ey of˚ultr‑dense wavelength switched network …...phot nw communications 1 3...

13
Vol.:(0123456789) 1 3 Photonic Network Communications https://doi.org/10.1007/s11107-018-00824-w INVITED PAPER Exploiting efficiency of ultra‑dense wavelength switched network for carrying metro network traffic Ya Zhang 1  · Xu Zhou 2  · Ning Deng 2  · Sanjay K. Bose 3  · Gangxiang Shen 1 Received: 24 June 2018 / Accepted: 17 December 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2019 Abstract In the 5G era, metro optical networks would need to meet more stringent quality of service requirements. They would have to operate with high spectral efficiency but with low latency and low power consumption. For this, we introduced a new paradigm based on the elastic optical network, called ultra-dense wavelength switched network (UD-WSN) in Zhang et al. (Proceedings of the Asia Communications and Photonics Conference, 2016), Shen et al. (EEE Commun Mag 56(2):189–195, 2017), and Zhou et al. (IEEE/OSA J Lightw Technol 35(11):2063–2069, 2016). UD-WSN was verified to be efficient in terms of system cost, spectrum efficiency, power consumption, and service connection latency when compared to other popular architectures, such as pure optical transport networks (OTN) and conventional OTN over dense wavelength division multi- plexing networks. This motivates us to explore further enhancements to UD-WSN in this paper for even better performance. Specifically, we consider a UD-WSN architecture without aggregation OTN switches to evaluate how the system cost can be reduced further by trading off the system performance. We also propose to implement partial OTN switching within the UD-WSN to exploit the benefit of traffic grooming to lower the system cost even further. Finally, we also implement spec- trum defragmentation to improve the spectrum utilization of the system. These schemes are studied through simulations to verify their effectiveness. Keywords Ultra-dense wavelength switched network (UD-WSN) · Metro optical network · Partial OTN switching · Spectrum defragmentation 1 Introduction The popularity of 4 K ultra-high-definition video, virtual realization (VR), augmented reality (AR), and cloud-based services has been increasing the Internet traffic dramati- cally in metropolitan areas. This puts high demands on the capacity of the current metro optical network and limits its bandwidth allocation flexibility. In the backbone networks, implemented primarily using high-speed optical channels, the data rates of major client services in today’s metro net- work may be at rates as high as 1, 2.5, or even 10 Gb/s, and this situation is likely to last in the foreseeable future [2]. For the metro optical network, pure OTN and the OTN over dense wavelength division multiplexing (DWDM) net- works [3] are the two most popular architectures in use today that enable efficient capacity utilization through electronic traffic grooming. In the pure OTN network, the optical layer consists of DWDM transmission links that directly inter- connect OTN switches through optical multiplexers and de- multiplexers. Pure OTN performs optical–electronic–optical Part of the work was presented in [1]. This work was jointly supported by the Project (YBN2016050116) with Huawei Technologies Co, National Natural Science Foundation of China (NSFC) (61671313), and the Science and Technology Achievement Transformation Project of Jiangsu Province, PR China (BA2016123). * Gangxiang Shen [email protected] 1 School of Electronic and Information Engineering, Soochow University, Suzhou 215006, Jiangsu Province, People’s Republic of China 2 Networks Research Department, Huawei Technologies Co., Ltd, Shenzhen, People’s Republic of China 3 Department of Electronics and Electrical Engineering, Indian Institute of Technology Guwahati, Guwahati, India

Upload: others

Post on 30-Aug-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Vol.:(0123456789)1 3

Photonic Network Communications https://doi.org/10.1007/s11107-018-00824-w

INVITED PAPER

Exploiting efficiency of ultra‑dense wavelength switched network for carrying metro network traffic

Ya Zhang1 · Xu Zhou2 · Ning Deng2 · Sanjay K. Bose3 · Gangxiang Shen1

Received: 24 June 2018 / Accepted: 17 December 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2019

AbstractIn the 5G era, metro optical networks would need to meet more stringent quality of service requirements. They would have to operate with high spectral efficiency but with low latency and low power consumption. For this, we introduced a new paradigm based on the elastic optical network, called ultra-dense wavelength switched network (UD-WSN) in Zhang et al. (Proceedings of the Asia Communications and Photonics Conference, 2016), Shen et al. (EEE Commun Mag 56(2):189–195, 2017), and Zhou et al. (IEEE/OSA J Lightw Technol 35(11):2063–2069, 2016). UD-WSN was verified to be efficient in terms of system cost, spectrum efficiency, power consumption, and service connection latency when compared to other popular architectures, such as pure optical transport networks (OTN) and conventional OTN over dense wavelength division multi-plexing networks. This motivates us to explore further enhancements to UD-WSN in this paper for even better performance. Specifically, we consider a UD-WSN architecture without aggregation OTN switches to evaluate how the system cost can be reduced further by trading off the system performance. We also propose to implement partial OTN switching within the UD-WSN to exploit the benefit of traffic grooming to lower the system cost even further. Finally, we also implement spec-trum defragmentation to improve the spectrum utilization of the system. These schemes are studied through simulations to verify their effectiveness.

Keywords Ultra-dense wavelength switched network (UD-WSN) · Metro optical network · Partial OTN switching · Spectrum defragmentation

1 Introduction

The popularity of 4 K ultra-high-definition video, virtual realization (VR), augmented reality (AR), and cloud-based services has been increasing the Internet traffic dramati-cally in metropolitan areas. This puts high demands on the capacity of the current metro optical network and limits its bandwidth allocation flexibility. In the backbone networks, implemented primarily using high-speed optical channels, the data rates of major client services in today’s metro net-work may be at rates as high as 1, 2.5, or even 10 Gb/s, and this situation is likely to last in the foreseeable future [2].

For the metro optical network, pure OTN and the OTN over dense wavelength division multiplexing (DWDM) net-works [3] are the two most popular architectures in use today that enable efficient capacity utilization through electronic traffic grooming. In the pure OTN network, the optical layer consists of DWDM transmission links that directly inter-connect OTN switches through optical multiplexers and de-multiplexers. Pure OTN performs optical–electronic–optical

Part of the work was presented in [1]. This work was jointly supported by the Project (YBN2016050116) with Huawei Technologies Co, National Natural Science Foundation of China (NSFC) (61671313), and the Science and Technology Achievement Transformation Project of Jiangsu Province, PR China (BA2016123).

* Gangxiang Shen [email protected]

1 School of Electronic and Information Engineering, Soochow University, Suzhou 215006, Jiangsu Province, People’s Republic of China

2 Networks Research Department, Huawei Technologies Co., Ltd, Shenzhen, People’s Republic of China

3 Department of Electronics and Electrical Engineering, Indian Institute of Technology Guwahati, Guwahati, India

Page 2: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

(OEO) conversions at each intermediate node of a service connection, which leads to high system costs, high power consumption, and long service connection latency.

As a more efficient architecture, the OTN over DWDM network reduces the OEO conversions by introducing optical bypass in the optical layer where reconfigurable optical add/drop multiplexers (ROADMs) are deployed in addition to an OTN switch at each network node. The optical bypass can eliminate unnecessary OEO conversions by establishing a long transparent lightpath, thereby reducing the network’s power consumption, its service connection latency, and the overall system cost. However, the OTN over DWDM net-work is still a type of OTN-based network, where the spec-trum granularity used is based on the conventional DWDM technology (e.g., 50 GHz [4]) and would therefore be very coarse for low-speed service connections (e.g., GE/10GE services). As a result, a large number of OEO conversions are still required for traffic grooming to ensure efficient capacity utilization. In addition to higher power consump-tion, these OEO conversions would lead to extra latency, which would not be desirable for services requiring low latency.

