Request-based Dynamic Bandwidth Allocation of Gigabit Passive Optical Network

doi:10.4236/opj.2013.32B040

Paper Menu >>

Journal Menu >>

Optics and Photonics Journal, 2013, 3, 165-170

doi:10.4236/opj.2013.32B040 Published Online June 2013 (http://www.scirp.org/journal/opj)

Request-based Dynamic Bandwidth Allocation of Gigabit

Passive Optical Network

Chih-Ta Chiu, Yung-Chung Wang

Department of Electrical Engineering, National Taipei University of Technology, Taipei, Chinese Taipei

Email: [email protected]

Received 2013

ABSTRACT

We propose request-based dynamic bandwidth allocation (DBA) of gigabit passive optical network (GPON). The opti-

cal line terminal (OLT) in GPON grants bandwidth to optical network units (ONUs). ONUs report request bandwidth

which depends on queue lengths of traffic containers (TCONTs) to the OLT. In the OLT, DBA of GPON supports a

request-based polling order to allocate bandwidth. Our request-based dynamic bandwidth allocation focuses on weight

assignments in the request-based polling order. Weight assignments allocate bandwidth in proportion to guaranteed and

request bandwidth. We use the C program to simulate results. Simulated results indicate improved performance in

queueing delay when total offered loads are or are not shared uniformly to TCONTs.

Keywords: GPON; DBA; Weight; Polling

1. Introduction

GPON [1], a network of data transmission between the

optical line terminal (OLT) and optical network units

(ONUs), provides point to multipoint (P2MP) broadband

access networks. To achieve P2MP, tree topology [2]

gives a high distance extension solution because low

lengths of fiber links make a low cost. In the upstream

direction, time-division multiplexing access (TDMA) [3]

is used. ONUs send packets in their own time slots to the

OLT. In the downstream direction, the OLT broadcasts

packets in the downstream direction to ONUs. Headers

of packets not only make the OLT assign time slots to

ONUs but also make ONUs report their queue lengths to

the OLT. By reporting queue lengths, time slots are as-

signed dynamically.

DBA [4], based on the TDMA protocol, improves

performance compared with performance of the static

bandwidth allocation. DBA is achieved by header fields

of packets [5]. Header fields in the downstream direction

includes allocation identifiers (AllocIDs) to identify

TCONTs, start time pointers to denote the start time of

time slots, and end time pointers to denote the end time

of time slots. Those fields specify that only one of

TCONTs is served at any time. Header fields in the up-

stream direction include AllocIDs to identify TCONTs

and dynamic bandwidth reports (DBRs) to report queue

lengths in TCONTs. DBRs notifies the OLT to allocate

bandwidth dynamically due to queue lengths of ONUs

are different. Bandwidth is allocated dynamically to sat-

isfy queue lengths as soon as possible so that perform-

ance is improved.

The literature of DBA is rich. Instead of constant time

slots, [6] achieves dynamic time slots. Headers in the

upstream and downstream direction are used to grant

bandwidth and report queue lengths, respectively. The

time interval within a time slot only depends on reported

queue lengths. Based on service level agreement (SLA),

[7] reserves a guaranteed and non-guaranteed bandwidth

allocation policy. Bandwidth is cut into multiple grids

which includes guaranteed or non-guaranteed bandwidth.

Then, grids are polled for ONUs so that the policy is

achieved. [8] determines which ONU is allocated guar-

anteed or non-guaranteed bandwidth. A call admission

control (CAC) generates characters of ONUs as

non-guaranteed, guaranteed, or delay guaranteed band-

width. Then, by evenly delay algorithm, ONUs are put in

grids. Then, by BGP, bandwidth is polled from the first

to the last grid for ONUs. [9] provides a multimedia ser-

vice by allocating high to low priority bandwidth called

fixed, assured, and best effort bandwidth. In [10] and

[11], a class-based bandwidth scheme provides the dif-

ferentiated service because minimum function in the

OLT compares critical values with request bandwidth to

choose a smaller one. The OLT allocates bandwidth to

queues of different traffic classes. Then, ONUs report

queue lengths according to traffic classes. [12] sets in-

ter-ONU and intra-ONU weights so that bandwidth can

be allocated in proportion to weights. Types of the

scheduling can be weighted or hierarchical round robin.

