optical-router-based dynamically reconfigurable photonic access network

Vol. 8, No. 1 / January 2009 / JOURNAL OF OPTICAL NETWORKING 51

Optical-router-based dynamicallyreconfigurable photonic

access network

Rajeev Roy,1,2,* Gert Manhoudt,2 and Wim van Etten1

1Telecommunication Engineering, University of Twente, P.O. Box 217, Enschede,7500 AE, The Netherlands

2AimValley B.V., P.O. Box 2194, Hilversum, 1200 CD, The Netherlands*Corresponding author: [email protected]

Received June 30, 2008; revised October 26, 2008;accepted November 10, 2008; published December 18, 2008 �Doc. ID 98116�

An optical-router-based dynamically reconfigurable photonic access networkis described. The network is a dense wavelength division multiplex overlayover legacy time division multiplexed passive optical networks (PONs), eachoperating in its native format. A single such fiber plant deployment can sup-port multiple logical PONs, each operating on a unique wavelength pair. Thisstack of logical PONs can be configured for optimal bandwidth delivery to theend user by controlling the number of optical network units (ONUs) supportedin every such logical PON. Microring-resonator-based optical routers are usedto dynamically manage the logical network topology to control the number ofONUs supported per logical PON. We present techniques to calculate an op-timal network configuration and its use in a typical evolving access scenario.The network view from a data transport perspective and from a control andmanagement perspective is also presented. © 2008 Optical Society ofAmerica

OCIS codes: 060.4250, 060.4510.

1. IntroductionThe use of fiber is a promising means to alleviate bandwidth bottlenecks in access net-works. Passive optical networks (PONs) have emerged as the preferred means todeploy fiber to the home/business/curb (FTTx). Current-generation PONs use a broad-cast transmission in the downstream direction from the head end (HE) to the enduser. The end user discerns the information broadcast and selectively picks up infor-mation intended for it. Upstream communication from the end user to the HE is, onthe other hand, a many-to-one communication on a single shared medium andrequires arbitration. Time division multiplexing (TDM) is one means of arbitration,and all current-generation commercial PON deployments use it.

The ever-increasing thirst for bandwidth is now driving the quest for technologydevelopment of platforms that can sustain this requirement. Use of dense wavelengthdivision multiplexing (DWDM) technologies have long been established in core andmetro networks as a means to increase the aggregate capacity of existing fiber plants.Current-generation TDM PONs, such as the IEEE Ethernet Passive Optical Networks(EPONs) [1], the ITU-T Gigabit-capable Passive Optical Networks (GPONs) [2], andthe Broadband optical access systems based on Passive Optical Networks (BPONs)[3], use WDM only as a means to achieve bidirectional (upstream and downstream)transmission over a single fiber and stop short of using it as a means to increase theaggregate capacity. The use of DWDM in PONs to increase the aggregate capacity hasseen significant interest in recent times [4]. The increased aggregate capacity of thenetwork can be used for increasing the number of end users, to increase the band-width per user, or a judicious combination of both. Recent surveys show the need tohave in addition a future-proof infrastructure [5] in access networks. The broadbandphotonics project (BBP) under the consortium of Freeband [6] projects aims to designsuch an optical-router-based agile and dynamic access network. The network is pro-posed as an upgrade path to existing TDM PONs operating in their native format.Bandwidth availability to the end user can be optimized by reconfiguring the networksuch that the hot spots can be allocated additional bandwidth from lean users [7].Related work requires additions and modifications to the specifications of protocol of

1536-5379/09/010051-26/$15.00 © 2009 Optical Society of America


operation for existing TDM PON specifications [8,9]. In this project we seek to keepthe operation of the TDM PON protocol unmodified and transparent to the DWDMoverlay. The network can be viewed as a logical stack of conventional PONs where theuse of optical routers enables change of the logical configuration of the network to doload balancing and consequently to optimize the bandwidth availability.

Section 2 introduces the concept of using such a dynamically reconfigurable accessnetwork. One such means to achieve dynamic reconfigurability in access networks isto use a BBP-like network. A description of this network has also been provided inSection 2. Section 3 presents a case study showing the motives for using a reconfig-urable access network. Section 4 discusses the methodology based on a linear pro-gramming (LP) formulation to calculate an optimal configuration of the network. Sec-tion 5 presents the use of this LP formulation in a typical evolving scenario tocalculate the network configuration of a BBP-like network. Section 6 describes differ-ent perspectives of the BBP network, first from a data perspective where an EPON istaken as the underlying TDM PON protocol and second from a control and manage-ment perspective. The paper is concluded in Section 7 with a summary of previoussections and a discussion.

2. Dynamically Reconfigurable Access Network2.A. Concept of Dynamic Reconfiguration in an Access NetworkThe overdeployment of fiber plant resources is an aspect that is of concern to networkoperators. An adaptive and scalable network can partly offset the need to overdeployresources and allow for a logical growth plan for increase in capacity of the network.Figure 1 presents a schematic depicting an access network as the proverbial last-mileinterface between the HE and customer premises equipment (CPE). The network canbe viewed as a switch that connects HE-based optical line termination units (OLTs) toCPE-based optical network units (ONUs). Figure 2 illustrates standard PON deploy-ments as an interconnect between the OLTs and the ONUs; the network is depicted asa static switch where every ONU is always associated with one particular OLT. Toincrease the aggregate capacity of the network, additional PON deployments have tobe made. Each one of the multiple PONs deployed is independent of one another, andfree capacity in any one of the PONs cannot be used in a practical way in any otherPON where there might be need for more bandwidth. To optimize the network deploy-ment, a priori knowledge of bandwidth usage patterns would be needed, and it wouldbe difficult to cater to an evolving scenario where more new users are added or exist-ing users have a changed demand pattern.

