IP Over WDM Networks Employing Renewable Energy Sources

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 1, JANUARY 1, 2011 3

IP Over WDM Networks Employing RenewableEnergy Sources

Xiaowen Dong, Taisir El-Gorashi, and Jaafar M. H. Elmirghani

Abstract—With network expansion, the energy consumption andCO� emissions associated with networks are increasing rapidly. Inthis paper we propose an approach for energy minimization in IPover WDM networks and furthermore propose the use of renew-able energy to further reduce the CO� emissions at a given en-ergy consumption level. We develop a Linear Programming (LP)model for energy minimization in the network when renewable en-ergy is used and propose a novel heuristic for improving renew-able energy utilization. Compared with routing in the electroniclayer, routing in the optical layer coupled with renewable energynodes significantly reduces the CO� emission of the IP over WDMnetwork considered by 47% to 52%, and the new heuristic intro-duced hardly affects the QoS. In order to identify the impact ofthe number and the location of nodes that employ renewable en-ergy on the non-renewable energy consumption of whole network,we also constructed another LP model. The results show that thenodes at the center of the network have more impact than othernodes if they use renewable energy sources. We have also inves-tigated the additional energy savings that can be gained throughAdaptive Link Rate (ALR) techniques where different load depen-dent energy consumption profiles are considered. Our optimizedREO-hop routing algorithm with renewable energy and ALR re-sults in a maximum energy saving of 85% (average of 65%) com-pared to a current network design where all nodes are statically di-mensioned for the maximum traffic in terms of IP ports and opticallayer and hence consume power accordingly. Furthermore, whenall the nodes have access to typical levels of renewable power weshow that the associated reduction in non-renewable energy con-sumption reduces the network’s CO� emissions by 97% peak, 78%average.

Index Terms—Adaptive link rate, IP over WDM, linear pro-gramming, renewable energy.

I. INTRODUCTION

I N the last ten years, the bandwidth of the Internet has grownby at least 50 to 100 times leading to an increase in the

energy consumption. Following the increase in the networkssize, the equipment (such as severs, amplifiers, routers, storagedevices and communication links) power consumption has in-creased rapidly [1]. Today the energy consumption of networksis a significant contributor to the total energy demand in manydeveloped countries; for example, in 2005 the energy consump-tion of the Telecom Italia network was more than 2 TWh which

Manuscript received May 11, 2010; revised September 17, 2010; acceptedOctober 04, 2010. Date of publication October 11, 2010; date of current versionDecember 16, 2010. This work was supported by the Engineering and PhysicalSciences Research Council, U.K.

The authors are with the School of Electronic and Electrical Engineering,University of Leeds, Leeds LS2 9JT, U.K.

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JLT.2010.2086434

is about 1% of the total Italian energy demand [2]. In the winterof 2007, British Telecom became the largest single power con-sumer in the UK accounting for 0.7% of the total UK’s energyconsumption [3]. Therefore, with the increase in data rates, thebottleneck facing the Internet’s expansion will probably becomeits energy consumption. In practice, 228 grams of CO approx-imately are produced by a network component that consumes1 kWh of traditional electricity power produced from coal ornatural gas [4]. A family vehicle typically emits 150 g/km CO ,therefore in a year a 1 kW router port contributes CO pollu-tion equivalent to approximately 13 k km journeys in a familyvehicle. If the network can be designed such that it eliminates1 kW non-renewable energy consumption, then this will lead toa significant reduction in CO emissions of about 2 tons everyyear.

Recently significant research efforts have been focused onreducing the energy consumption of ICT networks. There is asignificant literature body on power-awareness in mobile ad hocand wireless networks [5], [6] and computer architecture [7],[8]. However, many challenges need to be addressed to developand deploy energy efficient wire-line networks.

Gupta et al. [9] pioneered the introduction of the concept of“greening the internet” in 2003. Most of the existing work onenergy saving has considered local adaptation by implementinghardware-based techniques, such as sleep cycles and rate adap-tation. In [9], it is proposed to save energy by allowing networkinterfaces and other components to sleep when they are idle.In [10] two energy saving schemes are proposed. In the firstscheme, traffic is shaped into small bursts at the edge nodes toallow downstream line cards to sleep in between packet bursts.The second scheme is based on the fact that operating devices atlower frequencies and/or voltages can significantly reduce en-ergy consumption. The latter scheme is an important step to-wards power-proportional networking hardware.

Energy consumption and heat dissipation are increasingly be-coming primary objectives in router design. In [11], the authorsdiscuss the use of optics in routers to scale capacity and re-duce energy consumption. In [12] methods are investigated todecrease energy consumption in interconnection fabrics. A casestudy of power demands of standard router platforms is providedin [13] where a generic model for router power consumption isdeveloped.