To tackle these difficulties of the conventional OTN-based architectures, we proposed the ultra-dense wavelength switched network (UD-WSN) architecture in [5–7], particu-larly in the context of metro optical networks. UD-WSN is a special case of elastic optical networks [8], inspired by the ultra-dense WDM-PON (UDWDM-PON) [9] architecture in access networks. The unique feature of UD-WSN is that it allows spectrum granularity to be even finer than 12.5 GHz so as to realize the concept of “one wavelength per user.” In Zhou et al. [7], we physically verified this architecture. In [5, 6], based on this basic architecture, we analyzed its advan-tage from the performance perspectives of system cost, spec-trum efficiency, power consumption, and service connection latency in comparison with standard OTN-based networks.

In this paper, we further exploit potential opportunities for enhancing the performance of UD-WSN. In particular, to reduce the system cost, we consider a UD-WSN architecture without aggregation OTN switches to see how the system cost can be reduced by trading off system performance. To take advantage of the traffic grooming capability of an OTN switch while keeping the system cost low, we consider a partial OTN switching scenario for UD-WSN by properly distributing the traffic grooming capability in the network. Finally, we note that due to its usage of even finer spectrum granularities, spectrum fragmentations can become even worse in UD-WSN. We reduce the impact of this and further enhance network spectrum utilization by carrying out spec-trum defragmentation. These techniques are studied through simulations to verify their effectiveness.

The rest of the paper is organized as follows. In Sect. 2, we review the network architecture of UD-WSN and

introduce the issues that are subsequently examined. In Sect. 3, we introduce the UD-WSN architecture without aggregation OTN switches and evaluate its performance in comparison with a regular UD-WSN. In Sect. 4, we study partial OTN switching in UD-WSN. In Sect. 5, we evaluate the benefit of spectrum defragmentation in a UD-WSN. Case studies and performance analyses are presented in each sec-tion. Section 6 concludes the paper.

2 Ultra‑dense wavelength switched networks

The metro network bridges the backbone network and the access network as shown in Fig. 1. Many sub-1G/1G/10G service connections dominate today’s metro networks. For efficient service provisioning, electronic traffic grooming is implemented in today’s OTN-based networks with the 50-GHz spectrum granularity. To avoid extensive OEO conversions in the conventional OTN-based networks, UD-WSN combines the technical merits of elastic optical net-work (EON) and UDWDM-PON by adopting ultra-dense spectrum granularities (e.g., 5 GHz or 6.25 GHz) and ena-bles the “one wavelength per user” provisioning mode by establishing direct lightpath connections for users. Using such direct lightpaths would be obviously advantageous, as it would minimize OEO conversions. This would in turn reduce service connection latencies and the system’s power consumption.

2.1 Spectrum operation

The majority of service connections in the metro network are 1G, 2.5G, and 10G services, whose bandwidths would be much smaller than the capacity of a 50-GHz optical channel in a DWDM network. To accommodate these services effi-ciently, UD-WSN employs a much smaller spectrum granu-larity that is just enough to accommodate the bandwidth required by the metro service connections. Here, we consider four candidate granularities (e.g., 5 GHz, 6.25 GHz, 10 GHz, and 12.5 GHz) with a 2-bit/s/Hz spectrum efficiency using

Long-haul

transport

AccessAggregation Core

Backbone

EON

UDWDM-PON UD-WSN

Metro

metro core node

metro access node

OLT

Spliter

ONU

Fig. 1 Backbone, metro, and access optical networks

Page 3: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

QPSK modulation without polarization multiplexing. To illustrate the spectral benefit of UD-WSN, Fig. 2 compares the spectrum usage for the three different types of networks, including the DWDM network, EON, and UD-WSN, when carrying different types of services.

In a DWDM network (see Fig. 2a), spectrum granularity is typically 50 GHz according to the ITU-T standard [4]. An optical channel that occupies an entire DWDM wavelength is spectrally efficient to carry a 100-Gb/s service connection, while it would be inefficient to carry a 10-Gb/s service con-nection, even though the latter is currently dominant in the metro network. To improve spectrum utilization, the DWDM network needs to work with the OTN layer. The OTN over DWDM network can achieve high spectrum utilization by introducing the traffic grooming capability in the electronic layer. This would, however, be at the added cost of extra OTN equipment. This leads to a high system cost and high

power consumption, along with long service connection latencies.

EON shrinks the spectrum granularity to 12.5 GHz (see Fig. 2b) to mitigate the low spectrum utilization disad-vantage of the DWDM systems when carrying low-speed service connections. Moreover, EON can combine multi-ple spectrally neighboring frequency slots (FSs) to form a super-channel so as to provision a high-bandwidth service connection. However, this granularity still seems to be ineffi-cient for metro network services since these are largely over-provisioned to use a 12.5-GHz FS even when carrying only a 1-Gb/s or 10-Gb/s service connection. In view of this, UD-WSN further shrinks the spectrum granularity to an even finer one as shown in Fig. 2c, e.g., 5 GHz. As a result, more optical channels can be provisioned in the C-band, and these channels are efficient to provide just sufficient bandwidth for many metro network service connections. Moreover, these channels are switched in the optical domain, thereby avoid-ing OEO conversions. On the other hand, UD-WSN inherits the flex-bandwidth feature from EON, so it can also support super-channels with high bandwidth (e.g., > 100 Gb/s) by combining multiple spectrally neighboring FSs.

To enable UD-WSN, wavelength selective switches (WSSs) supporting ultra-fine switching granularity (e.g., 5 GHz) are the most critical components. Currently, new fine resolution photonic spectral processors supporting sub-1-GHz optical resolution and 50-MHz spectral addressabil-ity using a waveguide grating router with permanent phase trimming have been experimentally demonstrated [10, 11]. WSSs supporting 10-GHz filtering granularity and 1-GHz resolution are also commercially available from Lumentum, and it has been technically possible to support a WSS with an even finer FS granularity (e.g., 5 GHz).

2.2 Network architecture

Figure 3 shows the general network architecture of UD-WSN, which consists of core and aggregation segments. The core segment is the backbone part, while the aggrega-tion segment is an access part, aggregating traffic demands from distributed access nodes [6]. The access nodes con-nect to users, e.g., wireless base stations (BSs). We call the switching nodes in the core segment metro core (MC) nodes and those in the aggregation segment metro aggregation or access (MA) nodes.

In the core segment, UD-WSN employs coherent transceivers with 50-GHz optical bandwidth for direct end-to-end lightpath establishment between MC nodes. For flexible bandwidth allocation, the 50-GHz optical bandwidth of an optical transmitter is sliced into multiple optical sub-channels (or FSs), each with a limited gran-ularity (e.g., 5 GHz). The traffic flows from one source node to different destination nodes are mapped onto these

(a) DWDM

(b) EON

(c)UD-WSN

10Gb/s50 GHz

100Gb/s50 GHz

10Gb/s50 GHz

75 GHz100Gb/s

10Gb/s20Gb/s

50 GHz

50 GHz100Gb/s

Fig. 2 Spectrum usage in DWDM, EON, and UD-WSN systems [5]

Page 4: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

optical sub-channels, which are optically switched by the WSSs that support ultra-dense spectrum granularity (e.g., 5 GHz) and can be added/dropped at different MC nodes with one or multiple FSs. In the opposite direction, an optical receiver can receive multiple optical sub-channels from different MC nodes as long as they do not spectrally overlap. The architecture in this segment is generally symmetric because the traffic flows between a pair of MC nodes are typically bidirectionally symmetric and the same coherent transceivers are employed for the two end nodes.