C.-T. CHIU, Y.-C. WANG

166

In the DBA called prioritized weighted round robin

(PWRR) [5] [13], four kinds of bandwidth from high to

low priority bandwidth are fixed, assured, non-assured,

and best-effort bandwidth. Fixed and assured bandwidth

is guaranteed bandwidth while non-assured bandwidth

and best-effort bandwidth is surplus bandwidth. Critical

values include maximum and minimum bandwidth.

Critical values of non-assured bandwidth is in proportion

to critical values of assured bandwidth. The minimum

function compares request bandwidth with critical values

to choose a smaller one. Polling types are weighted

round robin (WRR). Time slots for bandwidth of four

priorities bandwidth are allocated every 125 us. [14]

gives different offered loads in its simulation to evaluate

performance of DBA.

We propose the request-based DBA called prioritized

adaptive request-based polling (PARP) in GPON. PARP

includes weight assignments and request polling. In

weight assignments, non-assured bandwidth is allocated

in proportion to weights. Two parameters are adaptable

for weights which are in proportion to assured and re-

quest bandwidth. Mention to the request-based polling,

instead of the fixed polling order of WRR, polling order

in PARP is based on the highest request bandwidth

among all request bandwidth within the same type of

TCONTs. Therefore, each type of TCONTs has its own

polling order. PARP is based on request bandwidth to

achieve DBA and polling. We use a C program to simu-

late results. Simulate results improve performance in

queueing delay when total offered loads are or are not

shared uniformly to TCONTs.

2. Proposed Method

2.1. Operations

GPON shown in Figure 1 includes OLT, ONUs and the

optical distribution network (ODN). OLT is responsible

for allocating bandwidth while ONUs are responsible for

reporting their queue lengths. ODN connects between the

OLT and ONUs to achieve broadcast in the downstream

direction and TDM in the upstream direction. Queues in

ONUs are TCONTs with queue lengths. Each TCONT is

identified by an AllocID so that the OLT can use Al-

locIDs to identify TCONTs.

Relationships between TCONTs and bandwidth are

shown in Figure 2. TCONT1 is allocated fixed band-

width which does not consider request bandwidth.

TCONT2 is allocated assured bandwidth. TCONT3 is

allocated assured and non-assured bandwidth. TCONT4

is allocated best-effort bandwidth. TCONT5 is the type

of test queues so that it is allocated fixed, assured, non-

assured, best-effort, and combined bandwidth. Priorities

to allocate bandwidth are shown in Figure 3. Fixed and

assured bandwidth is guaranteed bandwidth while non-

assured bandwidth and best-effort bandwidth is surplus

bandwidth. Order from high to low priority is fixed, as-

sured, non-assured, and best-effort bandwidth.

2.2. Assumptions

In [13], DBA for TCONT 3 is specified:

 For the surplus bandwidth allocation to TCONT 3,

setting all SImin parameters equal and varying the

ABsur parameters, a weighted round robin service is

enforced. If all the ABsur parameters are also set

equal, the service becomes equivalent to a simple

round robin.

 The guaranteed service rate is expressed as: AB-

min/SImax, while the surplus (non-guaranteed) ser-

vice rate is expressed as: ABsur/SImin. The guaran-

teed rate and the surplus rate sum up to the allowed

peak rate for TCONT 3.

 Regarding TCONT 3 AllocIDs, a GIR = 1/3PIR was

used.

Figure 1. GPON.

Figure 2. Relationships between bandwidth and TCONTs.

Figure 3. Priorities of the bandwidth allocation.

C.-T. CHIU, Y.-C. WANG 167

To compare our proposed method with PWRR, we

assume that PWRR in this paper sets all SImax and SImin

parameters equal and varies ABsur parameters so that a

WRR service is enforced. We follows the GIR = 1/3PIR.

Because the peak information rate (PIR) is three times of

guaranteed information rate (GIR), with setting SImax

and SImin equal, ABmin = 1/2ABsur is used for TCONT3.

In this paper, ABmin and ABsur for TCONT3 is

and , respectively.

max

TCONT

min

TCONT

2.3. Prioritized Adaptive Weighted Round Robin

We propose prioritized adaptive weighted round robin

(PAWRR) with weight assignments and without request-

based polling. Equation (1) and (2) are maximum band-

width. Maximum bandwidth is critical values. Maximum

bandwidth of TCONT2 is determined by Offered-

LoadTCONT2 while maximum bandwidth of TCONT3 is

determined by OfferedLoadTCO NT 3 . Equation (3) and (4)

are assured bandwidth. Assured bandwidth of TCONT2

and TCONT3 is allocated by comparing request band-

width with maximum bandwidth to choose a smaller one.

max .