In [14] the authors focused on saving energy in local area Eth-ernet networks. An IEEE 802.3 Energy Efficient Ethernet StudyGroup was established in November 2006 [15] to standardizethe Adaptive Link Rate (ALR) concept to reduce direct energyuse of Ethernet links. In [16] the authors have considered En-ergy-Aware Traffic Engineering where energy consumption is

0733-8724/$26.00 © 2010 IEEE

4 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 29, NO. 1, JANUARY 1, 2011

Fig. 1. Structure of the hybrid-power IP over WDM network.

taken into account in addition to the standard traffic engineeringobjectives. In [17] a novel energy reduction approach at the net-work level is proposed where the load dependent energy con-sumption information of the communication equipment is takeninto account when taking traffic-engineering decisions. A groupknown as “Green Grid” was also formed to increase the energyefficiency in data centers [18].

In this paper, we focus on reducing the CO emissions ofbackbone IP over WDM networks. A LP optimization model for“hybrid-power” IP over WDM networks and a new heuristic areset up to minimize the non-renewable energy consumption. Is-sues including how to use renewable energy (solar in this paper)more effectively, how to reduce the non-renewable energy con-sumption of transponders (the second most energy consumingdevice in a node), how to select the location of nodes using re-newable energy, and load dependent energy consumption areconsidered. The remainder of the paper is organized as follows:In Section II, the hybrid-power IP over WDM network architec-ture and the LP optimization model are introduced. Section IIIintroduces the new heuristic for minimizing non-renewable en-ergy consumption and CO emission. In Section IV the simu-lation results are presented and analyzed. Finally, the paper isconcluded in Section V.

II. LP MODEL FOR RENEWABLE ENERGY IP OVER

WDM NETWORK

The IP over WDM network includes two layers, the IP layerand the optical layer. In the IP layer, an IP router is connectedto an optical switch. The router aggregates data traffic from ac-cess networks. The optical layer can provide large capacity andwide bandwidth for the data communication between IP routers.Optical switches are connected to optical fiber links. On eachfiber, a pair of wavelength multiplexers/demultiplexers is usedto multiplex/demultiplex wavelengths [19]. The transponderscan provide OEO processing for full wavelength conversion ateach switch node. In addition, for long distance transmission,the EDFAs are used to amplify the optical signal on each fiber.

Similar to IP over WDM networks, the hybrid-power IP overWDM network architecture is composed of an IP layer and an

optical layer. However, the difference is that the power supplyof this new network is mixed being composed of non-renew-able energy and renewable energy. In this case, the total COemission of an IP over WDM network will be reduced if a por-tion of the non-renewable energy consumption is replaced byrenewable energy. Therefore, the problem becomes that asso-ciated with minimizing the non-renewable energy consumptionof the hybrid-power IP over WDM network.

In [19], a Mixed Integer Linear Programming (MILP) modelwas developed for minimizing the total energy consumption ofIP over WDM networks. We have formulated a model that buildson these concepts but is focused on minimizing the non-renew-able energy consumption when renewable energy is employedin the hybrid-power network. In this LP model, we assume thenetwork has the topology with nodes andphysical links. The nodes that have access to renewable energycan also be powered by non-renewable energy to guarantee QoSwhen the renewable energy output becomes low. The renewableenergy can power the ports, transponders, optical switches, mul-tiplexers and demultiplexers in a node. Fig. 1 gives the detailsof the hybrid-power IP over WDM network. The total non-re-newable energy consumption of the network is composed of:

1) Non-renewable energy consumption of ports without ac-cess to renewable energy

2) The non-renewable energy consumption of EDFAs

3) The non-renewable energy consumption of router portsthat have access to renewable energy (the non-renewableenergy may be used for example to guarantee control at alltime)

DONG et al.: IP OVER WDM NETWORKS EMPLOYING RENEWABLE ENERGY SOURCES 5

4) The non-renewable energy consumption of transpondersthat have access to renewable energy (again the non-renew-able energy may be used for example to guarantee controlat all time) and that of the transponders without access torenewable energy

5) The non-renewable energy consumption of opticalswitches that have access to renewable energy (simi-larly the non-renewable energy may be used for exampleto guarantee control at all time) and that of the opticalswitches without access to renewable energy

6) The non-renewable energy consumption of multiplexersand demultiplexers that have access to renewable energy(here also the non-renewable energy may be used for ex-ample to guarantee control at all time) and that of the mul-tiplexers and demultiplexers without access to renewableenergy

The LP model that minimizes the non-renewable energyconsumed is defined as follows.

Objective: minimize

(1)

Subject to:

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

The variables and parameters in the equations above are de-clared in Table I.

The aim of the objective function (1) is to minimize thenon-renewable energy consumption of the hybrid-power IPover WDM network. Equation (2) and (5) represent the flowconservation constraint in the IP layer and the optical layer.Equation (3) ensures that the traffic flow on each virtual linkdoes not exceed its capacity. The term representsthe total number of wavelength channels on each virtual linkpowered by either non-renewable energy or renewable energy.Equation (4) ensures that the limit on the number of router portsin each node is not exceeded. Equation (6) ensures that therenewable energy consumption of router ports and transpondersis not larger than the maximum output power of the renewableenergy source in each node. Equations (7) and (9) give the limiton the number of wavelength channels in each physical link. Equation (8) ensures that for each node the total number of

ports assembling data is equal to the number of router portsusing non-renewable energy and the number of ports usingrenewable energy. Equation (10) gives the limit on the totalnumber of multiplexers and demultiplexers in node .