In the aggregation segment, there are high traffic demands between an MC node and multiple MA nodes with small traffic demands between the MA nodes. UD-WSN employs an asymmetric transmission architecture to deploy high-capacity coherent transceivers at the MC nodes and low-cost non-coherent intensity-modulated and directly detected (IM-DD) transceivers at the MA nodes as shown in Fig. 3. The reason for adopting such an asym-metric architecture is the typical asymmetry of the bidi-rectional traffic flows between pairs of MC and MA nodes and the low-cost requirement for an MA node. At each MC node, a single coherent transceiver generates multiple optical sub-channels within a 50-GHz range, which are switched by the WSSs and dropped at different destination MA nodes. Each MA node employs a low-cost IM-DD transceiver to receive its corresponding sub-channels fil-tered by a blocker (for a chain topology) or an AWG (for a tree topology) supporting ultra-dense spectrum granular-ity. We refer to them as UD-blockers or UD-AWGs hereaf-ter. In the upstream direction, optical sub-channels gener-ated by different MA nodes are multiplexed and forwarded to a common coherent transceiver at the MC node subject to the condition that they are not spectrally overlapping and lie within the 50-GHz range acceptable by the coher-ent transceiver at the MC node.

2.3 Summary

In the previous part, we have introduced the motivation, the spectrum operation, and the general network architecture for UD-WSN. In [7], we have verified its possibility from the perspective of physical system realization. In [5, 6], we also conducted techno-economic analysis through simula-tions to verify its advantages from the perspectives of system cost, spectrum efficiency, power consumption, and service connection latency in comparison with the conventional OTN-based metro network technologies. In this study, we will further exploit the potential efficiency of UD-WSN for carrying metro network services by adopting new network architectures and efficient network operation strategies. Specifically, we will consider UD-WSN without aggrega-tion OTN switches for lower system cost and UD-WSN with partial OTN switching for better capacity utilization. We will also carry out spectrum defragmentation for UD-WSN to enhance its spectrum utilization.

3 UD‑WSN without aggregation OTN switches

In Fig. 3, an OTN switch is deployed to aggregate diverse low-speed traffic demands for the corresponding aggregation chain. The OTN switches are generally the most expensive items in the whole metro network. To reduce the system cost, we remove the OTN switches at the intermediate MC nodes (e.g., node C in Fig. 4) to directly establish lightpaths between core MC nodes and MA nodes, bypassing the inter-mediate MC nodes. The removal of an OTN switch at an intermediate MC node can reduce the system cost, power consumption, and service connection latency. However, a disadvantage of this would be that, without the traffic groom-ing capability at the intermediate MC node, each service of small granularity (e.g., GE) would need to be assigned

50 GHz

50 GHz

Coherent transceiver

Coherent transceiver

IM/DD transceiver

IM/DD transceiver

Coherent transceiver

50 GHz

Symmetric architecture

Asymmetric architecture

EML

ASIC

PD

PDM-IQ-Mod

ASIC

(AD, DA, DSP)

ICR

Laser

MC: metro coreMA: metro accessPDM-IQM: PDM IQ modulatorICR: integrated coherent receiverEML: externally modulated laserPD: photo detector

MC node

MA node

IM/DD transceiver

Core segment

Aggregation segment

Fig. 3 UD-WSN architecture [6]

Backbone MC node MC node MA node

A

B

C

D

S

A1 A2 A3 A4

B1 B2 B3 B4

C1 C2 C3 C4

D1 D2 D3 D4

Fig. 4 Architecture without aggregation OTN switches

Page 5: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

with an entire optical channel (e.g., with 5-GHz bandwidth), which would degrade the network’s spectrum utilization. Thus, there will be a trade-off between the parameters of system cost, power consumption, service connection latency, and spectrum utilization. In this section, we aim to evaluate this trade-off.

3.1 Routing and spectrum assignment

For performance evaluation of the proposed new network architecture, we consider its routing and spectrum assign-ment (RSA) problem under both static and dynamic traffic demands. We employ the spectrum window plane (SWP)-based heuristic algorithm [12, 13] for the performance eval-uation. Under static traffic demand, the main steps of the SWP-based algorithm are as follows.

SWP-based RSA algorithm

Step 1 Get the first request from a traffic demand list and calculate the number of FSs it requires

Step 2 Generate corresponding SWPs and find a suitable route from the lowest to the highest indexed SWPs with the minimum-hop strategy, and assign corresponding spectra to a lightpath along the route thus found

Step 3 Remove the served request from the demand list; if the demand list is empty, terminate the RSA process; other-wise, go to Step 1

Step 4 Calculate the total system cost and the total power consump-tion of the whole network

In Step 1, assume that each source–destination pair requests a bandwidth R (in units of Gb/s). We can calculate the number of FSs required by using F = R/(B·SE), where B is the bandwidth (in units of GHz) provided by each FS, and SE is the spectrum efficiency of the selected modulation for-mat (in units of bit/s/Hz). Here, QPSK without polarization multiplexing (PM) is employed, and therefore, the spectrum efficiency SE is 2 bit/s/Hz. In Step 2, we generate corre-sponding SWPs based on the number of FSs required. The detailed steps of generating the SWPs can be referred to in [12]. For each SWP, we employ the shortest path algorithm to find a suitable route. Finally, we choose the one that has the minimum number of hops and assign the corresponding spectra to a lightpath along the selected route. In Step 3, we continue applying the same process until all the service connections are provisioned. In Step 4, we calculate the total system cost and power consumption of the whole network.

Since the RSA performance was found to be related to the order in which the lightpath connections are established, as in [12] we implement a multi-iteration process to evaluate the performance for multiple shuffled demand sequences and then choose the one with the best performance.

The above SWP algorithm can also be employed to provi-sion dynamic lightpath services. Here, the traffic demands

are assumed to arrive and depart randomly and a demand request can be blocked if no suitable network resources can be found when it arrives.

3.2 Performance

We evaluate the performance of the new architecture with-out aggregation OTN switches based on a real-world metro network topology as depicted in Fig. 5. Note that this topol-ogy was derived from a real one of a network carrier, but due to commercial confidentiality issues, we cannot provide detailed information of this carrier. The network consists of two segments, a core segment and an aggregation seg-ment. Among all the MC nodes, there are two backbone MC nodes, which aggregate the traffic demands from the other nodes and function as gateways to the backbone network. The maximum distance between the MC nodes and the MA nodes is less than 300 km. Asymmetric and symmetric architectures are considered, respectively, for the aggrega-tion and core segments, which are highlighted as a chain and a ring. For a regular UD-WSN, the traffic demands are transmitted from all dispersed MA nodes to their associ-ated MC nodes which further forward the aggregated traffic demands (together with other local traffic demands) to the two backbone MC nodes. There is an OTN switch deployed at each intersecting node between the core segment and the aggregation segment (e.g., node 15 in Fig. 5). In contrast, for a UD-WSN without aggregation OTN switches, the OTN switch at each intersecting MC node is removed and optical sub-channels are directly established between the core MC node and MA nodes. In the aggregation segment, the propor-tion of the traffic demands of 1G, 10G, and 40G services is 3:6:1, and in the core segment, the proportion of the traffic demands of 10G, 40G, and 100G services is 8:1:1. Note that these ratios are from a real network equipment vendor and its customers. Due to commercial confidentiality issues, we

10

11

12

1314

15

16

17

18

19

20

21222324

1 23

4

56

89

26

27

28

29

30

31

25

35

34

33

32

39

38

37

36

43424140 464544

494847

5250

51

54 53

58

55

56

5761

60

59

65

64

62

63

Aggregation segment

Core segment MC node

MA node

Backbone MC node

0

7

Fig. 5 Test topology

Page 6: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

cannot provide detailed information on the exact prices and traffic demand volumes.

