dba

Evaluation of Dynamic Bandwidth AllocationAlgorithms in GPON Networks

Joanna Ozimkiewicz, Sarah Ruepp,Lars Dittmann, Henrik Wessing

Technical University of DenmarkDTU FotonikKgs. Lyngby

[email protected],{srru,ladit,hewe}@fotonik.dtu.dk

Sylvia SmolorzNokia Siemens Networks GmbH & Co. KG

MunichGermany

[email protected]

Abstract: In this paper, two approaches for Dynamic Bandwidth Allocation in GPON networks are proposed, andvalidated through simulations in the OPNET modeler. One approach address a Status Reporting scheme, where thebandwidth allocation originates from the client request. The second use a centralized Non Status Reporting scheme.Furthermore, parameters to cope with variances in the traffic pattern is quantified. The results on performance,scalability and efficiency show that Status Reporting is utilizing the bandwidth more efficient while the Non StatusReporting provides better QoS for real time services.

Key–Words:GPON, dynamic bandwidth allocation (DBA), simulation, status reporting, OPNET

1 IntroductionBandwidth requirement for providing new services isincreasing. Moreover, different types of users havevarying needs regarding the amount of bandwidth andtransmission delays. Network providers are forcedto think about new mechanisms that will distributethe bandwidth among the users and provide high net-work reliability. That leads to the increased interestin the optical networks suitable for Fiber to the Home(FTTH) and Fiber to the Building (FTTB) solutions.Creation of the networks in which each connected userobtains high QoS, despite the variation of payment istoo costly solution. This leads to high requirement forimplementing mechanisms that will be responsible forthe appropriate sharing of available resources.

One of the access network technologies intro-ducing high throughput, low delays and advancedbandwidth control is Gigabit Passive Optical net-work (GPON) [1]. GPON is designed to transportEthernet packets over the optical medium using theGPON Encapsulation Method (GEM) [3]. The phys-ical link is fragmented into GEM frames as specifiedin [1] [3]. Each downlink frame contains a Bandwidthmap (BWMap), with information about allowed trans-mission times for each Optical Network Unit (ONU)for the future uplink frame. Dynamic Bandwidth As-signment (DBA) algorithm is used to calculate the to-tal bandwidth assigned for end nodes [3].

This paper introduces two different DBA algo-rithms designed for GPON, namely, status and non

status reporting DBA [3] [4], respectively. DBA forGPON networks is not given in ITU-T GPON speci-fications, and hence often proprietary algorithms areused in OLT equipment. This paper provides a pro-posal for the DBA algorithms, together with the anal-ysis of the impact of available configurations on thenetwork performance, which has not been heavily re-searched until now. The research focuses on the net-work performance, its optimum configuration param-eters and QoS that can be delivered to the subscriberswith different traffic priorities. GPON network mod-elling and simulation has been performed using OP-NET environment [15]. Related work to PON net-works has been carried out in [10–14]

The remainder of this paper is organized as fol-lows: Section 2 provides background knowledge ofDBA and traffic types. In section 3, a DBA algorithmis proposed. In section 4, the simulation scenario isdescribed and simulation results are presented in sec-tion 5. Section 6 concludes the paper.

2 Dynamic Bandwidth Assignment(DBA)

GPON uses point-to-multi-point connections betweencentral Optical Line Termination (OLT), coordinat-ing network resources and Optical Network Units(ONU) located near the end users. Maximum allow-able distance is 20km. Varying raw transmission rates

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMSJoanna Ozimkiewicz, Sarah Ruepp, Lars Dittmann, Henrik Wessing, Sylvia Smolorz

ISSN: 1109-2734 111 Issue 2, Volume 9, February 2010

are available:1.24416 Gbit/s uplink, 2.48832 Gbit/sdownlink for asymmetric services and 2.48832 Gbit/suplink and downlink for symmetric services. Due tothe high available bandwidth, its allocation is basedon the Service Level Agreements (SLA) where Qual-ity of Service (QoS) can be granted according tothe demand. GPON is using transmission container(T-CONT) mechanism for provisioning differentiatedQoS. T-CONT is the logical connection between OLTand ONU, where multiple T-CONT types can be al-located in one ONU. GPON contains five different T-CONTS ouf of which types 2, 3 and 4 that were feasi-ble for dynamic bandwidth allocation were evaluated:

1. T-CONT type 1 for fixed rate traffic sensitive tojitter and delay.

2. T-CONT type 2 for on-off type traffic with welldefined bitrate and strict delay requirements isprovisioned with assured bandwidth. This band-width has to be granted to the T-CONTs’ traffic,if requested. If not used, bandwidth can be re-allocated to other T-CONTs, providing that it isavailable as soon as T-CONT type 2 requires it.

3. T-CONT type 3 is provisioned with assuredbandwidth and additionally, it can be grantednon-assured bandwidth if all available assuredbandwidth is utilized. It is suitable for variablerate, bursty traffic with requirements for averagerate guarantee.

4. T-CONT type 4 has no bandwidth guarantee butit has eligibility in best effort bandwidth sharing.It is suitable for variable rate, bursty traffic withno delay sensitivity.

5. T-CONT type 5 is consolidation of other T-CONTs, provisioned with both fixed and assuredbandwidth.

Collisions in such high throughput networks arevery costly. In order to avoid collisions the OLT al-locates upstream transmission intervals per T-CONTin a TDMA fashion. Transmission is coordinated us-ing GPON Transmission Convergence (GTC) framesin both uplink and downlink direction. In downstreamframe, OLT transmits BWMap containing timing in-formation indicating when each T-CONT is allowedto transmit data during future upstream frame. TheDBA automatically adjusts bandwidth grants to theneeds of a particular T-CONT. The DBA uses theT-CONT’s activity status as an input to the sched-uler. This activity status can be obtain either explicitlythrough T-CONT buffer status reporting (SR), or im-plicitly through transmission of the idle GEM frames

when T-CONT does not have enough data to trans-mit during all granted upstream allocation intervals.The implicit method is referred to as non status report-ing (NSR). Assured bandwidth is granted regardlessof the overall traffic load. Additional non-assured andbest effort bandwidth allocation depends on the avail-able upstream capacity remaining after allocation ofthe granted bandwidth. Non-assured and Best effortbandwidth can be overbooked, while this can neverbe the case for assured bandwidth. During bandwidthallocation the following prioritization is applied:

• Fixed traffic (highest priority)

• Assured traffic

• Non-assured traffic

• Best effort traffic (lowest priority)

For both SR and NSR DBA, the OLT traces theactivity status of each T-CONT during one DBA cy-cle. Obtained information becomes input to the sched-uler, which thereby allocates transmission opportuni-ties for the next DBA cycle.

3 Proposed Dynamic Bandwidth As-signment Algorithms

3.1 Non Status Reporting DBAIn the NSR algorithm, the OLT estimates bandwidthallocation for the next DBA cycleBt

a(c), required byeach T-CONT t, on the basis of the bandwidth usageduring the previous DBA cycleBt

u(c− 1).At the beginning of each cycle c, at first the

OLT assigns assured bandwidth for each T-CONTt (Bt

aAsr), based on the amount of data transmit-ted in the previous cycle. After this assignment, re-maining GPON bandwidth is divided between non as-sured traffic(Bt

aNasr). In the end the rest of avail-able bandwidth is distributed for the best effort traf-fic (Bt

aBE). The assured bandwidth granted for T-CONTs of type 2 and 3 for the new cycleBt

aAsr(c) isbased on the activity in the previous cycleBt

u(c− 1),as shown in eq. 1. The expansion factor(EF ) is usedto provide fast response for variation in traffic, andto ensure that assured bandwidth is always allocatedwhen required by a T-CONT.

BaAsrt(c) =

Maxtasr if Bt

u(c− 1) = MaxtAsr

Mintu if Bt

u(c− 1) ≤ MintAsr

Btu(c− 1) ∗ EF otherwise

(1)Equation 1 verifies thatBt

aAsr(c) does not exceedthe maximum allowed bandwidthMaxt

Asrfor T-

CONT t and is never lower than the minimum band-width Mint

Asr.