III. HEURISTIC APPROACH

In the multi-hop bypass heuristic proposed in [19] bandwidthutilization is improved by allowing traffic demands betweendifferent source-destination pairs to share capacity on commonvirtual links (lightpaths). Improving the wavelength bandwidthutilization results in fewer virtual links, and therefore, fewerIP router ports and lower energy consumption. However, inthe hybrid-power IP over WDM network architecture as weassume that renewable energy sources are available to a limitednumber of nodes, implementing the Multi-hop bypass heuristicwhich is based on shortest-path routing will only minimize thetotal energy consumption not taking into account whether thisenergy comes from renewable or non-renewable sources. Tominimize the utilization of non-renewable energy, we proposea new heuristic where the traffic flows are allowed to traverseas many nodes as possible that use renewable energy to ensurethat in addition to reducing the total number of IP router portsand transponders, the non-renewable energy consumption isminimized. This constraint may increase the propagation delay,however to maintain QoS, only the two shortest-path routes areconsidered. Due to the changing traffic pattern and the fact thatthe output power of renewable energy sources varies duringdifferent times of the day, the routing paths are dynamic. Thenew heuristic is known as Renewable Energy Optimization hop(REO-hop). The flowchart of the heuristic is shown in Fig. 2.

In this heuristic, all the node pairs are reordered based on theirtraffic demands from highest to lowest and an empty virtual linktopology is created. A node pair is then retrieved from the or-dered list, and its traffic demand is routed over virtual topology


TABLE IPARAMETERS �� AND VERIABLES �� USED FOR LP MODEL

so that it traverses the maximum number of nodes that userenewable energy. As mentioned, only the two routes which are

Fig. 2. Flowchart of the REO-hop heuristic.

shortest-path routes are considered, i.e., the two shortest-pathroutes are compared in terms of the number of nodes that userenewable energy and the one with the maximum number isselected. If sufficient free capacity is available on the virtualtopology, the selected route is accommodated and the remainingcapacity on all the virtual links is updated.

If the selected route with the maximum number of nodesusing renewable sources is not available, the other route is se-lected. In case the virtual topology cannot accommodate eitherroute, a new direct virtual link is established between the nodepair. Two virtual links are computed, one is a path with the max-imum number of nodes that use renewable energy, and the otheris the shortest-path route. The two virtual links are comparedand the one with lower non-renewable energy consumption isselected and is established. If the non-renewable energy con-sumption is the same for both paths, the shortest path is selectedin order to minimize the propagation delay. The new virtuallink is added to the virtual topology . The above process isrepeated for all the node pairs. Then, the remaining renewableenergy of each node is checked to determine whether it can sup-port the optical switch and the multiplexers and demultiplexers


Fig. 3. NSFNET test network with time zones.

Fig. 4. Average traffic demand in different time zones.

in a node. When all the traffic demands are routed on the virtualtopology , the objective function ((1)) is used to calculate thetotal non-renewable energy consumption of the network.

IV. SIMULATION AND RESULTS

To test the performance of the REO-hop architecture designheuristic and to evaluate the non-renewable energy consumptionof the architecture, the NSFNET network, depicted in Fig. 3, isconsidered as an example of a real world network.

The NSFNET network consists of 14 nodes and 21 bidirec-tional links. Solar energy is used as the renewable energy source.As the NFSNET network covers the US, different parts of thenetwork fall in different time zones, i.e., nodes will experiencedifferent levels of solar energy and traffic demands at any givenpoint in time. There are four time zones, Eastern Standard Time(EST), Central Standard Time (CST), Mountain Standard Time(MST) and Pacific Standard Time (PST). There is an hour timedifference between each time zone and the next, we use ESTas the reference time. Note that time zones dictate habits andtherefore network utilization and traffic demands in our case,however we use real sun rise and sun set data to determine thesolar energy available in a given city at a given point in time.

Fig. 4 shows the average traffic demand during different hoursof the day [20]. The average traffic demand between each node

Fig. 5. Solar energy in different nodes in different time zones.

pair ranges from 20 Gb/s to 120 Gb/s and the peak occurs at22:00 in these traffic profiles. We assume that the traffic demandbetween each node pair in the same time zone is random with auniform distribution and no lower than 10 Gb/s.

The solar energy power [21] available to a node is shown inFig. 5. As the output power of solar energy sources varies in dif-ferent hours of the day, we use the profile in [21] and the sun riseand sun set data associated with each node. The geographical lo-cation of nodes affects the sunset and sunrise time, and thereforehas impact on the solar energy generated in each node. Table IIgives the details of the solar energy output power of each node.The solar energy output is non-zero from 6:00 to 22:00 and themaximum output power occurs at 12:00.

A. Non-Renewable Energy Consumption of the Network

Nodes located at the core of the network with a high nodal de-gree are selected to use renewable energy as they are expectedto consume more power compared to nodes at the edge of thenetwork as more traffic flows are routed through them. Selectingthese nodes is expected to maximize the reduction in the non-re-newable energy consumption. Nodes 4, 5, 6, 7 and 9 are initiallyselected to use solar energy in our heuristic. Later we study theimpact of the location of nodes with solar energy and our LPmodel yields the optimum locations.