Table 1 gives the relative costs of network devices. We assume that the cost of the WSS that supports the switching granularity of 12.5 GHz, 10 GHz, 6.25 GHz, and 5 GHz is 1:1.1:1.2:1.3. The spectrum in each fiber is assumed to be 4 THz, which just supports 80 DWDM wavelengths with 50-GHz spacing. For static traffic demand, we shuffle an initial demand list 100 times to form shuffled demand sequences, for each of which we run the RSA heuristic algo-rithms and select the one with the best design performance (i.e., the minimum total cost of the network) as the final solution.

For dynamic traffic demand, we assume that the ser-vice connections arrive according to a Poisson distribu-tion and their holding times follow a negative exponential distribution. A total of 106 arrival requests were simulated to calculate the BBP, which is defined as the ratio of the total amount of blocked bandwidth to the total amount of requested bandwidth.

Figure 6 shows the simulation results of different network architectures from the perspectives of system cost, power consumption, and BBP. Here, we mainly present the results for spectrum granularities of 5 GHz and 12.5 GHz. The leg-ends “UD-WSN” and “w/o aggregation OTN” correspond to the cases of regular UD-WSN and UD-WSN without aggre-gation OTN switches, respectively. Based on the results in Fig. 6, we have the following key observations.

We see that the network without aggregation OTN switches has lower system cost and power consumption (see Fig. 6a–e). Its cost is 15.7% and 25.6% lower than the regular UD-WSN for the 12.5-GHz and 5-GHz spectrum granularities, respectively. Similarly, the power consumption of the network without aggregation OTN switches is 22.6% and 29.7% lower than the regular UD-WSN, respectively, for the two spectrum granularities. These are reasonable because the network without aggregation OTN switches establishes direct end-to-end lightpaths between MA nodes and the backbone MC nodes, which greatly reduces the num-ber of OTN switches and coherent transceivers required at the intersecting MC node (e.g., node 15 in Fig. 5), thereby significantly reducing the overall system cost and power consumption. However, as indicated before, because each

Table 1 Relative cost and power consumption of network devices

Network device Cost (unit) Power consumption (unit)

100G coherent TRx 600 3610G IM-DD TRx 60 525G IM-DD TRx 90 7.5WSS (50 GHz) 470 4.5WSS (12.5 GHz) 540 4.5EDFA 80 3.2UD-blocker 2 0UD-AWG 16 0OTN switch board 1200 72

(a) Cost (5 GHz) (b) Power consumption (5 GHz) (c) BBP (5 GHz)

(d) Cost (12.5 GHz) (e) Power consumption (12.5 GHz) (f) BBP (12.5 GHz)

1.3E+05

1.7E+05

2.1E+05

2.5E+05

2.9E+05

3.3E+05

100 200 300 400 500 600

Number of service connections

Cum

ulat

ive

CapE

x 15.7%UD-WSNw/o aggregation OTN

3.5E+03

6.5E+03

9.5E+03

1.3E+04

1.6E+04

100 200 300 400 500 600

Number of service connections

Pow

er c

onsu

mpt

ion

(W)

22.6%UD-WSNw/o aggregation OTN

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

12 14 16 18 20 22

Traffic load per node pair (Erlang)

Band

wid

th b

lock

ing

prob

abili

ty

UD-WSNw/o aggregation OTN

1.3E+05

1.9E+05

2.5E+05

3.1E+05

3.7E+05

4.3E+05

100 200 300 400 500 600

Number of service connections

Cum

ulat

ive

CapE

x 25.6%UD-WSNw/o aggregation OTN

4.0E+03

8.0E+03

1.2E+04

1.6E+04

2.0E+04

2.4E+04

100 200 300 400 500 600

Number of service connections

Pow

er c

onsu

mpt

ion

(W)

29.7%

UD-WSNw/o aggregation OTN

1.0E-02

1.0E-01

1.0E+00

10 12 14 16 18 20 22

Traffic load per node pair (Erlang)

Band

wid

th b

lock

ing

prob

abili

ty

UD-WSNw/o aggregation OTN

Fig. 6 Simulation results of different network architectures

Page 7: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

low-speed service (e.g., GE) needs to occupy the entire opti-cal sub-channel, this will cause severe spectrum wastage due to the lack of traffic grooming capability. As a result, the BBP of the network without aggregation OTN switches is higher as shown in Fig. 6c and f.

In summary, we may implement a UD-WSN in different flavors according to the system performance requirement and cost considerations. If we wish to design a UD-WSN with a lower cost and a reasonable good BBP performance, the network without aggregation OTN switches can be deployed.

4 UD‑WSN with partial OTN switching

Low-speed service connections (e.g., GE/10GE services) still dominate today’s metro networks, and this situation will probably not change in the near future. UD-WSN enables a spectrum granularity as small as 5 GHz. However, it is still not spectrally fine enough to efficiently accommodate the small metro services like GE when using a modulation format such as QPSK. To further improve the spectrum uti-lization of UD-WSN for carrying low-speed metro service connections, the FS granularity should be shrunk so that the low-rate services can just fit in the bandwidth. However, due to the physical limitation on optical components (e.g., the minimum spectrum resolution of a WSS) and the cost-performance ratio of the system, it is not practical and eco-nomical to do this. Traffic grooming [14–17] is an effective approach to improve the network overall spectrum utiliza-tion by allowing multiple low-speed service connections to be multiplexed onto the high-speed optical channels in the electrical domain. In order to carry low-speed service con-nections efficiently, we propose to deploy a partial set of OTN switches in UD-WSN to realize OTN switching [18, 19] and sparse traffic grooming [20–22] to avoid a significant increase in system cost and power consumption.

Figure 7 shows an example of the partial OTN switch-ing scenario. In this four-node network, nodes B and D are deployed with the OTN switches, while nodes A and C are deployed only as ROADMs. In this scenario, all the lightpaths that traverse an intermediate OTN switching node experience OEO conversions and electrical traffic grooming as provided by the OTN switch. The low-speed (e.g., 1G/2.5G) services whose bandwidths are much lower than the capacity of a FS (e.g., 5 GHz) can be electrically groomed onto a common lightpath for efficient capacity utilization. For example, S1 and S4 are assumed to be GE services and they are from different source nodes A and B. These two services can be electrically groomed at node B and transmitted along the same lightpath to node D, where only one FS is needed on link C–D. In contrast, because node C is a pure ROADM, its service S2 cannot be groomed

with S1 and S4 at node B though the FS carrying these two services still has remaining capacity to accommodate S2. Instead, a direct lightpath bypassing node B needs to be set up between the two nodes and the service is directly added and dropped via an add/drop port pair at nodes A and C.