Assignment of additional non-assured grantsBt

aNasr(c) eligible for T-CONTs of type 3 dependsboth on bandwidth used by the T-CONT in the previ-ous cycleBt

u(c − 1) and assured bandwidth alreadyassigned for current cycleBt

aAsr(c) (eq. 2). It is lim-ited by the total allowed bandwidthMaxt

Total, and no

minimum bandwidth assignment is guaranteed.

BtaNasr(c) =

Maxttotal −Maxt

Asr if Btu(c− 1) ∗ EF > Maxt

total

Btu(c− 1) ∗EF −Maxt

Asr if Btu(c− 1) > Maxt

Asr

0 otherwise(2)

When the non-assured bandwidth is overbooked, non-assured bandwidth assignments are scaled proportion-ally to fit into the total link capacityC for each T-CONT (eq. 3)

if : (∑

tBt

aAsr(c) +∑

tBt

aNasr(c)) > C

than :

BaNasrt(c) =Bt

aNasrt(c)∗(C−

∑tBt

aAsr(c))

∑tBt

aNasr(c)

(3)

Best effort bandwidth is assigned proportionally toBt

uBe(c − 1) scaled byEF (eq. 4), (eq. 6) and lim-ited by Maxtotal. OLT should try to grant to eachT-CONT minimum amount of bandwidth in each cy-cle to avoid T-CONT deadlock situation, where ONUhas no means of informing OLT about its bandwidthrequirements. In highly overbooked networks zerobandwidth allocation in the previous cycle indicatesthat the link was too congested to serve the best ef-fort traffic. ThereforeBt

aBe(c) is set to infinity, andscaled equally for all T-CONTs of type 4 (eq. 5).

BaBet(c) =

Maxttotal if Bt

u(c− 1) = Maxttotal

∞ if Btu(c− 1) = 0

Btu(c− 1) ∗EF otherwise

(4)

if : (∑

t BaAsrt(c) +∑

t BaNasrt(c) +∑

t BaBet(c)) > C

then :BaNasrt(c) =∑

tBaBet(c)∗(C−

∑tBaAsrt(c)+

∑tBaNasrt(c))

∑tBaBe(c)

(5)

Bta(c) < − >

Bta(c)

Btu(c− 1)

(6)

When the DBA cycle ends, the scheduler is providedwith assignments for a new cycle (eq.1 to eq.5) andcreates a bandwidth map for each frame in the newDBA cycle. Assured traffic is allocated for each T-CONT once per millisecond, thereby once in every 8upstream frames(eq. 7).

EF = ceil(Bt

aAsr(c) ∗ 8

frames per DBA cycle) (7)

Non-assured and best effort traffic grants are notserved in strictly regular intervals. Bandwidth grant

for each T-CONT may occur at most every third framein order to maximize throughput, since small band-width chunks lead to high header overhead and packetfragmentation. Allocations are served in a cyclic man-ner according to eq. 8 and starting from the T-CONTthat was last granted bandwidth.

BframeNasrt(c) =

BaNasrt(c) ∗ frames since last allocframes per DBA cycle

(8)

3.2 Status Reporting DBAIn the SR scheduling, the OLT requests the bufferoccupancy status from each T-CONT indicating thenumber of bytes waiting for transmission. At the be-ginning of a new frame, one T-CONT of each typegets a token. The T-CONT possessing a token isgranted allocation slots according to the request, pro-viding that it does not exceed its maximum bandwidthallowed for current cycle. At first, requests for assuredallocations are served. Later, bandwidth allocationsare given to consecutive non-assured bandwidth re-quests. In the end, best effort requests are served. Af-ter each T-CONT is served, the next in order T-CONTreceives the token. At the beginning of a new frame,T-CONTs with tokens received in the previous uplinkframes are served first.

4 SimulationsSeries of simulations were made using OPNET [15],to evaluate the efficiency of SR and NSR DBA al-gorithms for varying GPON network scenarios. AllGPON physical layer properties [5] together withframing and protocol overheads were accurately mod-eled. The transmission timing highly affects the re-sults, especially on the SR algorithm, where negotia-

Figure 1: Simulated network.



tion between OLT and ONU has to be performed be-fore calculation of bandwidth grants.

