Design of Optical Aggregation Network with Carrier ...

Design of Optical Aggregation Network with Carrier Edge Functions Virtualization Takashi Miyamura† Akira Misawa† Jun-ichi Kani∗ NTT Network Service Systems Laboratories, NTT Corporation, Musashino, Tokyo, Japan ∗ NTT Access Service Systems Laboratories, NTT Corporation, Yokosuka, Kanagawa, Japan E-mail: † [email protected] †

Abstract—NFV (network functions virtualization) enables carrier edge functions to be relocated from dedicated hardware to distributed pools of commodity servers, which reduces network cost and improves robustness to unpredictable demand changes. The key to this lies in how to place each VNF (virtual network function) on an adequate server among pools in conjunction with design of paths connecting servers with OLTs (optical line terminals). There have been extensive studies regarding VNF placement problems, but few have investigated the optimal placement of VNFs in consideration of carrier edge function virtualization. We thus study a design problem of carrier edge VNF placement in an optical aggregation network. We formulate a design problem as MILP (mixed-integer linear programming) and demonstrate the effectiveness of the proposed method through intensive mathematical experiments. The experiments showed that the proposed method reduced the maximum server load and network cost by up to by 29% and 21%, respectively.

I. Introduction Network convergence between fixed and mobile telecommunications networks has been gradually deployed in commercialized carrier networks, which provide various media services, such as Internet access, IP telephony, and video on demand. Service admission for such services is basically located at an edge router in carrier networks. NFV (network functions virtualization)[1], [2] is expected to be widely deployed in carrier networks and enables carrier edge functions to be relocated from dedicated hardware to distributed pools of commodity servers, which reduces network cost and improves robustness to unpredictable demand changes. Carrier edge function virtualization has been extensively studied [3], [4]. In carrier edge function virtualization, edge routers are replaced by edge servers (ESs) with VNFs (virtual network functions). A VNF, which runs on a commodity server, provides various carrier edge functions such as BAS (broadband access server) and SBC (session border controller) functions. Carrier edge function virtualization dramatically reduces capital and operational expenditure by deploying commodity servers instead of dedicated-purpose hardware. There have recently been extensive studies of all-optical aggregation network architecture [5], [6] in order to accomA. Misawa is currently with Chitose Institute of Science and Technology (CIST), Bibi, Chitose, Hokkaido, 066-8655 Japan.

modate large-volume traffic efficiently. Here, aggregation networks efficiently enable traffic to flow between access and backbone networks, as illustrated in Fig. 1. Our research group presented a system architecture for optical TDM (time division multiplexing)-based aggregation networks [5]. A dynamically reconfigurable TDM-WDM (wavelength division multiplexing) PON (passive optical network) ring architecture has also recently been proposed [6]. To cope with bursty traffic and accommodate numerous OLTs (optical line terminals), aggregation network architectures basically enable a wavelength path to be shared by multiple OLTs through DBA (dynamic bandwidth allocation) [7] or other optical TDM technologies. Existing studies [3], [4], [5], [6] mainly focused on developing and demonstrating a network architecture, but few have investigated the design of optical aggregation networks in consideration of VNF placement for carrier edge function virtualization. Each carrier edge VNF must be placed on an adequate server among server pools in conjunction with wavelength paths routed so that they connect servers with OLTs. The important thing here is that VNF placement on an adequate server should be performed in consideration of physical network design like wavelength path routing for efficient resource usage and improved performance. By adequately designing optical aggregation networks with carrier edge VNF placement, we can reduce network cost and improve robustness to unpredictable demand changes. In this paper, we thus study the problem of designing carrier edge VNF placement in optical aggregation networks. The main contributions of this paper lie in i) presenting mathematical formulations for finding optimal design of a TDM-WDM-based optical aggregation network with carrier edge VNF placement and ii) quantitatively evaluating the effectiveness of the optimal aggregation network design with VNF placement. The remainder of this paper is organized as follows. We briefly review related work in Section 2. We then describe an architecture of an optical aggregation network with carrier edge function virtualization in Section 3. Section 4 addresses to a model of the optical aggregation network and presents a mathematical formulation of the design problem. Results of intensive mathematical experiments are presented in Section 5. A brief conclusion is provided in Section 6.

