortest paths (k-SPs) in its first phase and performs RP in the second phase, after estimating bit error rate (BER) of lightpaths comprising of solution of the first phase. LERP simultaneously minimizes the amount of rejected traffic demands and required regenerators. The results clearly underline efficiency of LERP and indicate that the benefits obtained in terms of demand rejection and number of regenerators is due to a larger combinatoric in lightpath RWA and RP.

Authors in [8] proposed the Cross-Optimization for RWA and RP (COR2P) heuristic which minimizes both, the number of required regenerators and regeneration sites. The authors also introduced an original cost function contributing to capital and operational expenditures (CapEx/OpEx) optimization and further, compared COR2P to RP-CBR⁺ [6] and LERP [7]. The results show that for low traffic loads, COR2P does not reject any demands while RP-CBR⁺ presents non-zero blocking; whereas for high traffic loads, COR2P presents a blocking which is 100 times lower than that shown by RP-CBR⁺. Further, for identical scenarios at low or moderate traffic loads, in comparison to LERP, COR2P reduces both, the total number of regenerators and regeneration sites. The results also reveal the major drawbacks of COR2P viz., at high traffic loads, blocking performance maintained by COR2P is at the expense of increased number of required regenerators and at such loads, COR2P’s methodology of QoT-dependent ordering of demands is not very effective.

The aforementioned translucent network design studies are founded on traditional shortest path (SP) routing protocols since, in general, among all the routes, between a pair of nodes, SP algorithm sufficiently achieves the desired performance. However, since traffic characteristics and resource availability are not a part of its routing decisions, SP approach contributes significantly to network congestion due to its single-path routing, and also comes short of provisioning efficient network utilization that may lead to resource blocking, which as opposed to physical-layer blocking, occurs due to rejection of a connection owing to the unavailability of wavelengths. Further, owing to the heterogeneity of real networks and existence of non-uniform PLIs within such networks, SP may not always demonstrate the best signal quality. Hence, it may occur that any other route(s), apart from SP, exhibits better signal quality, thus requiring fewer intermediate regenerators and in turn lowering the overall network cost.

This can be justified by considering the simple topology shown in Figure 1 wherein; a request originates at the source node (N₁) demanding connection till the destination node (N₆). The weight on each link corresponds to the transmission delay and hence, among the different possible routes between N₁ and N₆, route N₁-N₂-N₃-N₆ is the SP. Let all-optical wavelength converters (AOWCs) be provisioned for wavelength contention resolutions and let the signal quality threshold (i.e. Q-factor threshold (Q_th)) be defined as the value below which the signal quality is unacceptable hence, necessitating regeneration. We assume that each link supports three wavelengths denoted as, and, with the wavelengths indicated on each link corresponding to a free (available) wavelength.

It can be observed from the figure that, if the SP is used for serving the request, then the demand will be blocked since wavelength continuity is not possible on link N₃-N₆ due to the unavailability of wavelength. Deploying an AOWC at node N₃ for wavelength conver-

sion (WC) to on one hand will relax the wavelength continuity constraint but on the other hand, will increase the network cost. Further, owing to the PLIs, signal quality falls below the pre-defined threshold value on links N₂-N₃ and N₃-N₆ which necessitates regenerator deployment at nodes N₂ and N₃ respectively, further increasing the network cost. On the other hand, if route N₁-N₄- N₃-N₆ is used, it can be observed that in order to serve the request, the path requires same amount of network components (i.e. two regenerators at nodes N₄ and N₃ respectively and an AOWC for WC to at node N₃) and demonstrates a similar cost as the SP. Further, paths N₁-N₂-N₅-N₆ and N₁-N₄-N₅-N₆ are observed to require lesser number of network components (i.e. one regenerator at nodes N₂ and N₄ respectively and an AOWC for WC to and at nodes N₅ and N₄ respectively) compared to the SP. Hence, these routes are the least cost paths as they demonstrate better signal quality among all other network routes thus, requiring fewer intermediate regenerators. Among the two least cost paths, if SP strategy is now applied in order to choose the candidate route, then path N₁-N₄-N₅-N₆ will be chosen since its total link weight is lower than that of the route N₁-N₂- N₅-N₆. Therefore, outcome of such a routing strategy, which is summarized in Table 1, is a candidate path that requires fewest amounts of regenerators and hence demonstrates least cost. Thus, in addition to simple SP routing, it is judicious to formulate an algorithm that selects candidate routes based on best signal quality.