All simulations were performed using GPON2.48832 Gbps links. No bandwidth is reserved forGPON network control and maintenance, since band-width requirement for these control channels is toosmall to have an effect on the simulation results.

Each ONU has three T-CONTs: type 2, type 3and type 4. Each T-CONT of same type has thesame bandwidth parameters (maximum and minimumbandwidth, source bandwidth). The packet sizes varyfor different scenarios, depending on the traffic loadgenerated per workstation. The source load is gener-ated with the normal distribution of variance equal tothe average packet size.

In the first part of simulations (A) optimumGPON DBA parameters were determined for SR fre-quency request, EF and DBA cycle (scenarios A.1,A.2 and A.3 correspondingly). In every case a tradeoff was to be made, since improvement on one net-work parameter often results in degradation of theother (increase of bandwidth utilization, results in anincreased transmission delay).

Second part (B) provides Fiber to the Build-ing (FTTB) simulation scenarios for observing net-work performance and traffic QoS for different net-work configurations. The modelled FTTB topologyis shown in in figure 1. The network consists of oneOLT and 48 office networks, by which two line cardsof OLT are fully utilized. Furthermore, each officenetwork consists of one ONU and six workstations,spaced within 200m from centered ONU node. Twoworkstations are used to model traffic load of one T-CONT. Each workstation randomly takes a worksta-tion of the same type as a destination. The worksta-tion can be from any network, and one workstationcannot be chosen more than once. The distance be-tween OLT and any of the ONUs is maximum allowedGPON distance of 20 km. FTTB simulation consistsof three scenarios:

• Scenario B.1 was carried out to compare the SRand NSR DBA methods, in the FTTB networkwith low traffic load. Since bandwidth utiliza-tion is not important in such a case, it is the traf-fic delay (especially uplink delay), that has to beverified.

• Scenario B.2 models FTTB network, highly uti-lized with data traffic and with majority of theavailable bandwidth reserved for the assured traf-fic of T-CONTs of type 2 and 3.

• Scenario B.3 is a nearly overbooked networkwith high utilization by best effort traffic and rel-atively low assured traffic load.

The following parameters are used to define thenetwork configuration:

• AssuredBW - Amount of bandwidth OLT hasreserved for a T-CONT. This bandwidth will begiven to that T-CONT if required. Otherwise itwill be granted to other T-CONTs

• SourceBW - Data load generated by one T-CONT. This bandwidth is generated by one ormore workstations connected to T-CONT’s ONU

•AssuredBW

GPONBW- Ratio between bandwidth reserved

for high priority traffic of T-CONTs of type 2 and3, and the GPON link datarate

•SourceBW

GPONBW- Ratio between total traffic load gen-

erated by all T-CONTs and the GPON link datarate

5 Results

5.1 Network TuningThree different sets of simulations were performedto find the most optimum values for SR request fre-quency, EF and DBA cycle length.

Status Report request frequencySRrFreq indi-cates how often OLT sends SR request to each T-CONT. If T-CONT is polled for SR too often, it de-creases overal system throughput and may lead to in-compatibilities between OLT and ONU, due to in-correct timing. On the other hand too seldom SRpoll results in increased transmission delays. Resultsfrom network simulation with DBA cycle of 32 ms,AssuredBW

GPONBW= 0.63 and SourceBW

GPONBW= 0.76 with vary-

ing SR request frequency is shown in figure 2.

Figure 2: Uplink delay vs. SR frequency

It can be seen that the smallest delay experi-enced by data belonging to each T-CONT type, isobserved withSRrFreq = 3, which corresponds to



one SR every 3 frames. LowerSRrFreq values resultin multiple bandwidth assignment for the same data,and correspond to unacceptable high delays. HigherSRrFreq does not have significant influence on theoverall amount of assigned bandwidth as can be seenin figure 3, however it linearly increases the data de-lay.