ĂƌƌŝĞƌĞĚŐĞ ĨƵŶĐƚŝŽŶ ŐŐƌĞŐĂƚŝŽŶŶĞƚǁŽƌŬ

ĚŐĞĂƚĂĐĞŶƚĞƌ ŽƌĞŶĞƚǁŽƌŬ

ĐĐĞƐƐŶĞƚǁŽƌŬ

K>d

Case2:

Case1:

dDͲtDƌŝŶŐ

KƉƚŝĐĂůĨŝďĞƌ

WϮDWǁĂǀĞůĞŶŐƚŚƉĂƚŚƐ

TwoOLTsareaccommodatedby differentESs • Consumed3WDMlinks. • Severloadiswellbalanced.

sE& ZKDŽƌKy

TwoOLTsareaccommodatedby thesameES

2

ĚŐĞ ƐĞƌǀĞƌ

• Consumedonly2WDMlinks. • Theservermaybeoverloaded.

1

1

2

Fig. 1. Aggregation network architecture with carrier edge virtualization

2

1

II. Related Work We briefly review related work on i) the optimization of optical multicast networks and ii) the optimal design of virtual resources. Here, we consider a TDM-WDM-based network as an aggregation network, which connects an OLT with a corresponding server accommodating carrier edge VNF using a shared P2MP (point-to-multipoint) wavelength path. To design a multicast network, a mathematical formulation of multicast trees in network cording was presented by Ahlswede et al. [8]. K¨oksal and Ersoy [9] and Hachisuka et al. [10] produced related work on optical multicast network design algorithms. Their work considers optical multicast that simply splits optical signals at intermediate nodes, so their models did not cover the sharing of a P2MP wavelength path with multiple destination nodes through DBA. In aggregation networks, as we described earlier, such capability is essential to efficiently accommodate bursty traffic from OLTs by using over-subscription or statistical multiplexing. For VNF placement and virtual network design, the optimization of VNF chaining problems in a packet-based network has been proposed [11]. The mapping of virtual networks on physical networks has also been investigated [12], and heuristic algorithms for VNF placement have been proposed [13], [14]. However, they did not consider a shared optical P2MP path network as a physical network. In summary, few researchers have investigated VNF placement in shared P2MP wavelength path networks.

:SourceNode(ES)

:DestinationNode(OLT)

:Corenode(ADM)

:WDMorFiberlink

1

2

:Sharedwavelengthpath

Fig. 2. Illustrative example

introducing NFV, we can efficiently accommodate unpredictable demand changes with limited network resources. The important feature of the architecture is that each VNF has one-to-one correspondence to a specific OLT of PONs. Thus, the architecture requires a mechanism for providing connectivity between an ES accommodating a VNF and a corresponding OLT. For this purpose, we consider a shared P2MP wavelength path network that only requires a set of conventional optical ADM (add/drop multiplexer) modules for an intermediate node. A group of ADMs consists of a ring topology, and each wavelength in the network forms a P2MP wavelength path shared by multiple OLTs that have the same target ES. The bandwidth of each P2MP wavelength path can be shared by multiple OLTs in accordance with DBA; thus, we can improve the resource utilization of each wavelength channel through statistical multiplexing. Optical burst contention resolution in the same wavelength channel is performed through a DBA mechanism. Here, DBA mechanisms are implemented on burst sender/receivers at ESs and OLTs. Technologies enabling such TDM-based shared wavelength path networks have been developed by Hattori et al. [5] and Carey et al. [6].

III. Optical Aggregation Network A. Architecture Overview Here, we describe an architecture of optical aggregation networks with carrier edge function virtualization. An overview of the architecture is illustrated in Fig. 1. In a conventional aggregation network, each edge router accommodates a group of PONs. In the optical aggregation networks considered in this paper, edge routers are replaced by ESs with VNFs. In the optical aggregation network architecture, resources are reallocated in accordance with the addition or withdrawal of subscribers and the variation in bandwidth usage for each subscriber. By

All wavelength channels are transmitted through the shared TDM-WDM ring topology. Intermediate ADMs just drop, copy, or pass through wavelength paths. Resource demand of each VNF also varies depending on the number of subscribers on an OLT and the duration of service usage. From the viewpoint of resource management, it is important to place each VNF to an adequate ES among server pools while considering bandwidth requirements and resource consumption in the shared TDM-WDM ring network.