A scalable and reconfigurable network, on the other hand, would meet require-ments of increasing the aggregate capacity in a phased manner and achieve more uti-lization of the installed network capacity. Use of DWDM to support multiple PONs ina single fiber plant deployment is an obvious choice to make the network scalable,where an OLT and multiple ONUs operating on a single wavelength pair would forma logical PON. Multiple wavelength pairs would then support numerous logical PONson the same fiber plant deployment. Unlike the static configuration where an increasein capacity would mean use of additional fiber resources, herein an increase in capac-ity would mean deployment of additional wavelength pairs without any corresponding

Fig. 1. Access network connecting HE-based OLTs to CPE-based ONUs.


increase in use of fiber resources. A single physical fiber plant deployment would inturn conceptually facilitate connectivity between every OLT and every ONU in thenetwork. Figure 3 illustrates the network as a dynamic switch in which every OLTcan be associated with any of the ONUs. If the OLTs operate on fixed wavelengthpairs, wavelength-agnostic ONUs would associate with any particular OLT based onthe add–drop wavelength pair toward it. The OLT–ONU association can be made tem-porally dynamic by changing the add–drop wavelength pair toward it. This wouldrequire the use of optical routers that can do a selective add–drop of wavelengthstoward the ONUs.

The nominal bandwidth available to an ONU in a logical PON depends on the num-ber of ONUs supported by that logical PON. Increasing the number of ONUs in thatPON decreases the nominal bandwidth available per ONU, and decreasing the num-ber of ONUs, on the other hand, increases it. The element of reconfiguration thusallows for optimal load balancing by configuring the network for maximal bandwidthavailability to the end user. Figure 4 illustrates the concept with just two wavelengthpairs and two OLTs supporting two logical PONs, the green PON and the blue PON.The optical router is configured to change the association of ONU 5 from the bluePON to the green PON, thus increasing the bandwidth available to ONU 1 left on theblue PON.

2.B. Broadband Photonics NetworkFigure 5 illustrates the BBP network schematic. The network is one way to imple-ment a dynamic switch in which a geographically spaced out community can be served

Fig. 3. Multiple logical PON deployments as a dynamic switch between OLTs and ONUs.

Fig. 2. Multiple PON deployments as a static switch between OLTs and ONUs.


by a single fiber plant deployment supporting multiple logical PONs. A physical ringconnects the HE to multiple remote nodes (RNs). Each RN subtends multiple CPEs[10], and the network provides a redundancy in the HE–RN connection with toleranceup to a single fiber break [11]. The logical topology of the network retains the archi-tecture of standard PONs. Figure 6 illustrates the logical connectivity between theHE and a diverse set of CPEs. The HE location supports multiple OLTs operating onstandard PON specifications with modified optics to transmit on the ITU-T50/100 GHz grid in the C-band. The HE in addition transmits an equal number ofcontinuous wave (cw) lasers on the ITU-T 50/100 GHz grid in the C-band. Every suchset of unique wavelength pairs thus supports a logical PON. The RN locations housemicroring-resonator-based reconfigurable optical add–drop multiplexers (ROADMs)that can selectively add–drop wavelength pairs toward any of the CPEs [12]. TheCPEs house an ONU that operates on standard PON specifications with optics modi-fied to receive downstream transmission in the C-band and to use a reflective semi-conductor optical amplifier (RSOA) to modulate the received cw transmission forupstream communication. The ONUs are thus wavelength agnostic and associate with

Fig. 4. Access network depicted as a two-stage switch with only two OLTs and five ONUs withONU 5 changing association from OLT 1 to OLT 2.

Fig. 5. BBP network schematic.


any OLT depending on the wavelength add–drop toward it. The wavelength pair onwhich any one ONU operates is thus decided at the HE. The network allows for use ofany PON technology or multiple technologies to be operated on a single fiber plantdeployment.

3. Case for Reconfigurability in an Access NetworkAccess networks can show significant variation in diurnal demands [13]. Even duringa particular period of day, there can be variation in the usage pattern by differenttypes of users. We consider a sample user base to be served by an access network andmotivate the need to have dynamic reconfigurability in the network such that an opti-mal load balancing can be done in an evolving demand situation.

3.A. Demand ProfilingStudies of typical user profiles are available for the United Kingdom [14]. This studyis used as a basis to create a demand profile in a typical developed country for peopleaccessing the Internet for work and leisure. Table 1 lists the user categories adaptedfrom [14]. Eight different user categories (A, B, C1, C2, D1, D2, E, and F) are defined.A total of 128 households is sought to be serviced by an access network deployment.The typical applications used over a day by this customer base are listed in Table 2.

Fig. 6. Logical connectivity between HE and CPEs. Two sets of logical PONs are illustrated.

Table 1. Categories of Users and Relative Numbers of ONUs

Category Description Number of ONUs (%)

A Single adult, retired 19 (14.8)B Two adults, retired 15 (11.7)

C1 Single male, working 14 (10.9)C2 Single female, working 8 (6.3)D1 Two adults, empty nesters 17 (13.3)D2 Two adults, working 17 (13.3)E Two adults with children 30 (23.4)F Single parent 8 (6.3)


The applications considered and the bandwidth usage thereof are estimated for thenear future. Future applications can be speculative; however, remote data backup,premises, and health monitoring are likely to be in common use. Applications such aspeer-to-peer file sharing and remote working are already bandwidth-intensive applica-tions to reckon with today.