Figure 3: Assigned Bandwidth vs. SR frequency

The Expansion factor EF determines the ratio forthe bandwidth assigned for current DBA cycle c tothe traffic used in the previous DBA cyclec− 1. HighEF results in greater bandwidth assignments for a T-CONT. This additional bandwidth compensates forthe potential variation in the traffic load. The trade offfor this lower delay is less efficient use of GPON link,since often this additional bandwidth is not fully uti-lized. Results from network simulation with DBA cy-cle of 32 ms,AssuredBW

GPONBW= 0.78 and SourceBW

GPONBW= 0.7

with varying EF is shown in figure 4. Decrease in

Figure 4: Uplink Delay vs. EF

in link delay of T-CONT types 2 and 3 correspondsto uplink delay increase for T-CONT of type 4, sinceincreased EF provides higher bandwidth allocationsfor T-CONTs of type 2 and 3, which results in smalleramount of bandwidth left for BE traffic from T-CONTof type 4 as show in figure 5. Expansion factor can

never be smaller than or equal to 1, since this wouldlead to a situation where the bandwidth for future cy-cle for T-CONTs of type 2 and 3 could either be con-stant or decrease. This would be an unacceptable casesince one cycle with lower data load would lead to re-duced bandwidth allocations for future cycles. In per-formed set of simulations with EF ranging between1 and 2, 1.25 turned out to be the best value, thatwould provide reasonable delay for all T-CONT types.Higher EF results in minimal uplink transmission de-lay decrease for T-CONTs of type 2 and 3, and rapiddelay increase for T-CONT of type 4. The optimalobserved value is 1.25.

DBA cycle length thus determines how often newcyclic bandwidth assignment for all T-CONTs shouldbe recalculated. Results from network simulationwith DBA cycle of 32 ms,AssuredBW

GPONBW= 0.78 and

SourceBW

GPONBW= 0.7 with varying EF is shown in figure

6.For the short DBA cycle length, T-CONTs of

types 2 and 3 suffer higher delays, while T-CONT oftype 4 experiences the least delay. The reason for T-CONTs of type 2 and 3 suffering higher delays, is thefact that, when traffic load measurements are taken fora shorter period, they are less uniform. With normaldistribution over longer time periods, traffic load tendsto be the average value of the normal distribution. The

Figure 5: Assigned Bandwidth vs. EF

Figure 6: Uplink Delay vs. DBA cycle



shorter the measuring period, the higher variation canbe observed. These delay values are not exceeding5 ms since the cycle length is short and bandwidthassigned per cycle can be relatively quickly recalcu-lated for new frame. Low DBA cycle time corre-sponds to more varying traffic assignments per cycle,while for larger cycle times they are more stable. Forhigher DBA cycle, less bandwidth is assigned per cy-cle, since during longer periods more information isgathered about T-CONTs load.

For T-CONT 4, the short DBA cycle time is bet-ter, since all bandwidth remaining after assignmentsto T-CONTs of type 2 and 3, will be distributed to theT-CONTs of type 4 that have completely utilized traf-fic assigned in the previous cycle. In order to ensurelowest delays for assured and non-assured traffic DBAcycle should be higher than 26 ms.

5.2 Fiber to the Building (FTTB)

The first FTTB scenario was simulated with thefollowing configuration: AssuredBW

GPONBW= 0.56,

SourceBW

GPONBW= 0.53 and equal source traffic load for

all T-CONTs. It provides a comparison of NSR andSR algorithms with relative low traffic load. The re-sults will show mainly the delay due to the schedul-ing and physical transmission delays, experienced bythe transmitted data. Simulation results show that forthe low traffic load of scenario B.1, both NSR and SRalgorithms serve all T-CONTs well with the SR aver-age uplink delay of 0.6 ms and NSR of 1.6 ms for allT-CONT types. Delay obtained using SR algorithm islower, since OLT provides almost immediate responsefor data transmission request. Delay using NSR algo-rithm is higher since bandwidth for assured and besteffort traffic is allocated every 1 ms.