2

B. Problem Statement

WŚLJƐŝĐĂůŶĞƚǁŽƌŬ

Figure 2 illustrates an example of a TDM-WDM ring network with two ESs and two OLTs. In Case 1, two pairs of ESs and OLTs are connected via two P2MP wavelength paths, respectively. On the other hand, in Case 2, two VNFs, located at the same ES, are connected with the corresponding OLT via a common P2MP wavelength path. Whereas Case 1 consumes more wavelength link resources than Case 2, the server load in Case 1 is adequately dispersed between two ESs. Thus, Case 1 is more tolerant of unexpected demand changes due to residual server resources. As illustrated in the above example, by adequately designing VNF location as well as shared P2MP wavelength paths, we can improve network resource efficiency and robustness to unpredictable demand changes. The important thing here is how to place each VNF on an adequate server among pools in consideration of efficient route selection between ESs and OLTs. Now, we describe the problem we are attempting to solve: how to place VNFs and allocate network resources so as to minimize network resource consumption and the maximum load of ESs on a given physical network.

>ŽŐŝĐĂůƚŽƉŽůŽŐLJ

Ϭ

00000000 00 00 00 00 00 00 00 00 00000000 00000000 00000000 00000000 00000000 00 00 00 00 00 00 00 00 00000000

WĂƚŚϬͲϭ 000000000 000000000 00 00 00 00 00 00 00 00 00 00000000 00 0 0 0 0 0 0 0 0

ϭ

ϭ

Ϯ

Ϭ

dƌĂĨĨŝĐϭнϮŝƐƌĞƐƚƌŝĐƚĞĚďLJ ďĂŶĚǁŝĚƚŚŽĨƐŚĂƌĞĚǁĂǀĞůĞŶŐƚŚ ƉĂƚŚƐ

^ŚĂƌĞĚǁĂǀĞůĞŶŐƚŚ ƉĂƚŚƐϬͲϭΘϮ >ŽŐŝĐĂůƉĂƚŚϬͲϭ 000000000000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 000000000000000

00000000 00000000 00 00 00 00 00 00 00 00 00000000

Ϯ

WĂƚŚϬͲϮ

dƌĂĨĨŝĐϭ

ϭ

>ŽŐŝĐĂůƉĂƚŚϬͲϮ dƌĂĨĨŝĐϮ

Ϯ

͗^

͗K>d

͗KD

͗sE&

Fig. 3. Relationship between physical network and logical network

P2MP wavelength paths considered in this paper. In a basic network optimization problem, source and destination nodes are pre-determined. However, in our model, the source-node (i.e. VNF) location changes depending on VNF location. We previously proposed mathematical formulations considering a part of such requirements [15]. Here, we extend our previous formulations [15] to apply them to shared P2MP wavelength path networks. The basic idea of the formulation is defining a multipoint-to-point flow from candidate ESs to each OLT and selecting the optimal ES among candidates. We introduce the following notations:

IV. Network Model A. Overview First, we describe an overview of an optical aggregation network model. The network consists of ESs, OLTs, intermediate nodes (OADMs), and physical WDM links connecting two adjacent nodes. We assume the following input given to the problem: • ESs accommodating VNFs, • OLTs generating traffic demand, • VNFs corresponding to each OLT, • Intermediate nodes (OADMs), • Physical WDM links, • Traffic demand of each OLT. Figure 3 shows an example that explains the relationship between physical and logical networks. In this example, two pairs of ESs and OLTs are attached with a TDMWDM ring network, and two VNFs are accommodated in ES 0. A shared P2MP wavelength path is established between ES 0 and OLTs 1/2 in the physical network. A bandwidth of the wavelength path is shared by two OLTs through DBA. Thus, each OLT can occupy up to the full bandwidth of the path, but the sum of average traffic from the two OLTs does not exceed the bandwidth. This enables a wavelength channel to be efficiently used by statistically multiplexing bursty traffic from OLTs.

• •

m and n denote end points of a physical link in the physical network, i and j denote source and destination nodes in the logical/physical network, respectively.