The user bandwidth demand during the day will vary depending on the applicationusage. Typical peaking of demand takes place around 18:00 hours to 21:00 hours[15,16]. A list of applications from Table 2 that are thought to be used by the differentuser categories in this busy hour is listed in Table 3. Each user in the different usercategories is thus considered to use one or more applications during this period. Theuse of applications by particular user groups is a subjective evaluation that can thenbe defined as a measurable parameter in terms of bandwidth usage. The number ofconcurrent applications that are used by any user category is modeled with a heavy-tailed generalized Pareto distribution for each of the categories. The presumption isthat most households use fewer concurrent applications, whereas only a few house-holds use many applications in parallel. A bandwidth demand profile called Scenario 1is generated from an application usage profile illustrated in Fig. 7 (shape parameter=2 and scale parameter=1 for generalized Pareto distribution). The graph representsthe number of ONUs (each household using only one ONU) as a function of the num-ber of applications used. Figure 8 illustrates the dynamic range of demands by the dif-ferent user categories with a weighted mean demand indication by each of the usercategories.

3.B. Static Network ConfigurationIn a static network configuration the demand is sought to be met by multiple stan-dard IEEE EPON deployments. A single network group of 32 ONUs is served by oneEPON deployment; for 4 such network groups (128 ONUs) a deployment of 4 EPONsis considered. A network group thus is used in context to define a group of usersserved by a single such EPON deployment. We consider two distinct compositions ofthe user types in the network groups. In Network Profile 1, all user categories are uni-formly distributed across the four network groups. In Network Profile 2, on the other

Table 2. Applications and Bandwidth Requirements

Type Application

DownstreamBandwidth(Mbits/s)

UpstreamBandwidth(Mbits/s)

Voice PSTN quality call 0.03 0.03CD quality call 0.13 0.13DAB/CD quality audio streaming 0.19High-quality digital audio streaming 6.00High-quality digital audio fast download (attwice real time)

12.00

Videoa CIF quality web conferencing 0.32 0.32SDTV quality web conferencing 0.38 0.38SDTV video streaming (MPEG 4) 2.00HDTV video streaming (MPEG 4) 9.00HDTV download (at twice real time) 18.00

Data General web browsing and e-mail download andupload

2.00 2.00

File download and upload (10 MB in 30 s) 2.67 2.67File download and upload (50 MB in 30 s) 13.33 13.33Peer-to-peer download and upload (60 MB in30 min)

0.27 0.27

Remote backup of data (400 GB in 30 days) 1.25Others Remote monitoring of health 0.00027

Remote premises monitoring (5 channel SDTVquality CCTV)

0.69

Online gaming 2.00 2.00

aNote: Video category calls include audio content in bandwidth calculations.


hand, there is a marked skew in the composition with each user group concentrated inone network group. Figures 9 and 10 illustrate the two network profiles.

Fig. 7. Application usage profile by all user categories in Scenario 1.

Table 3. Applications Considered as Used by Different User Categories inPeriod of Interest

UserCategory Application

Concurrent Applications in Period of Interesta

1 2 3 4 5 6 7 8

A HDTV video streaming � � � �

DAB/CD quality audio strm. � � �

SDTV quality web conf. � �

PSTN quality call �

B HDTV video streaming � � � �

General web brows. & email � � �

SDTV quality web conf. �

C1 HDTV video streaming � � � �

File dn/upload (50 MB in 30 s) � � �

HQ digital audio streaming � �

Online gaming �

C2 HDTV video streaming � � �

HQ digital audio streaming � �

General web brows. & email �

D1 HDTV video streaming � � � �

General web brows. & email � �

D2 HDTV video streaming � � � �

File dn/upload (50 MB in 30 s) � � �

General web brows. & email � �

E HDTV video streaming � � � � � �

SDTV video streaming � � � � � �

File dn/upload (50 MB in 30 s) � � � � �

HQ digital audio streaming � � �

DAB/CD quality audio strm. �

F HDTV video streaming � � �

SDTV quality web conf. �

aNote: Single instance, �; two instances, �.


On the basis of the application usage profile in Scenario 1 and the distribution ofuser types in the two network profiles, the aggregate bandwidth demand can be esti-mated for each of the four network groups. Figures 11 and 12 illustrate the aggregatedemands in the downstream direction for the two network profiles considered. In Net-work Profile 1, the aggregate free capacity across all deployments is 38.01%. The typi-cal network load for each of the EPON deployments is approximately 65.5% (the net-work load is calculated presuming a maximum downstream throughput of816 Mbits/s for an EPON with 32 ONUs) [17]. Network Profile 2, on the other hand,shows a marked skew in the demand, with the variation from 37% to 97% loading ofthe network. The limitations of a static network configuration become evident: eventhough from an aggregate network perspective there is 38% free capacity available, itwould be impractical to make use the free capacity from the lightly loaded networkgroups to meet demands for a loaded network group. Any increase in the demand ofusers because of change in the usage of applications or because new users have to beserved would severely degrade performance for some network groups. To mitigate thislimitation, a network operator might need to provision an additional EPON deploy-ment for this group to maintain a desirable quality of service (QoS). Figure 13 illus-trates the demand in the downstream direction with an additional PON deploymentfor users in Network Group 1 (now depicted as Network Group 1 and Network Group5). The aggregate free capacity of the whole network deployment increases to 50.40%.