4.1 UD‑WSN with sparse traffic grooming

For the partial OTN switching, we first need to decide which set of nodes should be deployed with OTN switches. Here, we consider two strategies, i.e., Highest Node Degree First (HDF) and Highest Probability Traversed by the Shortest Path First (HTF). HDF means that the node with the high-est nodal degree is first deployed with an OTN switch. HTF means that the node with the highest probability of being traversed by the shortest paths between all the node pairs is first deployed with an OTN switch.

For performance evaluation of the proposed partial OTN switching, we develop an efficient auxiliary graph (AG)-based multilayer traffic grooming algorithm for both static and dynamic traffic demand scenarios. The main steps of this algorithm are presented as follows.

AG-based multilayer traffic grooming algorithm

Step 1 For a certain service connection, calculate the number of FSs required

Step 2 Construct a corresponding auxiliary graph (AG) and set weights for different edges in the AG

Step 3 Find a shortest route (with the smallest sum weight) based on the AG

Step 4 Assign corresponding network resources along the route to provision the service connection

As before, in Step 1 we calculate the number of FSs required by using F = R∕(B ⋅ SE) . In Step 2, we construct an AG based on the current network spectrum usage and light-paths established previously. Figure 8a shows an example of physical topology, and Fig. 8b shows its corresponding AG built from the physical topology. In the AG, the nodes with the same name (regardless of the superscripts) correspond to the same physical node in the physical topology (e.g.,

DropNode A Node B Node DNode C

OTN Switch

Client ports

ROADM

S2

S3 S4 1S 4S

AddS1S2S3

S1+S4

Fig. 7 An example of partial OTN switching scenario

Page 8: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

nodes C1 and C2 correspond to the same node C). There are three types of edges in the AG, including lightpath edge, transponder edge, and fiber edge. We set the weights of the different types of edges as in Table 2

The fiber edge corresponds to a physical link in the physi-cal topology, which exists if the corresponding fiber link can provide enough spectrum resource for the establishment of the service connection. The weight of each fiber edge is set to 1 + 1∕W2 , where W is the total number of available spectrum windows (SWs) [12]. Here, SW is defined as a set of spectrally contiguous FSs. Assume that each fiber link contains 29 FSs and 5 FSs are required by a service con-nection, then a total of 25 SWs can be generated as shown in Fig. 9. We assume that the shaded FSs are occupied, and therefore, only 7 SWs are available for spectrum allocation implying that W should be 7. The lightpath edge corresponds to a lightpath previously established between a pair of nodes, which can be used to establish the new service connection if

it still has sufficient remaining capacity. The weight of each lightpath edge is set to 0.05·H, where H is the total number of physical link hops traversed by the lightpath. With a factor of 0.05, the weight of the lightpath edge is set to be much smaller than that of the fiber edge, which helps to maximize electrical traffic grooming and therefore improve network capacity utilization. The transponder edge is a kind of auxil-iary edge to bridge a fiber edge and a lightpath edge (shown as a dotted line in Fig. 8b). Since it does not correspond to any network capacity resource, we set its weight to be zero.

The AG is a type of virtual network topology specially built for jointly solving the multilayer traffic grooming and RSA problems. For example, for a service request between nodes A and D, the corresponding AG can be built as shown in Fig. 8b. In Step 3, we employ the shortest path algorithm to find the shortest route (with the smallest sum weight) between nodes A and D, which is A–A1–C2–C–D accord-ing to the constructed AG. In Step 4, if a route can be found, we assign corresponding network resources along the route to establish the service connection; otherwise, the service connection is blocked.

4.2 Performance

We evaluate the performance of the partial OTN switching scheme in terms of system cost, power consumption, and BBP. The test network topology is shown in Fig. 5, and the test conditions are the same as described in Sect.  3.2.

Figure 10 shows the simulation results of a UD-WSN with partial OTN switching. The legend “OTN-partial-x” corresponds to the results for UD-WSN with partial OTN switching whose spectrum granularity is x GHz, “OTN over DWDM” corresponds to the results for the OTN over DWDM network, and “OTN” corresponds to the results for the pure OTN. The legend “HDF-x” corresponds to the results for the HDF strategy with the spectrum granularity of x GHz; “HTF-x” corresponds to the results for the HTF strategy with a spectrum granularity of x GHz. The x-axis indicates the percentage of network nodes that are deployed with OTN switches in a UD-WSN. The following key obser-vations can be made.

First, the BBP performance is improved substantially when an increasing percentage of OTN switching nodes are deployed. However, a saturation phenomenon is also observed. When the percentage of the deployed OTN switches exceeds a certain threshold value (e.g., 21% for the HDF strategy as shown in Fig. 10a), further increasing the OTN switching nodes does not bring a significant per-formance improvement (see Fig. 10a and d). This implies that a UD-WSN with only a few OTN switching nodes can achieve almost as good performance as one where more (or even all) network nodes are deployed with OTN switches.

(a) Physical topology

A B C D

(b) An example of auxiliary graph

A B C D

A1

B1

B2

C1

C2 D1

Transponder edge

Fiber edge

Lightpath edge

Fig. 8 An example of AG-based traffic grooming

Table 2 Weights of different edges in an AG

Fiber edge Lightpath edge Transponder edge

1 + 1∕W2 0.05 ⋅ H 0

Fig. 9 An example of available spectrum windows (SWs)

Page 9: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

Second, we see that UD-WSN shows much lower cost and power consumption compared to the OTN over DWDM network when they both give similar BBP performance. For example, when the percentage of OTN switching nodes is 15%, the two schemes achieve similar BBP performance in Fig. 10a under a 5-GHz granularity; however, UD-WSN costs less and consumes less power. This can be seen in Fig. 10b and c when there are 15% OTN switching nodes.

Third, comparing the two OTN deployment strategies, the HTF strategy is more efficient and outperforms the HDF strategy, and the performance difference between the two is even larger when only a small percentage of the network nodes are deployed with OTN switches (see Fig. 10d). This is reasonable since the HTF strategy selects many “hub” nodes to be deployed with OTN switches (i.e., the nodes with the highest probabilities of being traversed by the shortest paths between all node pairs). However, HDF only considers the physical nodal degree but even a high nodal degree cannot ensure a large probability of the node being traversed by the shortest paths between all the node pairs.

5 UD‑WSN with spectrum defragmentation

UD-WSN is a special EON that allows for very fine spec-trum granularities, particularly targeting efficient provision-ing for many small traffic demands in the metro network. However, setting up and tearing down service connec-tions with different numbers of FSs under dynamic traffic demand can lead to severe spectrum fragmentations. This would make it difficult to find contiguous FSs of the required width for new service connections and will end up even-tually degrading the overall network spectrum utilization. Spectrum defragmentation [23] is an effective approach to mitigate spectrum fragmentation and consolidate the overall network spectrum utilization. Most of the existing studies mainly focus on spectrum defragmentation for the backbone EON [23–28] based on a 12.5-GHz spectrum granularity. Studying spectrum defragmentation for a metro optical net-work with smaller spectrum granularities (e.g., 5 GHz) has still not been examined in the literature. In this section, we evaluate how doing spectrum defragmentation can improve the spectrum utilization of UD-WSN when different spec-trum granularities are being used.