Given: • • • • • • • • • •

N : number of nodes in the network E: set of physical links V : set of physical nodes Volt : set of OLTs Ves : set of ESs Vin : set of intermediate nodes Eλ : maximum number of wavelengths per physical link Dj : Traffic demand generated at OLT j Cλ : capacity of a wavelength channel (in Gbps) Ces : capacity of each ES

Variables: • • • •

B. Notation

Lij : traffic demand on logical link ij i Cmn : required capacity of a P2MP path originating at ES i pij mn : physical path routing from node i to node j xkij : traffic routing from node i to node j regarding VNF k

We model a physical network as a directed graph G = (V, E). For the physical network design, the route and capacity of shared P2MP wavelength paths are determined i by variables Cmn and pij mn . In designing the logical netk work, xij is a binary variable that expresses the location

Second, we describe notations used in our formulation before presenting a mathematical formulation. Mathematical formulations of optical multicast path trees have been studied by K¨oksal and Ersoy [9], but their model cannot be applied to the variable source-node locations and shared

3

of VNF k, xkij

=

⎧ ⎪ ⎨1

if VNF k associated with OLT j is located at ES i otherwise

⎪ ⎩ 0

A mathematical formulation based on the above notations is presented in the next subsection. C. Mathematical Formulation

͗^

Finally, we present a mathematical formulation of shared P2MP wavelength path networks with carrier edge VNF placement. The objective of our design is to minimize network resource consumption while avoiding an increase in the maximum load of ESs on a given physical network. In the design, we need to determine how to do the following: • Place a VNF to an adequate ESs considering bandwidth requirement and the capacity of ESs. • Find routes for P2MP wavelength paths that minimize physical network resource consumption. • Define a group of OLTs that are accommodated by the same P2MP path. An outline of the MILP formulation is described below.

k

mn

k

In Eq. (1), the first term corresponds to the sum of wavelength resource consumption and the second term indicates the maximum load of ESs. Equation (2) shows a flow conservation law regarding physical path routing between ESs and OLTs. Please note that point-to-point path pij mn is not actually established in the physical network, but the variable indicates a conceptual flow [8] and is used for computing the route and capacity for P2MP i path Cmn . Equations (3), (4), (5) and (6) constrain link capacity considering shared P2MP paths. Equation (7) shows VNFs can be located at one candidate ES but the sum of traffic flows must equal 1. Equations (8) and (9) are constraints on the capacity of logical links and ESs.

i pij mn ≤ Cmn ,



i Cml −



l

V. Performance Evaluation A. Aims and conditions

j

Constraints: Physical network design: ⎧ ⎪ ⎨−Lij  l ∈ Ves  ij  ij pml − pln = Lij  l ∈ Volt , ⎪ ⎩ l l 0 l ∈ Vin

(3)

i Cln ≤ 0,

(4)

l ∈ Vin

To quantitatively evaluate the effectiveness of the proposed method, we performed intensive mathematical experiments that had two objectives: • to clarify the trade-off between the network cost and maximum server load. • to demonstrate the effectiveness of optimal VNF placement in consideration of optical network design in terms of reducing network cost as well as dispersing server load. Here, we describe conditions used in the experiments. We implemented our formulation on GLPK [16], which is open-source linear programming solver. A 10-node multiring network topology attached with 4 ESs and 6 OLTs is deployed in the experiments, as shown in Fig. 4. Traffic demand of OLTs is randomly generated in accordance with a uniform distribution. The experiments had three shared conditions: • number of wavelengths per link Eλ : 16 • capacity of a wavelength channel Cλ : 100 (Gbps) • capacity of each ES Ces : 500 (Gbps)

(2)

∀i ∈ Ves , ∀j ∈ Volt

l

i Cim ≥



Lij , ∀i ∈ Ves

(5)

j



i Cmn ≤ Eλ ,

(6)

i

Constraints: Logical network design:  xkij = 1, ∀j ∈ Volt k

 k

(7)

i

Dj · xkij ≤ Lij · Cλ , ∀i ∈ Ves , ∀j ∈ Volt Dj ·

xkij

≤ Ces , ∀i ∈ Ves

͗ZKD

Fig. 4. Ten-node 3-ring network topology attached with 4 ESs and 6 OLTs

Objective: Minimizing network cost and maximum server load ⎞ ⎛   k Cmn + α max Dj · xkij ⎠ (1) min ⎝ i