TABLE IISOLAR ENERGY OUTPUT POWER OF EACH NODE (20 KW MAXIMUM OUTPUT POWER)

Fig. 6. Number of wavelengths under the multi-hop-bypass heuristic in dif-ferent times of the day.

Typically a one square meter silicon solar cell can produceabout 0.28 kW of power [22]. We assume that the maximumsolar energy available to a node is 20 kW, therefore a total solarcell area of about 100 m is required. Later we examine theimpact of higher solar energy availability per node.

As mentioned under the Multi-bypass and REO-hop heuris-tics, some traffic demands are routed in the optical layer byoptical switches. We select a suitable size optical switch (theGlimmerglass’s 192 192 optical switch) based on the max-imum number of wavelengths used in each node. Although inthe Non-bypass heuristic the maximum number of wavelengthsused in each node is larger, we still use the same power con-sumption data for optical switches in the Multi-bypass heuristicand REO-hop heuristic due to the negligible difference in energyconsumption in different optical switch sizes compared to thepower consumption of a router port. Fig. 6 shows that under themulti-hop bypass (the heuristic requiring fewer wavelengths),the maximum number of wavelengths needed is 109.

Table III shows the simulation environment parameters interms of number of wavelengths, wavelength capacity, distancebetween two neighbouring EDFAs, and energy consumption ofdifferent components in the network. Some of the parameters

TABLE IIIINPUT DATA FOR THE SIMULATION

are similar to those in [19] which are also derived from Cisco’s8-slot CRS-1 data sheets [23], and others are derived from Glim-merglass’s 192 192 channels Sytem-600 data sheets [24] andCisco’s ONS 15454 data sheets [25].

The AMPL/LPSOLVE software is used to solve the LPproblem. Five different cases are considered: 1) Non-bypassheuristic without renewable energy nodes, 2) Non-bypassheuristic with renewable energy nodes, 3) Multi-hop-bypassheuristic without renewable energy nodes 4) Multi-hop-by-pass heuristic with renewable energy nodes, and 5) REO-hopheuristic with renewable energy nodes.

Fig. 7 shows the total non-renewable energy consumption.The curves “Non-bypass without solar energy” and “LP op-timal with solar energy” provide the upper and lower boundson the non-renewable energy consumption. We assume the


Fig. 7. Total nonrenewable energy consumption of different heuristics with andwithout solar energy.

traffic demands and the output power levels of the solar energysources given in Fig. 4 and Table II, respectively. Comparedto the “Non-bypass with solar energy” curve, both of theMulti-hop-bypass and REO-hop heuristics have reduced thenon-renewable energy consumption. The savings in the non-re-newable energy consumption, introduced by the REO-hopheuristic, increase at the time of the day when the solar en-ergy is significant (From Fig. 6, between 6:00 and 22:00).Compared with the upper bound, at 12:00, 14:00, 16:00 and18:00, the REO-hop heuristic saves non-renewable energy ofabout 1000 kW. Furthermore, REO-hop still out-performs theMulti-hop-bypass heuristic, from 0:00 to 4:00 when there isno solar energy in the network as REO-hop tries to route de-mands on virtual links with sufficient capacity rather than usingshortest-path routing as with the Multi-hop-bypass heuristic.

We have also examined the impact of different values of themaximum solar energy (40 kW, 60 kW and 80 kW) availableper node. A solar cell area of up to 300 m is needed to generatesuch values. Solar cell cladding with such surface area can bepractically built in a typical core routing node location. In Fig. 8,it is shown that increasing the maximum solar energy outputper node has linearly reduced the total non-renewable energyconsumption using our algorithms.

A similar trend to that observed in Fig. 8 is noticed for thereduction in CO emissions in Fig. 9. The total CO emissionsduring a 24 hour period have been reduced by about 47%–52%and 43%–49% under the REO-hop and Multi-hop-bypassheuristics, respectively compared to the non-bypass withoutrenewable case. As mentioned in Section I, about 228 gramsCO are produced through the consumption of 1kWh of non-re-newable energy. We are able to reduce the CO emissions byabout 1300 tones every year by using the REO-hop heuristic inNSFNET.

As the REO-hop heuristic routes demands dynamically basedon the output power of solar energy sources in nodes and theavailable capacity, the propagation delay under the REO-hopheuristic is expected to increase compared to the Multi-hop-by-pass heuristic which routes traffic based on the shortest path.However, as the REO-hop heuristic routes demands using only

Fig. 8. Total nonrenewable energy consumption in 24 hours for different valuesof the maximum solar energy output under different heuristics.

Fig. 9. Reduction in CO emissions in 24 hour period under differentheuristics.

Fig. 10. Average propagation delay of REO-hop and Multi-hop-bypassheuristics.

the two shortest-path routes, we can see in Fig. 10 that the prop-agation delay has not increased significantly (the increase is lessthan 0.3 ms, i.e., less than 10%) maintaining the QoS.


Fig. 11. Total nonrenewable energy consumption in 24 hour period under dif-ferent maximum solar energy power per node and different number of nodesemploying solar energy.