5.1 UD‑WSN with spectrum defragmentation

Spectrum defragmentation would efficiently smoothen the network spectra usage for newly arriving future service con-nections and would help to reduce network service block-ing. To study this, we consider two triggering mechanisms, i.e., batch defragmentation (BD) and defragmentation upon blocking (BTD), and two defragmentation strategies, i.e.,

(b)

(a)

(d)

(c)

BBP (HDF, offered load: 14 Erlang per node pair)

Cost (HDF, # of service connections = 500)

Power consumption (HDF, # of service connections = 500)

BBP of different strategies (offered load: 14 Erlang per node pair)

1.0E-02

1.0E-01

1.0E+00

0 3 6 9 12 15 18 21 24 27 30 33 36Percentage of OTN switching nodes (%)

Band

wid

th b

lock

ing

prob

abili

tyOTN -partial -5OTN -partial -6.25OTN -partial -10OTN -partial -12 .5

OTN

OTN over DWDM

2.5E+05

3.2E+05

3.9E+05

4.6E+05

5.3E+05

6.0E+05

0 3 6 9 12 15 18 21 24 27 30 33 36Percentage of OTN switching nodes (%)

Cum

ulat

ive

Cap

Ex

OTN-partial-5OTN-partial-6.25OTN-partial-10OTN-partial-12.5

OTN

OTN over DWDM

1.0E+04

1.5E+04

2.0E+04

2.5E+04

3.0E+04

3.5E+04

0 3 6 9 12 15 18 21 24 27 30 33 36Percentage of OTN switching nodes (%)

Pow

er c

onsu

mpt

ion (

W)

OTN-partial-5OTN-partial-6.25OTN-partial-10OTN-partial-12.5

OTN

OTN over DWDM

1.0E-02

1.0E-01

1.0E+00

0 3 6 9 12 15 18 21 24 27 30 33 36Percentage of OTN switching nodes (%)

Ban

dwid

th b

lock

ing

prob

abili

ty

HDF -5HDF -12.5HTF -5HTF -12.5

Fig. 10 Simulation results of partial OTN switching

Page 10: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

sequentially releasing and re-establishing service connec-tions (SR-D) and jointly releasing and re-establishing ser-vice connections (JR-D) [27, 28]. For the BD scheme, a defragmentation process is triggered after a certain number of services connections are released. For the BTD scheme, a defragmentation process is triggered whenever a service connection is blocked.

Figure 11 shows examples for the two defragmentation strategies. We assume that each link contains 10 FSs. For a four-node network as shown in Fig. 11a, Fig. 11b shows the current state of network spectrum resources where the service connections S1, S2, S3, and S4 and their cor-responding lightpaths are shown as dotted lines. We also assume that the remaining holding times of service con-nections S1, S2, S3, and S4 are 0.6 s, 0.2 s, 0.8 s, and 0.1 s, respectively. Figure 11c and d shows the states of network spectrum resources after applying the SR-D and JR-D strate-gies, respectively. In SR-D, we sequentially release and re-establish service connections S1, S2, S4, and S3. (This is done in order starting from the highest ending FS indexes.) In JR-D, we release the established service connections S1, S2, S3, and S4 altogether, and then run the routing and spectrum assignment (RSA) process for the service con-nections (based first on their bandwidth requirement and then on their remaining holding times,1 both in a descending order). For JR-D, if the RSA process fails, we will restore the original spectrum allocations of the lightpaths and consider the next attempt, i.e., SR-D. Only if the SR-D strategy also fails, would we block the service request. We can see that JR-D can achieve a more even FS distribution than that of SR-D after defragmentation because JR-D employs a global

optimization effort to release all the occupied spectra before re-establishing the lightpaths.

We employ the SWP-based heuristic algorithm for the RSA optimization in the spectrum defragmentation scheme. Here, we mainly introduce the defragmentation strategies, i.e., SR-D and JR-D.

5.1.1 SR‑D strategy

The following are the main steps of the SR-D strategy.

SR-D strategy

Step 1 Add all the established service connections in a request list R and sort them in a descending order based on their ending FS indexes

Step 2 Get the first service connection r from the request list R , tear down its corresponding lightpath, and employ the SWP-based heuristic algorithm with the first-fit strategy for RSA

Step 3 Remove the served request r from the request list R ; if the request list is empty, terminate the SR-D process; other-wise, go to Step 2

In Step 1, we sort all the service connections that are involved in the defragmentation process in a descending order based on their ending FS indexes. In Step 2, we tear down the first service connection r in the sorted request list R , and then employ the SWP-based heuristic algorithm with the first-fit strategy for routing and spectrum allocation. In Step 3, we continue applying the defragmentation process as described before for the remaining service connections until they are all defragmented.

5.1.2 JR‑D strategy

The following are the main steps of the JR-D strategy.

JR-D strategy

Step 1 Add all the established service connections in a request list R and sort them in a descending order based on the number of FSs required and the remaining holding time

Step 2 Tear down all the service connections in the request list RStep 3 Get the first service connection r from the request list R and

employ the SWP-based heuristic algorithm with the first-fit strategy for RSA

Step 4 If the RSA process fails, terminate the JR-D process and restore all the released service connections to the original state, and then refer to the SR-D process for another round of defragmentation

Step 5 Remove the served request r from the request list R ; if the request list is empty, terminate the JR-D process; other-wise, go to Step 3

In Step 1, we sort all the service connections that are involved in the defragmentation process in a descend-ing order based on the number of FSs required and the

(a) Network topology (b) Before defragmentation

(c) After SR-D (d) After JR-D

A

B

C

D

S1

S2

S4

A-B

B-C

C-D

D-A

Link

FS index0 1 2 3 4 5 6 7 8 9

S1

S1S2

S2

S3

S4

S4

A-B

B-C

C-D

D-A

Link

FS index0 1 2 3 4 5 6 7 8 9

S1

S1 S2

S2

S3

S4

S4

A-B

B-C

C-D

D-A

Link

FS index0 1 2 3 4 5 6 7 8 9

S1

S1S2

S2

S3

S4

S4

Fig. 11 An illustrative example of spectrum defragmentation

1 For the simulation, the service holding time for a demand is selected from an exponentially distributed random variable. The remaining holding time referred to here is the service time remaining out of this (earlier selected) value.

Page 11: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

remaining holding time. In Step 2, we first tear down all the service connections that are involved in the defragmenta-tion process. In Step 3, we employ the SWP-based heuristic algorithm with the first-fit strategy for RSA for the first ser-vice connection r in the sorted list R . In Step 4, if the RSA process failed, we terminate the JR-D process and restore all the released service connections to the original state before the JR-D process, and then, we implement the SR-D pro-cess. In Step 5, we continue applying the RSA process as described for the remaining service connections until they are all served if there is no failure of the RSA process.

In addition, for the purpose of performance comparison, we also develop the defragmentation algorithms for the pure OTN and the OTN over DWDM network. The main steps of the SR-D process and JR-D process are the same as those for UD-WSN, except that we employ the AG-based multi-layer traffic grooming algorithm of Sect. 4 for the network resource optimization.

5.2 Performance

We evaluate the performance of the proposed defragmenta-tion schemes in terms of their BBP based on the metro net-work topology shown in Fig. 5. The test conditions are the same as those described in Sect. 3.2. For the BD scheme, we assume that defragmentation is triggered whenever 200 ser-vice connections are released. Figure 12 shows the results of BBP performance under different combinations of trigger-ing mechanisms and defragmentation strategies. The legends “BD” and “BTD” correspond to the trigger mechanisms of BD and BTD, respectively. The legends “SR-D” and “JR-D” correspond to the defragmentation strategies of SR-D and JR-D. The following key observations and conclusions can be made.