͗K>d

(8)

B. Relationship between network cost and server load

(9)

First, we investigate how optimization parameter α affects the network cost and maximum server load. Here, parameter α, which is given in Eq. (1), indicates the

j

4





ZŽƵŶĚƌŽďŝŶ

1HWZRUNFRVW 



1HWZRUNFRVW

0D[LPXPVHUYHUORDG 1HWZRUNFRVW

0D[6HUYHUORDG







WƌŽƉŽƐĞĚ 

EĞĂƌĞƐƚ



















2SWLPL]DWLRQSDUDPHWHU˞









7UDIILFGHPDQG

Fig. 5. Trade-off between network cost and maximum server load

Fig. 6. Comparison of network cost



0D[LPXPVHUYHUORDG

optimization weight of two objectives. If we set α as 0, the proposed method just minimize the network cost without considering the maximum server load. The greater value of α indicates the increased weight of the maximum server load in the optimization. Here, the network cost is defined as the total amount of WDM links occupied by the all P2MP wavelength paths established on the physical network. We compared both performance indices while varying α from 0 to 1000. The average traffic demand of each OLT was set as 40 Gbps. The results are shown in Fig. 5. When parameter α is relatively low, our proposed method tends to optimize network cost instead of minimizing the maximum server load. As parameter α grows, our method tends to optimize the maximum server load. By choosing α between 0.1 and 1, we can efficiently reduce network cost while adequately dispersing server load.



EĞĂƌĞƐƚ 

ZŽƵŶĚƌŽďŝŶ



WƌŽƉŽƐĞĚ

 











7UDIILFGHPDQG Fig. 7. Comparison of maximum server load

to 29% and 21%, respectively. Moreover, Proposed indicated an slight increase of the network cost by about 5% compared to Nearest, which is used as a benchmark of the minimum network cost in our evaluation. Thus, we have concluded that both resource efficiency and robustness to demand changes can be improved by adequately selecting the locations of VNFs and wavelength path routing.

C. Effectiveness of optimal VNF placement Second, we evaluate the effectiveness of optimal VNF placement in consideration of optical network design in terms reducing both the network cost and maximum server load. On the basis of the above investigation into parameter α, we set α as 0.5 in this experiment. We compared the performance of our proposed method (Proposed) with two conventional sever selection methods (Nearest and Round robin ) [15]. Here, in Nearst, VNFs are always placed on the nearest ES from an OLT among four ESs. Thus, Nearst basically provides the best solution among three methods in terms of the network cost. In Round robin, each VNF is placed among four ESs in the round robin order, which can efficiently distribute server load among four ESs without considering the network cost. We evaluated network cost and the maximum server load while varying traffic demand. The results are shown in Figs. 6 and 7. The horizontal line is relative traffic demand normalized by the capacity of wavelength channel. Traffic demand 0.5 means 50 Gbps of average traffic demand from each OLT. Compared with Round robin, Proposed reduced the maximum server load and network cost by up

VI. Concluding Remarks We studied the optimal design of optical aggregation networks with carrier edge function virtualization, where a P2MP wavelength path is shared by multiple OLTs through DBA and the location of carrier edge functions can be selected flexibly among commodity server pools. The main contributions of this paper lie in i) presenting mathematical formulations for finding optimal design of a TDM-WDM-based optical aggregation network with carrier edge VNF placement and ii) quantitatively evaluating the effectiveness of the optimal network design with VNF placement. We formulated the design problem as MILP and demonstrated the effectiveness of the proposed method through intensive mathematical experiments. The proposed method improved resource efficiency and robustness to unpredictable demand changes. Our mathematical experiments showed that the proposed scheme can reduce

5

the maximum server load and network cost by up to 29% and 21%, respectively. Moreover, it also enhances flexibility in network design. By adequately choosing optimization parameter α, we can reduce network cost or utilize redundant network resources in order to disperse server load.