Fig. 8. Downstream bandwidth demand range by user categories in Scenario 1.

Fig. 9. Network Profile 1.


The limitation of a static network configuration comes from not being able to use upthe free capacity in the network with practically implementable means. A dynamicconfiguration on the other hand can be used to do load balancing from a networkwideperspective where instead of multiple fiber plant deployments the network is a singledeployment and it is easier to use free capacity of the network where it is needed. Ananalogy in real-world terms can be thought of with the problem of parking a big car ina small parking garage. Even if there are several small parking garages available andthe cumulative space available across all the garages is enough to park the big car, itwould not fit in any one single garage. However, if all the garages could be configuredas a single big garage with flexible individual boundaries being redrawn to suit thecar to be parked, the available space can be more efficiently used both in terms of thenumber of cars and the size of cars that can be parked in the garage. A dynamic net-

Fig. 10. Network Profile 2.

Fig. 11. Aggregate downstream demand in Network Profile 1.


work would be analogous to the big garage with flexible boundaries that can beredrawn to suit the demand presented to it.

4. Methodology for Reconfiguration of the NetworkThis section introduces the techniques that are used to configure a dynamic networkfor optimal utilization of resources. The bandwidth available is treated as a resourcethat has to be allocated subject to some constraints. The objective herein is to achieveload balancing across a whole network deployment such that the maximum possibledemand can be met within certain constraints. A LP technique is used to calculate anoptimal configuration of the network to do bandwidth allocation [18,19]. The recon-figuration of the network can be seen as bandwidth allocation on an inter-PON scalewhere the configuration of the logical PONs is altered to make resources available

Fig. 12. Aggregate downstream demand in Network Profile 2.

Fig. 13. Aggregate downstream demand in Network Profile 2 with an additional EPONdeployment.


where they are needed. This is distinct from any intra-PON bandwidth allocationschemes that will be operational within the scope of an individual logical PON. Theconcept of inter-PON bandwidth allocation would need a service-aware architecturewhere a predictive demand requirement can be advertised by end users and the net-work resources are subsequently allocated in an optimal way [20].

A BBP-like network allows for two diverse paths of connectivity between an OLTand an ONU. Each OLT can have a clockwise downstream/anticlockwise upstream oran anticlockwise downstream/clockwise upstream communication with every ONU.The communication direction between the HE and every CPE is decided on the basisof minimizing the number of RN crossings to reach a CPE. A cost weight is then asso-ciated for every possible OLT–ONU association. There is a further cost weightrequired to provision any OLT. The eventual objective is to minimize the total cost ofbandwidth supply. The cost factors can be suitably defined to bias allocation of OLTsto ONUs if desired by the network operator. Once an OLT–ONU association is formed,the cost for this association can be reduced while increasing the cost for this particu-lar ONU to associate with any other OLT. This prebiases the ONU to remain associ-ated with the OLT with which it was originally associated before the network wasreconfigured. The cost parameter also decides whether a certain OLT–ONU associa-tion is possible; an infinite cost of supply would imply that the association is not pos-sible. This is used when a particular type of OLT cannot serve a particular type ofONU, for instance, if there is a mix in protocols of operation such as EPON and GPONor if there is a mix of speeds such as EPON and the upcoming 10 G EPON or if thenetwork operator would like to restrict certain OLT–ONU associations to a set of usertypes or user locations. In the current context it is presumed that all ONUs can beserved by all OLTs; however, there might be network configurations where some OLT–ONU associations are not possible because of physical limitations. The supply cost insuch cases is also infinite.

It has been shown with pricing models that end users can advertise for a desiredbandwidth and specify a tolerable relaxation in the demand. This gives flexibility tothe network operator to provision a lesser bandwidth if the situation so demands. Theusers, on the other hand, benefit from lower tariffs if they are more flexible in therange specified for the bandwidth relaxation vis-à-vis the desired demand [21]. In thisformulation we try to allocate the desired bandwidth by the ONUs, and if it is aninfeasible or unbounded solution, then we iteratively relax the demand of the ONUsup to the level specified by it and then find a solution. If a solution is still not found byrelaxing the demands up to specified levels, the relaxation is increased in discreteunits for all ONUs till a solution can be found. The relaxation comes into play only ifthe original demand cannot be met or the network operator forces a relaxation in thedemand. The optimization routine tries to associate OLT with ONUs based on thebandwidth demands and the constraints listed. The parameter and variable declara-tion for the problem formulation are listed below, and the state diagram of a heuristicalgorithm to calculate the association of ONUs with OLTs is presented in Fig. 14:

Num�ONU Number of ONUs supported in the networkNum�OLT Number of OLTs supported in the networkOLT�DS�capj Maximum bandwidth that can be provi-

sioned in the downstream by the jth OLTOLT�US�capj Maximum bandwidth that can be provi-

sioned in the upstream by the jth OLTBW�Demand�DSi Desired bandwidth demand of the ith ONU

in the downstream directionBW�Demand�USi Desired bandwidth demand of the ith ONU

in the upstream directionBW�Relax�DSi Relaxation advertised in the downstream

bandwidth demand for the ith ONUBW�Relax�USi Relaxation advertised in the upstream band-

width demand for the ith ONUCostji Cost weight to provision an association

between the jth OLT and the ith ONUCost�OLTj Cost weight to provision the jth OLT