First, UD-WSN with a smaller spectrum granularity can achieve lower BBP after spectrum defragmentation. This is reasonable since with a smaller spectrum granularity UD-WSN can accommodate low-speed metro services (e.g., GE/10GE connections) more efficiently with less spectrum wastage and can establish more end-to-end optical chan-nels. Second, from the results of Fig. 12a–d, we can see that UD-WSN with a 5-GHz spectrum granularity can achieve a BBP close to those of the pure OTN and OTN over DWDM networks. This is because the spectrum capacity of a 5-GHz spectrum granularity can achieve full spectrum utilization for the traffic demands whose granularities are 10G, 40G, and 100G. This avoids spectrum wastage when establishing these network services. Third, comparing the BBP perfor-mance of the different defragmentation schemes, we can see that BTD JR-D performs best (see Fig. 12e). This is reasona-ble because the BTD JR-D scheme triggers defragmentation whenever a service connection blocking occurs and releases all the occupied spectra before re-establishing lightpaths.

(a) BD SR-D

(b) BD JR-D

(c) BTD SR-D

(d) BTD JR-D

(e) Defragmentation versus non-defragmentation (5 GHz)

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

10 12 14 16 18 20

Traffic load per node pair (Erlang)

Band

wid

th b

lock

ing

prob

abili

ty

UD-WSN-5UD-WSN-6.25UD-WSN-10UD-WSN-12.5OTN over DWDMOTN

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

10 12 14 16 18 20

Traffic load per node pair (Erlang)

Ban

dwid

th b

lock

ing

prob

abili

ty

UD-WSN-5UD-WSN-6.25UD-WSN-10UD-WSN-12.5OTN over DWDMOTN

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

10 12 14 16 18 20

Traffic load per node pair (Erlang)

Ban

dwid

th b

lock

ing

prob

abili

tyUD-WSN-5UD-WSN-6.25UD-WSN-10UD-WSN-12.5OTN over DWDMOTN

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

10 12 14 16 18 20

Traffic load per node pair (Erlang)

Ban

dwid

th b

lock

ing

prob

abili

ty

UD-WSN-5UD-WSN-6.25UD-WSN-10UD-WSN-12.5OTN over DWDMOTN

1.0E-03

1.0E-02

1.0E-01

1.0E+00

12 14 16 18 20

Traffic load per node pair (Erlang )

Ban

dwid

th b

lock

ing

prob

abili

ty

BD JR-DBD SR-DBTD JR-DBTD SR-DUD-WSN-5

Fig. 12 Simulation results of spectrum defragmentation

Page 12: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

6 Conclusions

Metro optical networks will be required to operate with high spectral efficiency, low power consumption, and low service connection latency in order to meet the stringent quality of service (QoS) required for today’s cloud-based services. For this, we proposed a novel metro optical network architec-ture called UD-WSN, which enables an even finer spectrum granularity, i.e., as small as 5 GHz. UD-WSN was verified to have spectrum efficiency close to the pure OTN and OTN over DWDM network but shows lower cost, lower power consumption, and shorter end-to-end connection latency. Due to the promising potential of this approach, this paper focuses on techniques to enhance the performance of UD-WSN further. Specifically, we considered UD-WSN without aggregation OTN switches for lower system cost and UD-WSN with partial OTN switching for better capacity utili-zation. We also carried out spectrum defragmentation for UD-WSN to enhance its spectrum utilization. Studies indi-cate that the architecture without aggregation OTN switches can greatly reduce the system cost and power consumption, but its spectrum utilization is degraded because low-speed services may occupy entire optical channels and this would also significantly degrade the BBP performance. We sug-gest implementing a UD-WSN in different flavors accord-ing to different system performance requirements and cost considerations. Simulation studies of UD-WSN with partial OTN switching indicate that it can achieve a BBP close to that of pure OTN and OTN over DWDM networks at lower cost and with lower power consumption. A saturation trend was also observed, i.e., when the percentage of OTN switching nodes reaches a certain value, increasing the per-centage of OTN switching nodes does not provide a com-mensurate improvement in the BBP performance. Finally, the evaluation of spectrum defragmentation indicated that spectrum defragmentation with a smaller spectrum granu-larity can achieve better spectrum utilization, and spectrum defragmentation with a 5-GHz spectrum granularity can achieve a BBP close to those of the pure OTN and OTN over DWDM network.

It should be noted that all the results and conclusions obtained in this paper were based on specific network resource optimization algorithms. Though we believe that these algorithms are efficient enough, there may be other efficient algorithms for the above resource allocations. Therefore, as future studies, more efficient resource allo-cation algorithms can be proposed and evaluated for the proposed UD-WSN. However, we believe that the overall performance trends would not be significantly different from those found in this study when these new algorithms are adopted.

References

1. Zhang, Y., Zhou, X., Sheng, Y., Deng, N., Shen, G.: Spectrum defragmentation and partial OTN switching in ultra dense-wave-length switched network (UD-WSN). In: Proceedings of the 19th International Conference on Transparent Optical Network, pp. 1–4 (2017)

2. IHS Technology: Networking ports market tracker: 1G, 2.5G, 10G, 40G, 100G abstract (2016). https ://techn ology .ihs.com/55053 4/netwo rking -ports -1g-25g-10g-25g-40g-100g-marke t-track er-regio nal-h1-2016. Accessed 21 Nov 2018

3. Freiberger, M.: In-network experiences with installed OTN switched metro core optical systems. In: Proceedings of the Opti-cal Fiber Communication Conference Exhibition, pp. 1–3 (2015)

4. ITU-T SG15: Spectral grids for WDM applications: DWDM fre-quency grid. ITU-T G.694.1 (2012)

5. Zhang, Y., Zhou, X., Deng, N., Shen, G.: Ultra dense-wavelength switched network (UD-WSN): a cost, energy, and spectrum effi-cient metro network architecture. In: Proceedings of the Asia Communications and Photonics Conference, pp. 1–3 (2016)

6. Shen, G., Zhang, Y., Zhou, X., Sheng, Y., Deng, N., Ma, Y., Load, A.: Ultra-dense wavelength switched network: a special EON par-adigm for metro optical networks. IEEE Commun. Mag. 56(2), 189–195 (2017)

7. Zhou, X., Jia, W., Ma, Y., Deng, N., Shen, G., Lord, A.: An ultradense wavelength switched network. IEEE/OSA J. Lightw. Technol. 35(11), 2063–2069 (2016)

8. Jinno, M., Takara, H., Kozicki, B., Tsukishima, Y., Sone, Y., Matsuoka, S.: Spectrum-efficient and scalable elastic optical path network: architecture, benefits, and enabling technologies. IEEE Commun. Mag. 47(11), 66–73 (2009)

9. Sales, V., Segarra, J., Polo, V., Velásquez, J.C., Prat, J.: UDWDM-PON using low-cost coherent transceivers with limited tunabil-ity and heuristic DWA. IEEE/OSA J. Opt. Commun. Netw. 8(8), 582–599 (2016)

10. Marom, D.M., Rudnick, R., Goldshtein, N., Golani, O., Sinefeld, D.: Realization of sub-1 GHz resolution photonic spectral pro-cessors for flexible optical networks. In: Proceedings of the 41th European Conference on Optical Communication, pp. 1–3 (2015)

11. Goldshtein, N., Sinefeld, D., Golani, O., Rudnick, R., Pascar, L., Zektzer, R., Marom, D.M.: Fine resolution photonic spectral pro-cessor using a waveguide grating router with permanent phase trimming. IEEE/OSA J. Lightw. Technol. 34(2), 379–385 (2016)