B. Number and Location of Nodes That Use Renewable Energy

In Section IV.A, we have only selected the five nodes with thehighest nodal degree in the NSFNET network to use renewableenergy. In this section we study the impact of the number andlocation of nodes that use renewable energy. First, we investi-gate how the non-renewable energy consumption is affected bythe maximum output power of renewable energy sources andthe number of nodes that use renewable energy. In Fig. 11 thenumber of nodes with access to renewable energy increases byadding to the list the node with the next larger node ID, i.e.,the set of nodes using renewable energy takes the followingvalues: . Weexamine a range of values for the maximum output power of re-newable energy sources from 0 to 80 kW.

In Fig. 11, it is clear that increasing the number of nodesthat use renewable energy and increasing the maximum outputpower per renewable energy source reduces the total non-renew-able energy consumption of network. However, we can see fromthe figure that the relation is not linear between the number ofnodes and the non-renewable energy consumption which indi-cates that some nodes have more impact on the non-renewableenergy consumption than others if they are selected to use re-newable energy.

To investigate the impact of the location of nodes that userenewable energy on the total non-renewable energy consump-tion, a new LP model is developed with the objective of opti-mizing the selection of nodes that use renewable energy suchthat the non-renewable energy consumption savings are maxi-mized. The new LP model is subject to the sameconstraints inSection II, except that (6) is replaced with (12) and a new con-straint is added (13).

In this model we consider time as a variable. Therefore, isadded to all the variables in Table I where is the time point oftime set .

The new LP model is defined as follows.Objective: maximize

(11)

Subject to:Equations: (2), (3), (4), (5), (7), (8), (9), (10)(Every variable in the equations above has had the time vari-

able augmented)

(12)

(13)

where if node uses renewable energy, otherwise ,and is the total number of nodes with access to renewableenergy.

Equation (12) ensures that the renewable energy consumptionin each node is within the maximum output power of its associ-ated renewable energy source at any time of the day. In practice,the energy produced from solar cells can be stored in batteries;which relaxes the constraints on the availability of solar energyas a function of the day time. The use of energy storage elementsis however, not included in the current formulations. Equation(13) implies that the total number of nodes that use renewableenergy is limited to which is set in advance.

The optimization results are given in Fig. 12(a) under dif-ferent values of the maximum renewable energy output power(20 kW to 80 kW), assuming and the traffic demandshown in Fig. 4. We can see that the optimal node selection doesnot change.

To verify the optimization results, in Fig. 12(b) we evaluatethe total non-renewable energy consumption in a 24 hour pe-riod under the REO-hop heuristic where we assume that onlya single node uses renewable energy (note that Fig. 12(a) usedLP). We evaluate the performance under different values of themaximum solar energy per node. It is clear that the total non-re-newable consumption is lower when the nodes in the center ofthe network (4, 5, 6, 7, and 9) use renewable energy.

In Fig. 13, the delay and power consumption performanceare evaluated under the REO-hop heuristic with a maximumrenewable power of 60 kW per node when central networknodes or the periphery nodesare selected to use renewable energy. It is clear that the formernode set results in a higher reduction in the non-renewableenergy consumption compared to the latter node set. Alsoin Fig. 13 it is clear that selecting the node setresults in a lower average propagation delay compared to thenode set . Therefore, the optimal selection ofnodes using renewable energy results in better utilization of therenewable energy resources.

C. Node Non-Renewable Energy Consumption

In this section we investigate the reduction in the non-renew-able energy consumption experienced by each node individu-ally under different heuristics. We compare the scenarios whenno renewable energy sources are used and when some nodesuse renewable energy (nodes 4, 5, 6, 7 and 9). Assuming thetraffic demand in Fig. 4 and that the maximum solar energy


Fig. 12. (a). Optimum node location of nodes with access to renewable energyfor different values of maximum available solar energy per node (LP-optimal).(b). Total non-renewable energy consumption in 24 hour period with differentnodes using renewable energy.

Fig. 13. Non-renewable energy consumption and the average propagationdelay under two different node selection scenarios using renewable energy.

output power is 60 kW, Fig. 14 shows the non-renewable en-ergy consumption of all nodes in the NSFNET network. It is

Fig. 14. Non-renewable energy consumption of the nodes in a 24 hour periodunder different heuristics.

clear that under both cases the non-renewable energy consump-tion of nodes with the Non-bypass heuristic has a large vari-ance because nodes at the center of the network consume moreenergy as more traffic flows are routed through them. Com-pared with the Non-bypass heuristic, the Multi-hop-bypass andREO-hop heuristics have significantly reduced the non-renew-able energy consumption and its variance. However, nodes at thecenter of network still have slightly more non-renewable energyconsumption than the nodes at the edge of the network. As ex-pected, the REO-hop heuristic results in further reductions com-pared to the Multi-hop-bypass heuristic. It is also clear in thefigure that as nodes at the center of the network use renewableenergy, their non-renewable energy consumption is significantlyreduced when renewable energy sources are introduced to thenetwork.