In this paper, we formulate a framework that corroborates the offline version of PLI-RWA problem in translucent networks where, given a network topology and the estimate of traffic demands, both, the static PLI-RWA and the RP problems are solved jointly.

The contributions of this paper are twofold:

• An innovative PLI-Signal Quality Aware RWA (PLISQARWA) algorithm is introduced that 1) guarantees zero blocking due to signal degradation and wavelength contention, and 2) minimizes the total required network components by a) considering signal quality to route connections over paths requiring fewest numbers of regenerators, and b) maximally using placed regenerators for wavelength conversion (WC) before resorting to AOWCs, which in recent years have been demonstrated to be practical [9].

• An electro-optical hybrid translucent node architecture [10] is proposed in view of a latency efficient technology capable of delivering a cost effective implementation suitable for large scale deployment. The proposal is motivated from the fact that for next generation optical networks, a) OEO conversions will be the primary bottleneck [11], and b) provisioning demands with the least possible network latency will be necessitated [12].

The rest of the paper is organized as follows. In Section 2, we describe the network model and the performance evaluation metric used in our study. Section 3, describes details of the PLI-SQARWA algorithm for translucent network design and further, presents performance comparison results of PLI-SQARWA and COR2P. In Section 4, we present details of the proposed hybrid translucent node architecture. We then proceed to performance comparison of the network equipped with the hybrid node and node architectures existing in literature. Finally, Section 5 presents the conclusion of this study.

2. Network Model

Considering the typical configuration of a translucent network, Figure 2 describes within such a network, the architecture of a generic node equipped with a pool of 3R regenerators. In such architecture, the switching node consists of an optical cross connect (OXC) connected to a regenerator pool linked via transponders. Signals which fulfill the QoT requirements and do not require WC, transit through the node optically without undergoing any OEO conversions. On the other hand, if signal quality does not meet the QoT requirements and/or wavelength continuity constraint is not satisfied, the signal is regenerated and/or wavelength converted (assuming that regenerators also have WC capability) and is then re-injected into the switching fabric to join the next node along its route. We assume fiber links to be deployed using standard non-zero dispersion shifted fibers and erbium-doped fiber amplifiers (EDFAs) to be deployed every span (typically every 80 Km) in order to recover from fiber losses.

Existing studies use BER as the metric to evaluate the

quality of a lightpath as it captures the effects of all impairments. The BER is given as [13]

(1)

where Q represents the Quality-factor (Q-factor) which is given as [13]

(2)

In Equation (2), and representing the mean values of current and, and and representing the variance values for bit “1” and bit “0” respectively are given as [14]

(3)

(4)

(5)

and

(6)

where represents the signal power at the j^th (center) receiver at frequency f_j, S_sp the power spectral density of ASE noise, B₀ the optical filter bandwidth, R₀ the responsivity, e the electron charge, K_B the Boltzmann constant, B_e the electrical bandwidth of the receiver, T the receiver temperature and R_L the load resistance.

In the current study, we consider an Intensity Modulation/Direct Detection (IM/DD) system based on on-off keying (OOK) modulation and the lightpath QoT evaluation is based on the realistic estimation of signal quality considering the simultaneous impact of three effects viz. stimulated Raman scattering (SRS), four wave mixing (FWM) and amplified spontaneous emission (ASE) noise. We use the Q-factor model from our previous work [14] in order to evaluate blocking probability (BP), which is the performance metric used in our study. BP is defined as the probability that a connection cannot be accepted and is given as

(7)

When the receiver Q-factor associated with a connection request is below the threshold, the connection is blocked.

3. Translucent Network Design

In this section we describe the translucent network design problem statement which can be stated as follows:

Given

a)    A network topology;

b)    A set of available wavelengths per fiber link;

c)    The offered traffic comprising of a set of static demands;

d)    An admissible Q-factor threshold value i.e. Q_th.

Objective

Find a candidate route that uses fewest numbers of regenerators and assign wavelength(s) on each of the links along the found lightpath route. For wavelength contention resolutions, ensure the maximal use of WC capability of placed regenerators, before deploying AOWCs, in effect, minimizing the required number of total network components. Consequently, the aim of translucent network design is to satisfy maximum number of connection requests (i.e. obtaining lowest possible blocking) simultaneously reducing the overall network cost.