Scenario B.2 shows NSR and SR algorithm per-formance with high traffic load. The following con-figuration was used:AssuredBW

GPONBW= 0.82

SourceBW

GPONBW=

0.89 and equal source traffic load for all T-CONTs.In congested networks, SR is more efficiently usingnetwork resources. With the SR DBA, transmissiondelays of 0.65 ms for T-CONTs of type 2 and 3 aresimilar to the slightly loaded network in B.1. Best ef-fort traffic for T-CONT type 4 also experiences verylow delay of 1 ms as shown in figure 7(a). Bandwidthutilization with SR reaches 90% of the total availablebandwidth. It can be seen that although T-CONT oftype 4 is granted the highest bandwidth it experienceshighest delay, due to the fact that bandwidth for besteffort traffic is assigned from upstream frame leftoversafter assignments for other T-CONTs. For that rea-son T-CONT of type 4 obtains bandwidth grants lessregularly, which results in higher delay. This irreg-

ularity also causes multiple assignment of bandwidthfor the same data. T-CONTs of type 4 are asked forSR every third frame. Since priority is given for T-CONTs of type 2 and 3, T-CONTs of type 4 usuallydo not get bandwidth immediately after OLT receivestheir status reports. Bandwidth for data is assignedby OLT one or two frames after receiving informa-tion about waiting data in response to the status re-port. In the mean time, second SR request is sentand T-CONT informs again about data waiting in thequeue. The problem occurs if OLT assigns bandwidthfor data before receiving second SR request. In thiscase OLT gets duplicated information about amountof data waiting in the queue. This leads to the du-plicated bandwidth grant, which will most likely notbe fully utilized by the T-CONT. Certainly T-CONTsof type 2 and 3 may experience the same duplicatedbandwidth assignment in congested network. How-ever T-CONTs of type 4 are more vulnerable to suchcases, due to its higher serving irregularity. Dupli-cated bandwidth assignment could be eliminated byintroduction of the status request time stamps. OLTshould register the time, when the last status reportwas obtained from a particular T-CONT. This infor-mation could be used in order to verify whether anydata assignment has been issued for that T-CONT inthe mean time.

NSR algorithm is slightly less bandwidth effi-cient, with data bandwidth utilization of 86%. T-CONTs of type 2 and 3 are given more additional butunused bandwidth. As a result not enough bandwidthremains for T-CONTs of type 4. Average of 648 Mbpsassigned to T-CONTs of type 4 is not sufficient toserve 744 Mbps source traffic load, as shown in fig-ure 7(a). Considering the fact that again T-CONTs oftype 4 often obtain small bandwidth slices left fromeach frame, their data packets will be heavily frag-mented. As a result data throughput for T-CONT oftype 4 is very low. 15% of the best effort traffic doesnot receive bandwidth. In this case transmission delayof that T-CONT type becomes a minor issue.

SR DBA proved to be a better choice for this net-work configuration, since it offers low transmissiondelays for both real-time and best effort traffic, andenough bandwidth is assigned for all T-CONT types.

The third scenario B.3 was chosen to test NSRand SR algorithms with high traffic load, dominatedby best effort traffic of T-CONT 4. The following net-work configuration was used:AssuredBW

GPONBW= 0.3 and

SourceBW

GPONBW= 0.9, source load for T-CONTs 2 and 3

- 432 Mbps each and for T-CONT 4 - 1462 Mbps.Results of the simulation are shown in figure 8(a)and 8(b) . The clear advantages and disadvantagesof the two DBA algorithms can be observed in this



(a) Assigned bandwidth

(b) Uplink transmission delay

Figure 7: Simulation results for Scenario B.2.



(a) Assigned bandwidth

(b) Uplink transmission delay

Figure 8: Simulation results for Scenario 3.



high utilized network with high percentage of best ef-fort traffic. In this scenario T-CONT of type 2 assuredbandwidth is:AssuredBW = 1.1 ∗ SourceBW

When NSR DBA is used, T-CONTs of type 3are experiencing smallest delay of average 1.7 ms,while best effort traffic the highest 7.4 ms. T-CONTsof type 3 obtain the best service due to their maxi-mum bandwidth settings. Due to the decrease of safebandwidth part given to T-CONTs of type 2, they areexperiencing slightly higher delay than T-CONTs oftype 2. However this average of 3.6 ms delay, isstill acceptable, since real-time data requires delay ofmaximum 5 ms. The maximum bandwidth setting1.1 times higher than source bandwidth is enough toeliminate the risk of traffic rejection, and it increasesthe bandwidth utilization to 90%. This utilization ishigher than in the previous scenario, where NSR ob-tained utilization of only 86%. This indicates, that thedifference of bandwidth utilization between NSR andSR algorithms is not particularly due to the estimationof bandwidth assignment itself, but mostly due to theadditional security bandwidth given to T-CONTs oftype 2 and 3.