D. Non-Renewable Energy Consumption Under Adaptive LinkRate With the REO-Hop Heuristic

Results were obtained in the previous sections under the as-sumption that equipment has two energy states (on and off), i.e.,the equipment consumes full power when switched-on. How-ever, energy consumption can be decreased by deploying en-ergy-efficient components. Several factors affect the power con-sumption of telecommunication equipment, such as traffic load,temperature and QoS policies [13]. Load is considered as one ofthe factors that have the highest influence on equipment energyconsumption. In this section, we investigate the effect of ALRon the non-renewable energy consumption.

Different energy profiles are proposed to provide a more ac-curate definition of the dependency between equipment energyconsumption and traffic load. Fig. 15 shows different energyprofiles for telecommunication equipment [13] where energyconsumption is a function of the load on the network compo-nent. The latter is expressed as a percentage of the total capacityof the network component. We consider (i) ‘On-off’ energy pro-file [13] (ii) ‘Linear’ energy profile: Here the energy consump-tion depends linearly on the traffic load, e.g., in switch archi-tectures like Batcher, Crossbar and Fully-Connected [13], [26](iii) ‘Log10’ energy profile employed in equipment that uses hi-bernation techniques such as the low-power idle technique for


Fig. 15. Different energy profiles.

Ethernet [27], [28]. In this approach data is sent as fast as pos-sible to allow the equipment to quickly return to the low-poweridle state. (iv) ‘Log100’ energy profile: This profile is consid-ered as a middle function between the ‘On-off’ and the ‘Log10’profiles. (v) ‘Cubic’ energy profile: Typical in equipment thatuses Dynamic voltage Scaling (DVS) and Dynamic FrequencyScaling (DFS) [13].

Under the same assumptions of Section IV.A and with 20 kWsolar power at the five optimum nodes or 80 kW at all nodes,Fig. 16 shows the non-renewable energy consumption of theREO-hop heuristic where we consider the different energy pro-files in Fig. 15 for router ports and transponders (the two mostenergy consuming sub-systems in the node). It is clear fromFig. 16 that the non-renewable energy consumption curves’behaviour is subject to the energy profile curves in Fig. 15.The largest reduction in non-renewable energy consumptionoccurred under the ‘cubic’ profile. Compared to the ‘on-off’profile, the ‘cubic’ profile results in reducing the non-renewableenergy consumption by up to 9% between 12:00 and 20:00.

It should be noted that in Fig. 16 the energy profiles of Fig. 15are only applied to partially loaded wavelengths while an ‘on-off’ profile is applied to fully loaded wavelengths. For examplea traffic demand of 70 Gb/s between a node pair calls for twowavelengths, one fully loaded, the other partially loaded. Therelatively small (9%) reduction in energy consumption associ-ated with ALR is commensurate with this. Note that when allthe nodes have access to 80 kW solar power then the energyconsumption in Fig. 16 continues to decrease beyond 6 am dueto the availability of solar power at the nodes. At 22:00 the twosets of curves converge, however the total energy consumptionis still lower when all nodes have access to solar power due tosolar power availability in more nodes.

The sum of the total power consumed by all nodes, whereeach node is dimensioned based on the largest number of routerports needed over the 24 hour period is 2010 kW which is thepeak shown in Fig. 7 in the non-bypass case without solar en-ergy. The energy saving between such a network and the re-sults in Fig. 16 (clustered curves where only 5 nodes use 20 kWrenewable energy) is significant and the maximum is approxi-mately 85%. The average savings in this case are approximately

Fig. 16. Total non-renewable energy consumption of the REO-hop heuristicunder different energy profiles.

Fig. 17. Reduction in non-renewable energy consumption of the REO-hopheuristic under different energy profiles.

65% and vary slightly between the five profiles. These savingsare shown in Fig. 17. Note that the 85% and 65% savings are al-most real energy savings since the renewable energy is low hereand has limited effect. The savings come from our architecturedesign (photonic switching instead of electronic routing) andpowering down unused router ports and transponders. Fig. 18shows the non-renewable energy consumption of the individualnodes in the network under the REO-hop heuristic when onlythe 5 optimum nodes have access to 20 kW renewable energyeach, and when all the nodes have access to 80 kW renewableenergy each. It helps appreciate the typical power consumptionlevels per node and the impact of 20 kW renewable power atfive nodes (4, 5, 6, 7, 9) and 80 kW renewable at all nodes.

Fig. 16 also shows the total non renewable energy consump-tion when all the NSFNET nodes have access to 80 kW renew-able energy each. The maximum reduction in non-renewableenergy consumption (reduction in CO emissions) in this casecompared to the peak in Fig. 7 is 97%. This is also shown inFig. 17. The average savings here are approximately 78% withsmall variation between the five profiles.


Fig. 18. Non-renewable energy consumption at each node under REO-hop heuristic and ‘cubic’ profile with: (a) 20 kW solar power at nodes (4, 5, 6, 7, 9),(b) 80 kW at all nodes.