Subject to

i)   Signal quality constraint: For any network lightpath, at its destination node, the corresponding Q-factor value must not fall below the threshold value.

ii)   Delay constraint: For any network lightpath, the latency incurred must not exceed a given timeout T_max, which is assumed to be user adjustable.

iii)   Minimum cost constraint: For any network lightpath, minimum number of components must be deployed along the route i.e. minimum overall cost must be incurred along the route.

3.1. PLI-Signal Quality Aware RWA (PLI-SQARWA) Algorithm

In this sub-section, we present details of the PLISQARWA algorithm that comprises of three phases. In order to improve the scalability of our approach, we decompose the problem into the RWA and wavelength converter placement (WCP) and the RP sub-problems. The PLI-SQARWA flow-chart is described by diagram in Figure 3.

Phase I. Signal Quality Aware Routing (SQAR) Algorithm

In Phase I, PLI-SQARWA employs the signal quality aware routing (SQAR) algorithm that selects a candidate route requiring fewest numbers of regenerators. The SQAR algorithm comprises of the following three steps:

Step 1. Consider the Q-factor values of individual links and summate them to find the total Q-factor as follows

(8)

where N represents the total number of links between the source and destination, the total Q-factor of a lightpath and Q_x the Q-factor value on x^th link of the lightpath. Using evaluated total Q-factors of all the lightpath routes, find a route termed as “Best Path” that has the “best (highest)” Q-factor and go to Step 2.

Step 2. Evaluate the SPs, considering that apart from finding Best Paths, aim of SQAR algorithm like any other routing algorithm is also to find a route which demonstrates the least distance and go to Step 3.

Step 3. Compare Best Paths with SPs in terms of number of regenerators required for each path by using the RP algorithm detailed in phase III and select the path that uses fewest numbers of regenerators as the candidate route. In case when both, Best Path and SP use the same number of regenerators, prefer SP over the Best path and go to Phase II.

Phase II. Wavelength Assignment (WA) and Wavelength Converter Placement (WCP) Algorithm

In Phase II, the carried lightpaths from phase I are sequentially processed. For each lightpath, the first-fit WA (FF-WA) technique [1] is used to assign wavelengths and when wavelength continuity is not possible, we resort to wavelength converters. It must be noted that in view of reducing the total number of required network components and hence minimizing the overall network cost, in this phase, before deploying any AOWCs to resolve wavelength contentions, our aim is to maximally use the WC capability of regenerators that will be required to be appropriately placed along lightpaths which demonstrate non-admissible QoT. Based on the number of placed regenerator(s), such non-admissible QoT lightpaths will be divided into section(s) and between any two consecutive sections, the existence of a regenerator with WC capability will imply that wavelength continuity for such lightpaths will have to be ensured only on each section. On the other hand, for lightpaths demonstrating admissible QoT, regenerators will not be required to be placed along the lightpath and thus, such lightpaths will consist of only one section (i.e. from source node to destination node).

We assume that all the wavelengths (λ) and nodes (n_i) are numbered consecutively in the order λ₁, λ₂, ···, λ_W and n₁, ···, n_i, ···, n_n for 1 < i < n, respectively, where W is the same maximum available wavelength numbers on each link, n₁ and n_i represent the source and destination nodes of a lightpath route respectively. The WA and WCP process comprises of the following seven steps:

Step 1. Initially, consider 1^st node (n₁) of the candidate route as the source node and go to Step 2.

Step 2. Starting from source node, scan all the fiber links along the path and check if any common wavelength is available along all the links of the route. If any such wavelength exists, assign one with the smallest index (say for e.g. λ₁) to the lightpath. Starting from the source node, check Q-factor of the route and if the route demonstrates Q-factor below the pre-defined threshold value, go to Phase III in order to place regenerators along the route appropriately, otherwise terminate. Else, go to Step 3.

Step 3. If no common wavelength exists along the lightpath route, find the node from where a common wavelength is not available, considering that a common wavelength could be available on all the links up to an intermediate node n_i, for 1 < i < n, along the route or starting from the source node itself, no common wavelength could be available along the route. Assign current node as the source node and go to Step 4.

Step 4. Evaluate Q-factor at the next node. If Q-factor value at next node is below the threshold value, place a regenerator at the current node (i.e. source node) and in addition to regeneration, use the placed regenerator for WC to λ₁ and go to Step 6. Else, go to Step 5.

Step 5. If Q-factor value at next node is above the threshold value, a regenerator is not required at the current node (i.e. source node). Hence, in order to satisfy the WC requirement, deploy an AOWC at the current node for WC to λ₁ and go to Step 6.