Bandwidth assignment for the T-CONT of type 3with NSR DBA is much higher than bandwidth as-signment for T-CONT of type 3 with SR. This indi-cates that T-CONTs of type 3 are granted much morebandwidth when EF is used. Despite the reduction ofassured bandwidth for T-CONT of type 2, best efforttraffic from T-CONTs of type 4 gets bandwidth as-signments irregularly, due to high grants for T-CONT3. In any case, delay below 10 ms for best effort trafficis sufficient.

The total data utilization of the network is 90%,both for NSR and SR DBA. Over 5 ms delay for T-CONTs of type 2 with SR DBA indicates that thisDBA method tends to duplicate bandwidth grants.When new DBA cycle starts and OLT tends to allo-cate more bandwidth to T-CONT of type 2, due to thecongestion in T-CONT’s queue. The bandwidth as-signment is duplicated from time to time, which leadsto reaching maximum bandwidth limit slightly beforeend of current DBA cycle. Queues get congested, andthey are being emptied at the beginning of next cycle,causing duplicated bandwidth assignment. Due to thisuneven distribution of bandwidth grants over the cyclein SR DBA, it is necessary to have bandwidth limitsslightly higher than the expected source traffic.

NSR DBA seems to be better for the real-timetraffic, due to the fact that bandwidth assignments aregiven regularly every 1 ms. Delay of traffic belongingto T-CONTs of type 4 is, on the other hand, lower thanwhen SR DBA is used (2.4 ms comparing to 7.4 ms).

The NSR DBA delay for T-CONTs of type 3 re-mains around 1.7 ms, as in previous scenarios. It can

be observed that throughout the simulations this tendsto be the average value for T-CONTs with a total ofassured and non-assured traffic grants not less than1.25 * source bandwidth. This delay seems to be in-dependent of the traffic load of traffic with other pri-orities, provided that no overbooking for the high pri-ority traffic has been performed. In order to maintainlow delays for real-time traffic, when network is be-ing planned, it should be the amount of assured band-width granted for T-CONTs of type 2 and 3, and notthe source bandwidth itself, which should be includedin the calculation of the load. No overbooking can bemade for this bandwidth grants.

6 Conclusion

SR algorighm utilizes bandwidth more efficiently thanNSR. In the low loaded networks where a lot of band-width is available, SR algorithm allocates around 5%more bandwidth than T-CONTs source load. Effi-ciency of NSR algorithm depends highly on the ex-pansion factor used, and the assured bandwidth pa-rameter of the T-CONT. If high assured bandwidth isgranted, with expansion factor of 1.25, T-CONTs ob-tain bandwidth grants for up to 129% of the sourcebandwidth. This security bandwidth allocated for as-sured traffic is not fully utilized by data. This leads todegradation in throughput for data transmission. Ob-viously T-CONTs with the lowest priority suffer thehighest delay and data drops due to the low band-width utilization, since those T-CONTs can only uti-lize bandwidth chunks left after assignment of assuredand non-assured bandwidth grants.

In NSR DBA, the compromise of high QoS forT-CONTs with high priority data and T-CONTs withlow priority data is the reduction of the assured band-width grants. It was verified that the assured band-width parameter set to 110% of the source bandwidthis sufficient to keep the average delay below 5 ms.Low average delay is obtained through the periodi-cal bandwidth allocations, with period of 1ms. Thisimplies that selection of the T-CONT’s bandwidth pa-rameters should be performed very carefully, since itis possible for the multiple T-CONT types to obtainreasonable service quality.