V. CONCLUSION

In this paper, we have proposed the use of renewable energyin IP over WDM networks to reduce the non-renewable en-ergy consumption and consequently CO emissions. We havedeveloped an LP optimization model and a highly efficientheuristic based on Multi-hop-bypass, known as REO-hop, tooptimize the use of renewable energy in the hybrid-power IPover WDM architecture. Simulation results show that com-pared to the Multi-hop-bypass, the REO-hop heuristic has re-duced the non-renewable energy consumption by 47%–52%while maintaining QoS. We have also investigated the impactof the maximum solar energy per node and the optimal locationof nodes that use renewable energy on the total non-renewableenergy consumption. An LP optimization model has been de-veloped to optimize the selection of nodes using renewableenergy. The results show that selecting nodes at the center ofthe network to use renewable energy results in higher reduc-tions in the total non-renewable energy consumption comparedto nodes at the edge. Also ALR where load dependent energyconsumption is assumed has been investigated. The results haveshown that the ‘cubic’ energy profile results in the highest re-duction in the non-renewable energy consumption. Our opti-mized REO-hop routing heuristic with renewable energy andALR results in a maximum energy saving of 85%, average of65%, compared to a current network design where all nodes arestatically dimensioned for the maximum traffic in terms of IPports and optical layer. These results are based on the networkparameters used, NSFNET, the traffic profiles given and solarpower variation shown in this work. Furthermore, when all thenodes have access to 80 kW renewable power each, the totalenergy savings and the associated reduction in non-renewableenergy consumption reduce the network’s CO emissions by97% peak, and 78% average.

ACKNOWLEDGMENT

The authors would like to thank their collaborators atthe University of Cambridge, Cambridge, U.K., for usefuldiscussions.

REFERENCES

[1] B. G. Bathula, D. Satyanarayana, and J. M. H. Elmirghani, “Energyefficient optical burst switched (OBS) networks,” in Proc. IEEEGLOBECOM’09 Workshops, 2009, pp. 1–6.

[2] “The Environment,” [Online]. Available: http://www.telecomitalia.it/sostenibilita2006/English/B05.htm1 Telecom Italia Website, URL:

[3] “BT Announces Major Wind Power Plans,” BT Press, Oct. 2007 [On-line]. Available: http://www.btplc.com/News/Articles/Showarticle.cfm?ArticleID=dd615e9c-71ad-4daa-951a-55651baae5bb

[4] “Wind Farms to Power BT,” 2009 [Online]. Available: http://www.btplc.com/sharesandperformance/Annualreportandreview/Sharehold-ermagazine/May2009/windfarms.htm?Terms=380,4305

[5] C. Jones, M. Sivalingam, P. Agrawal, and J. Chen, “A survey of energyefficient network protocols for wireless networks,” Wireless Networks,vol. 7, no. 4, pp. 343–358, Jul. 2001.

[6] A. Vahdat, A. Lebeck, and C. Schlatter-Ellis, “Every joule is precious:The case for revisiting operating system design for energy efficiency,”in Proc. 9th ACM SIGOPS European Workshop, Kolding, Denmark,Jun. 2000.

[7] V. Raghunathan, M. Srivastava, and R. Gupta, “A survey of techniquesfor energy efficient on-chip communication,” in Proceedings of DesignAutomation Conference’03, Anaheim, CA, Jun. 2003, pp. 900–905.

[8] T. Pering, T. Burd, and R. Bordersen, “The simulation and evaluationof dynamic voltage scaling algorithms,” in Proceedings of the Interna-tional Symposium on Low Power Electronics and Design, Monterey,CA, Aug. 1998, pp. 76–81.

[9] M. Gupta and S. Singh, “Greening of the internet,” Proc. ACM SIG-COMM, Aug. 2003.

[10] S. Nedevschi, L. Popa, G. Iannaccone, Y. Ratnasamy, and D.Wetherall, “Reducing network energy consumption via sleepingand rate-adaptation,” in Proc. USENIX Symp. Networked Syst. Des.Implementation, San Francisco, CA, 2008, pp. 323–336.

[11] I. Keslassy, S. Chuang, K. Yu, D. Miller, M. Horowitz, O. Solgaard, andN. McKeown, “Scaling internet routers using optics,” in Proc. ACMSIGCOMM, Karlsruhe, Germany, Aug. 2003, pp. 189–200.

[12] A. Wassal and M. Hasan, “Low-power system-level design of VLSIpacket switching fabrics,” IEEE Trans. Comput.-Aided Design (CAD)Integr. Syst., vol. 20, no. 6, Jun. 2001.

[13] J. Camilo, C. Restrepo, C. Gruber, and C. M. Machuca, “Energy profileaware routing,” in Proc. 1st Int. Workshop Green Commun. IEEE Int.Conf. Commun., Dresden, Germany, Jun. 2009, pp. 1–5.

[14] K. J. Christensen, C. Gunaratne, B. Nordman, and A. D. George, “Thenext frontier for communications networks: Power management,”Comput. Commun., vol. 27, pp. 1758–1770.

[15] The Green Grid, , Aug. 2006 [Online]. Available: http://www.thegreen-grid.org/pages/overview.html

[16] J. Chabarek, J. Sommers, P. Barford, C. Estan, D. Tsiang, and S.Wright, “Power awareness in network design and routing,” in Proc.INFOCOM, 2008, pp. 457–465.