Num�ONUj Number of ONUs that can be associatedwith the jth OLT

Penalty Penalty weight for shortfall in allocation ofbandwidth for an ONU

OLT�ONUassocji Boolean variable to denote association of ithONU with jth OLT

Com�OLTj Boolean variable to denote whether the jthOLT is provisioned

DS�allocji Allocation in the downstream from the jthOLT to the ith ONU

US�allocji Allocation in the upstream from the jth OLTto the ith ONU

ONU�DSi Total allocation in the downstream for theith ONU

ONU�USi Total allocation in the upstream for the ithONU

ONU�DS�relaxi Relaxed allocation in the downstream forthe ith ONU

ONU�US�relaxi Relaxed allocation in the upstream for theith ONU

The major constraints are defined as follows:

(a) Each ONU can be supplied by only one OLT:

�j

OLT � ONUassocji = 1, ∀ i. �1�

(b) Each OLT can provision bandwidth only if it is provisioned:

OLT � ONUassocji � Com � OLTj, ∀ i,j. �2�

(c) The total supply to ONUs is equal to the demand that can be met:

ONU � DSi = BW � Demand � DSi − ONU � DS � relaxi, ∀ i, �3�

ONU � USi = BW � Demand � USi − ONU � US � relaxi, ∀ i. �4�

(d) Total allocation to ONUs is the sum of all allocations:

ONU � DSi = �j

DS � allocij, ∀ i, �5�

ONU � USi = �j

US � allocij, ∀ i. �6�

Fig. 14. State diagram for reallocation of ONUs.


(e) Each OLT can only supply as much bandwidth as it can support in the down-stream and the upstream directions:

�i

DS � allocji � OLT � DS � capj, ∀ j, �7�

�i

US � allocji � OLT � US � capj, ∀ j. �8�

(f) Each OLT can supply only if the ONU is associated with the OLT:

DS � allocji � OLT � ONUassocji*OLT � DS � capj, ∀ i,j, �9�

US � allocji � OLT � ONUassocji*OLT � US � capj, ∀ i,j. �10�

(g) Each OLT can support a limited number of ONUs:

�i

OLT � ONUassocji � Num � ONUj, ∀ j. �11�

The objective function is defined as minimizing the overall supply cost:

min��i,j

�DS � allocji + US � allocji�*Costji + �j

Cost � OLTj*Com � OLTj

+ �i

�ONU � DS � relaxi + ONU � US � relaxi�*Penalty� �12�

5. Case Study with Dynamic Network ConfigurationWe take the same customer base of 128 ONUs to be served by a BBP-like dynamicallyreconfigurable network. The referral to network groups, as done earlier, is redundantin this case as potentially all ONUs can be served by any of the OLTs. However, westill refer to ONUs with reference to the erstwhile Network Groups 1–4, in NetworkProfile 2, to simplify the depiction in text and figures.

In this section we consider an evolving scenario by changing the application usageprofile for all the user categories. The new scenarios (Scenario 2–4) are created bychanging the scale parameter of the generalized Pareto distribution function from 2through 4. The function is used to generate the distribution of the number of ONUsfor the number of applications used by each user category. As the application profilechanges, the demands of the ONUs are going to change. Figures 15–17 illustrate Sce-narios 2–4, wherein increasing users are using multiple concurrent applications. Thedemand in the downstream direction per ONU from Scenario 1 to 4 is illustrated inFigs. 18–21. Aggregate bandwidth demands depend on the number and type of appli-cations used by different end users. As more users tend toward more concurrent appli-cations, the aggregate demand increases. The initial demand profile as illustrated inFig. 17 required the use of at least four standard EPONs, and it was discussed that afifth deployment might be necessary to accommodate such an evolving scenario.

Using techniques discussed earlier, we try to associate the ONUs with OLTs froman initial condition where there is no specific bias for any particular ONU to be asso-ciated with any particular OLT; the bias is decided by the cost weight factor. The TDMunderlay for the PON operation is taken as EPON. All OLTs are assumed to have amaximum downstream throughput capacity of 800 Mbits/s and a maximum upstreamthroughput capacity of 750 Mbits/s. The initial demand can be met with just threewavelength pairs. In all the cases the upstream demand is not a limiting factor. As thescenario evolves from Scenario 1 to Scenario 2, ONUs are reallocated such that thestated requirement is met. In this case again three wavelength pairs suffice to meetthe demand. A total of six reallocations take place as a result of change in userdemands. Once the ONUs are reallocated to other OLTs, the cost to associate themwith any other OLT is doubled. This is to reduce the possibility of repeated realloca-tions for ONUs that have already been reallocated once. From Scenario 2 to 3 a totalof four reallocations take place. As Scenario 4 evolves, the fourth wavelength pair isrequired to meet the increased demand. A total of 17 reallocations take place fromScenario 3 to 4. Figures 22–25 illustrate the OLT-to-ONU allocation with the demand


profile changing from Scenario 1 to Scenario 4 for sample Network Group 1. If the net-work operator is constrained to not using the fourth wavelength pair for this set of128 users, the bandwidth allocation to the ONUs is relaxed by the demand advertised.Figure 26 illustrates a case wherein all ONUs have advertised for a 10% relaxation.As Scenario 4 evolves from Scenario 3, instead of using a fourth wavelength pair, thedemand of ONUs is relaxed up to the advertised limit, and the OLT–ONU associa-tions are calculated. The figure shows the OLT–ONU associations for Network Group3 in this scenario, without any relaxation and with 10% relaxation advertised. Thenumber of ONUs associated with the OLTs for each of the scenarios is illustrated inFig. 27. Allocation by the OLTs in each of the scenarios in the downstream is illus-trated in Fig. 28. All optimization routines are solved with commercially availableAIMMS software with a XA14 solver. The load profiles are generated and analyzedwith MATLAB programs.