12. Cai, A., Shen, G., Peng, L., Zukerman, M.: Novel node-arc model and multiiteration heuristics for static routing and spec-trum assignment in elastic optical networks. IEEE/OSA J. Lightw. Technol. 31(21), 3402–3413 (2013)

13. Wang, C., Shen, G., Bose, S.K.: Distance adaptive dynamic routing and spectrum allocation in elastic optical networks with shared backup path protection. IEEE J. Lightw. Technol. 33(14), 2955–2964 (2015)

14. Zhu, H., Zang, H., Zhu, K., Mukherjee, B.: A novel generic graph model for traffic grooming in heterogeneous WDM mesh networks. IEEE/ACM Trans. Netw. 11(2), 285–299 (2003)

15. Wan, X., Li, Y., Zhang, H., Zheng, X.: Dynamic traffic grooming in flexible multi-layer IP/optical networks. IEEE Commun. Lett. 16(12), 2079–2082 (2012)

16. Zhang, S., Martel, C., Mukherjee, B.: Dynamic traffic grooming in elastic optical networks. IEEE J. Sel. Areas Commun. 31(1), 4–12 (2013)

17. Cai, A., Shen, G., Peng, L.: Optimal planning for electronic traffic grooming in IP over elastic optical networks. In: Proceedings of the Asia Communications Photonics Conference, pp. 1–3 (2013)

Page 13: Ey of˚ultr‑dense wavelength switched network …...Phot Nw Communications 1 3 withanentireopticalchannel(e.g.,with5-GHzbandwidth), whichwoulddegradethenetwork’sspectrumutilization

Photonic Network Communications

1 3

18. Walklin, S.: Leaf-spine architecture for OTN switching. In: Pro-ceedings of the International Conference on Computing Network Communication, pp. 95–99 (2017)

19. Vizcaíno, J.L., Ye, Y., López, V., Jiménez, T., Krummrich, P.M.: OTN switching for improved energy and spectral efficiency in WDM MLR networks. In: Proceedings of the Optical Fiber Com-munication Conference Exhibition, pp. 1–3 (2016)

20. Shen, G., Grover, W.D., Cheng, T.H., Bose, S.K.: Sparse place-ment of electronic switching nodes for low blocking in translucent optical networks. IEEE J. Opt. Netw. 1(12), 424–441 (2002)

21. Awwad, O., Al-fuquha, A.I., Guizani, M.: Genetic approach for traffic grooming, routing, and wavelength assignment in WDM optical networks with sparse grooming resources. In: Proceedings of the International Conference Communications, pp. 2447–2452 (2006)

22. Shen, G., Tucker, R.S.: Sparse traffic grooming in translucent opti-cal networks. IEEE/OSA J. Lightw. Technol. 27(20), 4471–4479 (2009)

23. Wang, X., Kim, I., Zhang, Q., Palacharla, P., Sekiya, M.: A hitless defragmentation method for self-optimizing flexible grid optical networks. In: Proceedings of the 38th European Conference on Exhibition Optical Communications, pp. 1–3 (2012)

24. Yin, Y., Wen, K., Geisler, D.J., Liu, R., Yoo, S.J.B.: Dynamic on-demand defragmentation in flexible bandwidth elastic optical networks. Opt. Exp. 20(2), 1798–1804 (2012)

25. Cugini, F., Paolucci, F., Meloni, G., Berrettini, G., Secondini, M., Fresi, F., Sambo, N., Poti, L., Castoldi, P.: Push-pull defragmen-tation without traffic disruption in flexible grid optical networks. IEEE/OSA J. Lightw. Technol. 31(1), 125–133 (2013)

26. Wang, R., Mukherjee, B.: Provisioning in elastic optical networks with non-disruptive defragmentation. IEEE/OSA J. Lightw. Tech-nol. 31(15), 2491–2500 (2013)

27. Wang, C., Shen, G., Chen, B., Peng, L.: Protection path-based hit-less spectrum defragmentation in elastic optical networks: shared backup path protection. In: Proceedings of the Optical Fiber Com-munications Conference Exhibition, pp. 1–3 (2015)

28. Wang, C., Shen, G., Peng, L.: Protection lightpath-based hitless spectrum defragmentation for distance adaptive elastic optical networks. Opt. Exp. 24(5), 4497–4511 (2016)

Ya Zhang is currently a Master student with Soochow University in China. His current research interest focuses on optical networks.

Xu Zhou is currently a Research Engineer with Huawei Technologies Co., Ltd, China. His research interest focuses on all optical switching and transport networks. He has authored or co-authored more than 10 peer-reviewed technical papers and has more than 10 patents granted or pending.

Ning Deng is currently a principal engineer with Huawei Technologies Co., Ltd. He has been working on advanced technologies in optical networking and photonic switching. He also actively contributes to both interna-tional and domestic standard bodies. He serves as an Editor for multiple ITU-T and CCSA standards. He is a Senior Member of IEEE. He has more than 20 patents granted or pending, and over 40 papers in referred

international conferences and journals.

Sanjay K. Bose got his B.Tech degree from IIT Kanpur in 1976 and his Master’s and Ph.D. from S.U.N.Y. Stony Brook, USA, in 1977 and 1980, respectively. After working with the Corporate R&D Centre of the Gen-eral Electric Co. in Schenectady, NY, USA, till 1982, he joined IIT Kanpur as an Asst. Professor and became a Professor there in 1991. He left IIT Kanpur in 2003 to join the faculty of the School of EEE, NTU, Singa-

pore. In Dec 2008, he left NTU to join IIT Guwahati where he is cur-rently a Professor in the Department of EEE. Prof. Bose has been work-ing in various areas in the field of Computer Networks and Queueing Systems and has published extensively in the area of optical networks and network routing. Prof. Bose is a Senior Member of IEEE, a Fellow of IETE (India), and a member of Sigma Xi and Eta Kappa Nu. More details on Prof. Bose can be found on his web page at http://www.iitg.ernet .in/skbos e/.

Gangxiang Shen is a Distinguished Professor with the School of Electronic and Informa-tion Engineering of Soochow University in China. Before he joined Soochow University, he was a Lead Engineer with Ciena, Linthi-cum, Maryland. He was also an Australian ARC Postdoctoral Fellow with University of Melbourne. His research interests include integrated optical and wireless networks, spectrum-efficient optical networks, and

green optical networks. He has authored and co-authored more than 150 peer-reviewed technical papers, among which one of the papers received the highest citations among all the papers published in IEEE/OSA JOCN. He was a Lead Guest Editor of IEEE JSAC Special Issue on “Next-Generation Spectrum-Efficient and Elastic Optical Transport Networks,” and a Guest Editor of IEEE JSAC Special Issue on “Energy-Efficiency in Optical Networks.” He is an associated editor of IEEE/OSA JOCN, and an editorial board member of Optical Switching and Networking and Photonic Network Communications. He has served as TCP chairs and members for various international conference in the area of optical networking, including general TPC co-chair of ACP 2018 and symposium lead chair of GLOBECOM 2017, and TPC members for OFC and ECOC. He is a “Highly Cited Chinese Research Scholar” selected by Elsevier (from 2014 to 2017) and an “Excellent Young Research Scholar” sponsored by NSFC. He was a Secretary for the IEEE Fiber-Wireless (FiWi) Integration Sub-Technical Committee. He is serving as a member of IEEE ComSoc Strategic Planning Stand-ing Committee and an IEEE ComSoc Distinguished Lecturer (2018–2019).