[17] N. Vasic and D. Kostic, “Energy aware traffic engineering,” in Proc.1st Int. Conf. Energy-Efficient Computing Networking (E-ENERGY),2010, pp. 169–178.

[18] IEEE 802.3 Energy Efficient Ethernet Study Group, [Online]. Avail-able: http://grouper.ieee.org/groups/802/3/eee_study/index.html


[19] G. Shen and R. S. Tucker, “Energy-minimized design for IP over WDMnetworks,” Opt. Commun. Networking, vol. 1, pp. 176–186, 2009.

[20] Y. Chen and C. Chou, “Traffic modeling of a sub-network by usingARIMA,” in Proc. Info-Tech and Info-Net Conf., 2001, vol. 2, pp.730–735.

[21] H. Wang, H. Yang, and H. Wu, “A fine model for evaluating output per-formance of crystalline silicon solar modules,” in Proc. 4th IEEE WorldConf. Photovoltaic Energy Conversion, 2006, vol. 2, pp. 2189–2192.

[22] J. Zhao, A. Wang, P. P. Altermatt, S. R. Wenham, and M. A. Green,“24% efficient silicon solar cells,” in Proc. 24th Photovoltaic Spe-cialists Conf. Photovoltaic Energy Conversion , 1994, vol. 2, pp.1477–1480.

[23] Cisco CRS-1 Specification Data Sheet, [Online]. Available:http://www.cisco.com

[24] Glimmerglass System-600 Data Sheet, [Online]. Available:http://www. glimmerglass.com

[25] Cisco ONS 15454 Data Sheet, [Online]. Available: http://www.cisco.com

[26] T. T. Ye, “Analysis of power consumption on switch fabrics in networkrouters,” in Proc. 39th Design Autom. Conf., New Orleans, LA, Jun.10–14, 2002, pp. 524–529.

[27] R. Hays, “Active/idle toggling with low-power idle,” presented at theIEEE802.3az Task Force Group Meeting, Jan. 2008.

[28] IEEE802.3az, Energy Efficient Ethernet Study Group, [Online]. Avail-able: http://www.ieee802.org/3/az/index.html

Xiaowen Dong received the B.E. degree in electronic engineering from South-west Jiaotong University, Chengdu, China, in 2005, and the M.E. degree (withFirst Class Honours) in electronic engineering from the National University ofIreland, Maynooth, Ireland, in 2008. He is currently working toward the Ph.D.degree in electronic and electrical engineering at the University of Leeds, Leeds,U.K.

From 2005 to 2007, he was a Wireless Communication System Engineer withWuhan Research Institute (WRI), Wuhan, China. His research interests includeenergy-aware networks, energy-aware routing methods, and protocols in com-munication networks.

Taisir El-Gorashi received the B.S. degree in electrical and electronic engi-neering from the University of Khartoum, Khartoum, Sudan, in 2004, the M.Sc.degree in photonic and communication systems from the University of Wales,Swansea, U.K., in 2005, and the Ph.D. degree from University of Leeds, Leeds,U.K., in 2010.

Currently, she is a Postdoctoral Research Fellow with the University of Leeds,Leeds, U.K.. Her research interests include next-generation optical network ar-chitectures, storage area networks and Grid networks, and green ICT.

Jaafar M. H. Elmirghani is Director of the Institute of Integrated Informa-tion Systems, School of Electronic and Electrical Engineering, University ofLeeds, Leeds, U.K. He joined the University of Leeds in 2007 and, prior tothat (2000–2007), as a chair in optical communications with the Universityof Wales Swansea, he founded, developed, and directed the Institute of Ad-vanced Telecommunications and the Technium Digital (TD), a technology incu-bator/spin-off hub. He has provided outstanding leadership in a number of largeresearch projects at the IAT and TD. He is the coauthor of Photonic switchingTechnology: Systems and Networks (Wiley, 1998) and has authored or coau-thored over 300 papers. His research interests are in optical systems and net-works and signal processing.

Dr. Elmirghani is Fellow of the IET and the Institute of Physics. He wasChairman of IEEE COMSOC Transmission Access and Optical Systemstechnical committee and was Chairman of IEEE COMSOC Signal Processingand Communications Electronics technical committee, and an editor of theIEEE Communications Magazine. He was founding Chair of the AdvancedSignal Processing for Communication Symposium which started at IEEEGLOBECOM’99 and has continued since at every ICC and GLOBECOM.He was also the founding Chair of the first IEEE ICC/GLOBECOM opticalsymposium at GLOBECOM’00, the Future Photonic Network Technologies,Architectures and Protocols Symposium. He chaired this Symposium, whichcontinues to date under different names. He was the recipient of the IEEECommunications Society Hal Sobol Award, the IEEE COMSOC ChapterAchievement Award for Excellence in chapter activities (both in 2005), theUniversity of Wales Swansea Outstanding Research Achievement Award in2006, and the IEEE Communications Society Signal Processing and Commu-nication Electronics Outstanding Service Award in 2009.

Date post:	30-Sep-2016
Category:	Documents
Upload:	jmh
View:	216 times
Download:	2 times

IP Over WDM Networks Employing Renewable Energy Sources

Documents