If a generic situation is considered, where the sys-tem load may vary, the most reliable solution for effi-cient handling of delay sensitive traffic is NSR algo-rithm. This is although for low traffic load and highadditional bandwidth, delays, using NSR, are higherthan the case when SR is used. Regular assignmentof grants every 1 ms gives guarantee for appropriatehandling. If the generated high priority traffic doesnot exceed the available bandwidth and no overbook-



ing on assured bandwidth is done, data traffic doesnot experience delay higher than 5 ms. NSR is notvulnerable to the duplicated bandwidth grants, whichimplies that bandwidth allocated for each T-CONT isused more efficiently.

Based on the simulation results, it can be con-cluded that NSR DBA is more reliable in providingagreed QoS, and, with appropriate network configu-ration optimized for amount of connected ONU’s andtheir traffic requirements, it can sufficiently serve alltraffic types. The SR algorithm does not provide andguarantee QoS in all network configurations, but itis efficiently using available resources. NSR is morerecommended for networks, where transmission de-lay has to be maintained below a specified minimumlevel. This solution however requires more bandwidthdue to the over-allocated bandwidth.

The choice of the algorithm should be made de-pending on the QoS SLAs agreed upon with particu-lar customers and the dominant traffic type in the net-work.

Acknowledgements: Special thanks goes to OmerAlptekin Yurdal for all the support. The research wassupported by the Technical University of Denmark(DTU) and Nokia Siemens Networks.

References:

[1] ITU-T, G.984.1 Gigabit-capable Passive Opti-cal Networks (GPON): General characteristics,Mar. 2003.

[2] ITU-T, G.983.2 Broadband optical access sys-tems based on Passive Optical Networks (PON),Jul. 2005.

[3] ITU-T, G.984.3. Gigabit-capable Passive Opti-cal Networks (G-PON): Transmission conver-gence layer specification, Mar. 2008.

[4] ITU-T, G. 983.4 A broadband optical access sys-tem with increased service capability using dy-namic bandwidth assignment, Nov. 2001.

[5] ITU-T, G.984.2 Gigabit-capable Passive Opti-cal Networks (GPON): Physical Media Depen-dent (PMD) layer specification, Mar. 2003.

[6] Joanna Ozimkiewicz,Competing BroadbandAccess Networks, Master’s thesis, Danish Tech-nical University, Nokia Siemens Networks, Jan.2009.

[7] OPNET Technologies, Inc.,http://www.opnet.com.

[8] Aida Salihovic, Matija Ivekovic,Gigabit PassiveOptical Network GPON, Jun. 2007.

[9] Rich Baca and Muneer Zuhdi,Technologicalchallenges to G-PON operation, Feb. 2008.

[10] Joanna Ozimkiewicz and Sarah Ruepp andLars Dittmann and Henrik Wessing and SylviaSmolorz, Dynamic Bandwidth Allocation inGPON Networks, Proc. CISST Conference atHarvard University, 2010

[11] M. Thomsen and T. S. Lyster and M. Berger andB. Mortensen and B. Srensen,Development plat-form for dynamic bandwidth allocation schemesin future MPCP enabled Ethernet Passive Op-tical Network (EPON),WSEAS Transactions onCommunications, 2006, vol. 5, issue 1, pages92-98

[12] C. G. Park and Y. Lee and D. H. Han and B.Kim, Waiting time analysis of gated polling sys-tem for bandwidth allocation in a passive opticalnetwork, WSEAS Transactions on Communica-tions, 2005, vol 4 , issue 10, p. 963-970

[13] S. V. Kartalopoulos, Security and bandwidthelasticity aspects of the CWDM/TDM-PON net-work, WSEAS Transactions on Communica-tions, 2006, vol 5, issue 8, p. 1461-1468

[14] T. Orphanoudakis and H.-C. Leligou and J. D.Angelopoulos,Next generation ethernet accessnetworks: GPON vs. EPON, Proceedings of the7th WSEAS International Conference on Elec-tronics Hardware, Wireless and Optical Commu-nications (EHAC ’08), 2008

[15] OPNET Technologies, OPNET Modeler,www.opnet.com



dba

Documents