Typical hold times for media-based services will be long during the presumed busyhour [22]. We try to analyze the network from a perspective of handling short-duration high-bandwidth requirements from additional new sources, for example, ifsomeone requires high bandwidth for download of files for a few minutes. Two suchcases are considered: in the first case new users require a chunk of 50 Mbits/s for afinite duration during the period of interest, and in the second case the requirement isfor 100 Mbits/s for a finite duration. It is also assumed that the fourth wavelengthpair is now configured for use as a broadcast channel with 20 HDTV and 30 SDTVchannel transmissions. This allows for only 560 Mbits/s of free capacity in the down-stream direction by potential new users on this wavelength pair. The bandwidth allo-cation to the incumbent ONUs is relaxed by a certain percentage, and the optimiza-tion routine is used to seek additional space that can be created for new users. The


Fig. 18. ONU demands in Scenario 1.


service requests and processing are modeled as an M /M /m /m /s queue [23], where s isthe number of such blocks of channels that can be accommodated in the network. Theblocking probability is calculated using the Engset formula [24]. Figure 29 illustratesthe service blocking probability as a function of traffic generated per ONU for a casewhere only 32 new ONUs are requesting services. Figure 30 presents a different pic-ture where we analyze the additional new ONUs that can be accommodated with aservice blocking probability of 1%, as a function of the traffic generated per ONU.

6. Perspective Views of the BBP NetworkThe BBP network uses the IEEE EPON as the TDM underlay for PON specifications.This section presents the BBP network from data transport and control and manage-ment perspectives. The first view is an Ethernet perspective where the network is




examined at the level of transport of an Ethernet frame to and from a content–serviceprovider in the edge to the end user. The methodology to do network reconfigurationwas described in Section 5; however, this requires additional control and managementthat is beyond the scope of existing TDM protocols. In this network we have an out-of-band channel that is used to do the control and management of the network. The sec-ond view presented in this section deals with this perspective where we examine thenetwork from a control and management point of view.

6.A. Network from an Ethernet PerspectiveThe BBP network is an access network that stretches between the OLTs and ONUs;however, this network forms a part of the Ethernet access network that spans from


Fig. 22. OLT-to-ONU association for ONUs for Network Group 1, Scenario 1.


the server of a service or content provider to the equipment such as a personal com-puter, voice-over-IP (VoIP) phones, digital TV, etc. in the end-user network. Figure 31illustrates the Ethernet perspective of the network from the content–service providerto the end user. The EPON segment is the actual reconfigurable access network underconsideration. Forwarding in the EPON segment is different from that done in a stan-dard local area network (LAN) segment. Unlike a LAN segment, the destinationaddress (DA) field of the Ethernet frame is not used, but instead the logical link iden-tifier (LLID) is used for forwarding. Figure 32 illustrates the buildup of the EPONframe that indicates modification of the preamble field to accommodate the LLID. To




emulate a point-to-point link in a point-to-multipoint scenario, each OLT maintains aunique media access control (MAC) instantiation for every ONU associated with it (anadditional MAC is used for broadcast in the downstream direction). In the down-stream direction the OLT writes a LLID in the preamble of each frame that identifiesthe target ONU. ONUs only listen to frames carrying LLIDs assigned to them (and toframes with the multicast mode bit set to 1) and drop all other frames, irrespective ofthe MAC DA in the frame. In the upstream direction every ONU writes its own LLIDinto the preamble of the outbound frame. The OLT that maintains multiple MACinstantiations uses the LLID as a virtual port to emulate a point-to-point service toeach ONU. The OLT learns the MAC addresses of devices connected to an ONU by


Fig. 26. OLT-to-ONU association for ONUs for Network Group 3, without any and with 10% re-laxation advertised by ONUs.


Fig. 27. Number of ONUs associated with OLTs for different scenarios.

Fig. 28. Allocation in downstream per OLT for different scenarios.

Fig. 29. Blocking probability as a function of traffic generated (blocks of 50 and100 Mbits/s) for 32 ONUs.


inspecting the source address (SA) of the inbound upstream frames and associatesthem with the LLID of the particular ONU that sent the frames. In the downstreamdirection, the OLT attaches the same LLID for the outbound frames with the sameDAs [1,16]. The LLID–MAC association table is derived from the IEEE 802.1D auto-matic learning feature [25].

To distinguish Ethernet streams belonging to different end users over the sameEthernet link, the virtual LAN (VLAN) tagging mechanism is used as defined in IEEE802.1ad-2005 [26]. The OLT has the capacity to attach a VLAN tag to any frame inthe upstream direction and remove it in the downstream direction. Figure 33 illus-trates the layout of the VLAN tag and its position in the Ethernet frame. In the

Fig. 30. Number of additional ONUs that can be served by the network with a blockingprobability of 1%.