Step 6. Assign next node along the lightpath route as the source node and go to Step 7.

Step 7. Check if the source node is the destination node. If yes, then terminate, else go to Step 2.

Phase III. Regenerator Placement (RP) Algorithm

In this phase, PLI-SQARWA applies an efficient RP algorithm that minimizes number of placed regenerators over a lightpath. The RP algorithm comprises of the following seven steps:

Step 1. Initially, consider source node i.e. 1^st node (n₁) of the candidate route as current node and go to Step 2.

Step 2. Move to next node along the route and evaluate Q-factor at this node. Continue and go to Step 3.

Step 3. If evaluated Q-factor at the current node is higher than or equal to the pre-defined threshold value i.e. if a regenerator is not required, then go to Step 6. Else go to Step 4.

Step 4. If current node demonstrates Q-factor less than the pre-defined threshold value, place a regenerator at the preceding node and go to Step 5.

Step 5. Assign the node where the regenerator is placed as the source node and go to Step 6.

Step 6. Assign next node as current node. Check if current node is the destination node and go to Step 7.

Step 7. If current node is the destination node, then terminate, else, evaluate Q-factor at this node and go to Step 3.

3.2. Performance Comparison of PLI-SQARWA and COR2P

In this sub-section, we first precise our simulation environment and characteristics and then present the performance comparison results between the two algorithms: PLI-SQARWA and COR2P.

3.2.1. Simulation Assumptions

We investigate the 18 node NSFNET network as depicted in Figure 4. The network is assumed to have 16 wavelengths per fiber link with 50 GHz channel spacing and each carrying a capacity of 10 Gbps.

In the current study, Q-factor threshold takes the value of 6 which corresponds to a BER of 10^–9. We consider permanent lightpath demands (PLDs) which are offline requests that consist of pre-known connection demands with data rate equal to full capacity of the wavelength channel and are thus established through a full lightpath. For the simulations, other transmission system parameters are adopted from our previous studies [10,14].

3.2.2. Simulation Results

This sub-section analyses the performance of PLISQARWA by means of comparison with COR2P algorithm. The COR2P algorithm has been executed for range of different values of a set of specific parameters that adjust in order to determine how the solution space is explored heuristically. The results shown in this paper correspond to those parameters which provided the best performances and their values are reproduced so as to allow the results to be repeatable: 1) for computation of k-SPs associated to each demand, the value of k is set to 5; 2) COR2P cost function is simplified by setting the ratio C_C/C_O to 1, where C_O is the unitary OpEx cost corresponding to the cost for managing a single active regenerator at a node and C_C is the unitary CapEx cost corresponding to carrier’s investment for installing a regeneration pool at a node; and 3) simulations cover 6 traffic loads that range from 100 to 600 connection demands and for each load, 10 static traffic matrices are generated stochastically according to uniform distribution. Thus, the presented results are average values of ten simulation runs.

In order to appropriate the investigation of number of non-accepted requests by both algorithms, for all traffic load values and comparison of the two algorithms, we chose simulation parameters such that cases admitting a solu-

tion did not guarantee 0% of resource blocking. In Figure 5, variation of BP with traffic load has been plotted for both algorithms. It can be observed from the figure that:

1)    For low traffic loads, both, COR2P and proposed PLI-SQARWA show zero BP. This can be attributed to the fact that for both the algorithms:

i)      Regeneration and WC is provisioned which eradicates blocking due to signal degradation and wavelength contention respectively, and

ii)    Owing to low loads, the network is not overloaded and ample resources are available.

2)    For intermediate and high traffic loads, resource blocking starts to occur for both the algorithms. It can be observed that COR2P shows higher resource blocking compared to the proposed PLI-SQARWA algorithm. This can be attributed to the fact that:

i)      For COR2P, as the load increases, owing to the use of k-SP routing, higher network overloading occurs which results in higher resource blocking, and

ii)    For PLI-SQARWA, rather than the SPs, majority of the Best Paths are selected as the candidate routes which leads to larger resources being available even at higher traffic loads. The low resource blocking observed in the case of PLISQARWA is due to the selection of few SP routes as candidate routes. It can also be observed from the figure that if only Best Paths are used for routing, then zero resource blocking is achieved.

It can also be deduced from Figure 5 that for low loads, both the algorithms accept all demands whereas, between moderate and high loads, both the algorithms are able to maintain acceptable blocking.

Table 2 presents the average total number of required components (i.e., regenerators + wavelength converters) for both the algorithms at different load values. In the