[7] IEEE P802.3av Task Force, ʠ 10Gb/s Ethernet Passive Optical Network, ʡ http://www.ieee802.org/3/av/ [8] Rudolf Ahlswede, Ning Cai, Shuo-Yen Robert Li, and Raymond W. Yeung, “Network Information Flow,” IEEE Transactions on Information Theory, Vol.46, No.4, Jul 2000. [9] F. K¨ oksal and Cem Ersoy, “Multicasting for all-optical multifiber networks,” OSA Journal of Optical Networking, Vol. 6, No. 2, 219-238, 2007. [10] Y. Hachisuka, H. Hasegawa, and K. Sato, ʠ Impairment-aware multicast tree design for hierarchical optical path networks, ʡ Photonics in Switching 2012 (PS 2012), Th-S33-O08, Corsica, Sep 2012. [11] Bernardetta Addis, Dallal Belabed, Mathieu Bouet, and Stefano Secci, “Virtual Network Functions Placement and Routing Optimization,” CloudNet 2015 - IEEE 4th International Conference on Cloud Networking, Oct 2015, Niagara Falls, ON, Canada. IEEE, pp.171-177. [12] N. M. Mosharaf Kabir Chowdhury, Muntasir Raihan Rahman, and Raouf Boutaba, “Virtual Network Embedding with Coordinated Node and Link Mapping,” INFOCOM 2009: 783-791, Apr 2010. [13] Mari Otokura, Kenji Leibnitz, Yuki Koizumi, Daichi Kominami, Tetsuya Shimokawa, and Masayuki Murata, ʠ Application of evolutionary mechanism to dynamic virtual network function placement, ʡ The IEEE 24th International Conference on Network Protocols (ICNP), Workshop: Control Operation and Application in SDN protocols (CoolSDN Workshop), Nov 2016. [14] Tachun Lin, Zhili Zhou, Massimo Tornatore, and Biswanath Mukherjee, “Demand-Aware Network Function Placement,”, IEEE Lightwave Technology Journal of, vol. 34, pp. 2590-2600, 2016. [15] Shigeyuki Yamashita, Daiki Imachi, Miki Yamamoto, Takashi Miyamura, Shohei Kamamura, and Koji Sasayama, “A New Content-Oriented Traffic Engineering for Content Distribution: CAR (Content Aware Routing)” IEICE Trans. Commun., vol. E98-B, no.4, pp. 575-584, Apr. 2015. [16] “GNU Linear Programming Kit,” https://www.gnu.org/software/glpk/

References [1] ETSI, GS NFV001,ʠ Network Functions Virtualisation (NFV); Use Cases, ʡ1 V1.1.1, Oct. 2013. http://www.etsi.org/standards [2] Rashid Mijumbi, Joan Serrat, Juan-Luis Gorricho, Niels Bouten, Filip De Turck, and Raouf Boutaba, ”Network Function Virtualization: State-of-the-Art and Research Challenges”, Communications Surveys & Tutorials IEEE, vol. 18, pp. 236-262, 2016, ISSN 1553-877X [3] Joon-Myung Kang, Hadi Bannazadeh, Hesam Rahimi, Thomas Lin, Mohammad Faraji, and Alberto Leon-Garcia, “SoftwareDefined Infrastructure and the Future Central Office,” Communications Workshops (ICC), 2013 IEEE International Conference on, Jun 2013. [4] Akira Misawa, Konomi Mochizuki, Hideo Tsuchiya, Masahiro Nakagawa, Kyota Hattori, Masaru Katayama, and Jun-ichi Kani, “Virtual Edge Architecture with Optical Bandwidth Resource Control,” IEICE Transactions on Communications, Vol. E99.B (2016) No. 8 pp. 1805-1812. [5] Kyota Hattori, Toru Homemoto, Masahiro Nakagawa, Naoki Kimishima, Masaru Katayama, and Akira Misawa, “Optical Layer 2 Switch Network with Bufferless Optical TDM and Dynamic Bandwidth Allocation,” IEICE Transactions on Electronics, Vol. E99.C (2016) No. 2 pp. 189-202. [6] Daniel Carey, Nicola Brandonisio, Stefano Porto, Alan Naughton, Peter Ossieur, Nick Parsons, Giuseppe Talli, and Paul Townsend, “Dynamically Reconfigurable TDM-DWDM PON Ring Architecture for Efficient Rural Deployment,” ECOC 2016; 42nd European Conference on Optical Communication; Proceedings of, Sep 2016.

6