Fig. 31. Ethernet view of the network.


upstream direction the OLT performs a simple LLID-to-VLAN translation. VLANs aretagged to a particular ONU and not to a particular LLID. The use of VLANs distin-guishes the flow of traffic to and from different ONUs and avoids direct ONU-to-ONUcommunication. With the use of VLANS, the HE bridge forwarding between ONUsbecomes impossible in the entire access network as the IEEE 802.1 specification stipu-lates that MAC addresses learned per VLAN can never be forwarded between VLANseven if the DA is known in some other VLAN. The use of VLANs thus also forcesintra-ONU traffic to be routed via the IP server of the service–content provider thatcan then be monitored and billed. In the downstream direction the OLT performs aVLAN-to-LLID translation. The bandwidth delivery to an ONU is monitored in termsof the bandwidth allocated to VLANs associated with any one ONU. This representsthe bandwidth pipe flowing to and from any one ONU with the OLT it is associatedwith.

6.B. Network from a Software and Control Plane PerspectiveThe management and maintenance of the network is operated in a software platformimplemented on a PC at the HE. The elements controlled in the HE are the following:

• A HE bridge that is used to distribute the downstream traffic over different OLTsand to aggregate the upstream traffic.

• OLTs in the HE to check ONU registrations, allocation of LLIDs to ONUs, andmonitoring of multipoint control protocol (MPCP) entities.

• Calculation of an optimized network configuration and association of OLTs toONUs based on bandwidth demand. Maintenance of a resource manager to translateservice requests to bandwidth demand requests as an aggregate of VLANs to each ofthe ONUs.

• Monitoring and configuration of the RNs to add–drop different wavelengthstoward the ONUs depending on the ONU-to-OLT assignment process.

Fig. 32. EPON frame buildup.

Fig. 33. VLAN tag layout and position.


• The maintenance of the LLID–MAC associations of ONUs and migration of theseentries once an ONU is served by a different OLT.

• The maintenance of the VLAN to port identification in the HE bridge and migra-tion to another port when the ONU is served by a different OLT.

The HE-based controller can communicate directly with the OLTs and theHE-based switch through a management LAN interface. The HE control and manage-ment channel is implemented in an out-of-band bidirectional communication channelbased on 1490/1310 nm optics [27]. Figure 34 illustrates the communication stack forthis. The HE-to-CPE communication is realized through a data plane, a LAN connec-tion is made from the HE controller to one of the traffic ports of the HE bridge, andthe management traffic is routed to the ONUs. The OLTs at the HE also communicatewith the associated ONUs via existing MPCP and operations, administration, andmaintenance (OAM) protocols and in addition from the OLT application layer to theONU application layer. Figure 35 illustrates the communication stack for this. Themanagement model of the network with k OLTs, m RNs, and n ONUs is illustrated inFig. 36. In this model, each ONU can support one or more connections with a uniqueidentity. The connections are associated with a unique LLID issued by the OLT, aVLAN identifier, and a set of service level agreement (SLA) parameters. In this modeleach ONU can be associated with multiple end users and each end user can opt forone or more content–service providers.

7. ConclusionA description of a dynamically reconfigurable optical-router-based access network hasbeen presented. The network makes use of optical routers to change the logical con-figuration of the network to do load balancing in the network. A varying range ofdemands from a diverse category of end users can be met by dynamically adapting thenetwork configuration. A technique to calculate the network configuration in an evolv-

Fig. 34. Communication protocol stack between HE controller OLTs and HE controller RNs.

Fig. 35. Communication protocol stack between OLT and ONUs.


ing demand scenario has been discussed. Use of the technique has been demonstratedwith the example of a use case scenario with multiple customer types and applicationusage profiles to be served by an access network deployment. The dynamic configura-tion has been compared vis-à-vis a static network configuration meeting similardemand from the users. It has been shown that a dynamic network configurationallows for benefits over a static network configuration in terms of supporting spikes indemands, in terms of an increase in potential number of end users, and in terms ofscalability by increasing the capacity of the network on a requirement basis. For thesample use case, a static configuration required the use of the equivalent of a greaternumber of wavelength pairs than an equivalent deployment of a BBP-like network.The network also offers the possibility of supporting multiple existing TDM PON tech-nologies in their native format in a single fiber plant deployment. For instance, asingle HE can be used to deploy both EPON- and GPON-based equipment. The net-work further facilitates the selective upgrade of users to higher speeds as and whensuch specifications are standardized. Redundancy is provided in the connectivitybetween the HE location and the RN locations for increased fault tolerance. A dataperspective of the network has been provided for the BBP network where EPON istaken as the underlying TDM PON protocol. The use of the DWDM overlay necessi-tates control and management of elements that are beyond the scope of TDM PONoperations in their native format, and an additional out-of-band control and manage-ment plane has been implemented on the basis of an ad hoc IP connectivity betweenthe network elements. The network perspective from a control and management viewhas been discussed.

Appendix AThe generalized Pareto distribution for location parameter �, shape parameter �, andscale parameter � is defined as

f��,�,��x� =1

��1 +

��x − ��

��−1/�−1�

. �A1�

Appendix BFor a M/M/m/m/s queue with a mean arrival rate of � and an exponentially distrib-uted hold time of 1/� with s servers the service blocking probability of P is given by

Fig. 36. Management model of the broadband photonics network.

B


PB =

1

m!� �

��m

�n=0

m �s − 1 − m�!sm−n

n!�s − 1 − n�! � �

��n

. �B1�

AcknowledgmentsThe work was funded by the Dutch Ministry of Economic Affairs through the BSIKBroadband Photonics project under contract BSIK 03025. The authors acknowledgethe technical input of the project group. Special thanks are due to Anne Roc’h for tech-nical help with MATLAB simulations and discussions on solving optimization rou-tines.

References and Links1. IEEE 802.3-2005, Section 5, “Carrier sense multiple access with collision detection (CSMA/

CD) access method and physical layer specifications,” IEEE Standard.2. ITU-T G.984.1-6, “Gigabit-capable passive optical networks,” ITU-T Standards.3. ITU-T G.983.1-3, “Broadband optical access systems based on passive optical networks

(PON),” ITU-T Standards.4. A. Banerjee, Y. Park, F. Clarke, H. Song, S. Yang, G. Kramer, K. Kim, and B. Mukherjee,

“Wavelength-division-multiplexed passive optical network (WDM-PON) technologies forbroadband access: a review [Invited],” J. Opt. Netw. 4, 737–758 (2005).

5. “Frost & Sullivan: bandwidth demand, future proofing driving optical deploymentsglobally,” S. M. Hardy, ed., Lightwave Direct Newsletter, May 2008, www.lw.pennet.com.

6. Freeband project web site, www.freeband.nl.7. T. Koonen, M. G. Larrode, J. Yang, A. Ng’oma, G. Rijckenberg, P. Urban, H. Yang, H. van

Den Boom, and H. de Waardt, “Recent research results on flexible optical access andin-building networks,” in Proceedings of the 12th European Conference on Networks andOptical Communications, D. W. Faulkner, R. Neat, and E. Vannin, eds. (Acreo AB, 2007),pp. 153–160.

8. F. T. An, K. S. Kim, D. Gutierrez, S. Yam, E. Hu, K. Shrikhande, and L. G. Kazovsky,“SUCCESS: a next-generation hybrid WDM/TDM optical access network architecture,” J.Lightwave Technol. 22, 2557–2569 (2004).

9. M. P. McGarry, M. Reisslein, M. Maier, and A. Keha, “Bandwidth management for WDMEPONs,” J. Opt. Netw. 5, 637–654 (2006).

10. P. Urban, E. J. Klein, L. Xu, E. G. C. Pluk, A. M. J. Koonen, G. D. Khoe, and H. de Waardt,“1.25–10 Gbit/s reconfigurable access network architecture,” in Proceedings of the 9thInternational Conference on Transparent Optical Networks (IEEE, 2007), pp. 293–296.

11. R. Roy and W. van Etten, “Design of a survivable multi-wavelength photonic accessnetwork,” in Proceedings of the 2nd International Conference on Access Networks (IEEE,2007), pp. 1–6.

12. E. Klein, P. Urban, G. Sengo, L. T. Hilderink, M. Hoekman, R. Pellens, P. van Dijk, and A.Driessen, “Densely integrated microring resonator based photonic devices for use in accessnetworks,” Opt. Express 15, 10345–10355 (2007).

13. K. Fukuda, K. Cho, and H. Esaki, “The impact of residential broadband traffic on JapaneseISP backbones,” Comput. Commun. Rev. 35, 15–21 (2005).

14. Broadband Stakeholder Group, “Predicting UK future residential bandwidthrequirements,” www.broadbanduk.org.

15. Amsterdam Internet Exchange, www.ams-ix.net.16. Munich Internet Exchange, www.inxs.de.17. G. Kramer, Ethernet Passive Optical Networks (McGraw-Hill, 2005), pp. 173–181.18. J. Kalrath and J. M. Wilson, Business Optimization: Using Mathematical Programming

(Macmillan Business, 1997).19. J. Bisschop, “AIMMS: optimization modelling,” www.aimms.com.20. V. Souza, B. Tremblay, P. Karlsson, J. R. I. Riu, K. Persson, Z. Ghebretensae, and W. Zhao,

“A QoS enabled public ethernet access network,” in Proceedings of the 12th EuropeanConference on Networks and Optical Communications, D. W. Faulkner, R. Neat, and E.Vannin, eds. (Acreo AB, 2007), pp. 244–251.

21. A. Elmaghraby, A. Kumar, M. M. Kantardzic, and M. G. Mostafa, “Bandwidth allocation ina dynamic environment using a variable pricing policy,” in Proceedings of the 7thInternational Symposium of Computers and Communications (IEEE, 2002), pp. 589–594.

22. A. Erramilli, E. H. Lipper, and J. L. Wang, “Some performance considerations for massmarket broadband services,” in Proceedings of the 1st International Workshop onCommunity Networking Integrated Multimedia Services to the Home (IEEE, 1994), pp.109–116.

23. Y. Lu, F. Kuipers, M. Janic, and P. V. Mieghem, “E2E blocking probability of IPTV andP2PTV,” in Proceedings of the 7th International Conference on Networking (IFIP, 2008), pp.445–456.


24. P. van Mieghem, “Queuing models,” in Performance Analysis of Communications Networksand Systems (Cambridge, 2006), pp. 279–282.

25. IEEE 802.1D, “Media access control (MAC) bridges,” IEEE Standard.26. IEEE 802.1ad-2005, “Virtual bridged local area networks, amendment 4: provider bridges,”

IEEE Standard.27. R. Teune, R. Roy, and W. van Etten, “Demonstration of IP based control and management

for a reconfigurable photonic access network,” in Proceedings of the 10th InternationalConference on Transparent Optical Networks (IEEE, 2008), pp. 145–148.

optical-router-based dynamically reconfigurable photonic access network

Documents