TCONT TCONT

B OfferedLoad2

(1)

max .

TCONT TCONT

B OfferedLoad (2)

max

min( ,).

TCONTTCONT TCONT

assured request

BBB (3)

max

min( ,).

TCONTTCONT TCONT

assured request

BBB (4)

Equation (5), (6), and (7) are surplus, minimum band-

width, and weight assignments, respectively. Surplus

bandwidth is allocated for non-assured bandwidth by

weights of TCONT3. Weights of TCONT3 are assigned

in proportion to assured bandwidth and request band-

width. Parameters of α and β are the influence of assured

and request bandwidth, respectively. The sum of α and β

is 1.0. Equation (8) is non-assured bandwidth. Non-as-

sured bandwidth is allocated by choosing the smallest

one among request bandwidth, minimum bandwidth, and

frame bytes bandwidth. Frame bytes bandwidth is unused

bandwidth after allocating bandwidth.

()

TCONT TCONT

surplustotali assuredi assured

BB BB



 



. (5)

min

TCONT

TCONT surplus NTCONT







(6)

()

TCONT

TCONT TCONT

assured request

TCONT

NTCONT TCONT

i assuredi request













(7)

333

min

min( ,,).

TCONTTCONT TCONT

non assuredrequestframe bytes

B BBB





Equation (9) and (10) are minimum and best-effort

bandwidth, respectively. Best-effort bandwidth is allo-

cated by comparing request bandwidth with frame bytes

bandwidth to choose a smaller one.

min .

TCONT

rame bytes



 (9)

min

min( ,).

TCONTTCONT TCONT

best effortrequest

BBB

4

(10)

2.4. Prioritized Adaptive Request-based Polling

We proposed prioritized adaptive request-based polling

(PARP) with weight assignments and request-based

polling. Figure 4 is the scheme of request-based polling.

Instead of a count-down timer, we use the 0-1 trigger

called SImax or SIm in to determine which TCONT can

be allocated bandwidth. When a TCONT can and can not

allocate bandwidth, trigger value is 1 and 0, respectively.

In the beginning, we copy requests of N TCONT2, 3,

and 4 as TempTCONT2, TempTCONT3, and TempTCONT4

in number u time slot, respectively. Then, we use

MatchTCONT2, MatchTCONT3, and MatchTCONT4 to

choose the highest request bandwidth from TempT-

CONT2, TempTCONT3, and TempTCONT4, respectively.

If MatchTCONT2 is 0, the polling order follows the order

of weighted round robin due to all request bandwidth of

TCONT2 is zero for no comparison. If MatchTCONT2 is

not 0, the polling order is the highest request bandwidth

TCONT2. After MatchTCONT2, it turns to MatchT-

CONT3. If MatchTCONT3 is 0, the polling order follows

the order of weighted round robin due to all request

bandwidth of TCONT3 is zero for no comparison. If

MatchTCONT3 is not 0, the polling order is the highest



(8)

Figure 4. Request-based polling.

C.-T. CHIU, Y.-C. WANG

168

request bandwidth TCONT3. After MatchTCONT3, it

turns to MatchTCONT4 . If Ma tch T CO NT4 is 0, the poll-

ing order follows the order of weighted round robin due

to all request bandwidth of TCONT4 is zero for no com-

parison. If MatchTCONT4 is not 0, the polling order is

the highest request bandwidth TCONT4. Simax and

Simin follow the position of MatchTCONT2, MatchT-

CONT3, and MatchTCONT4 to set its value as 1 for the

bandwidth allocation. Then, DBA works for the band-

width allocations of TCONTs whose SImax or SImin is 1.

After DBA, its value is set 0.

2.5. Examples

The example includes one ODN between three ONUs

and one OLT. Each ONU includes three TCONTs which

are one TCONT2, one TCONT3, and one TCONT4. The

OLT includes a table about request bandwidth shown in

Figure 5. In this example, we use parameters in unit of

bytes.

rame bytes

B comes from 125 × 103 / 8 / 1 = 15625

bytes. OfferedLoadTCONT2 is 6000 bytes and Offered-

LoadTCONT3 is 3000 bytes so that is 6000 bytes

and is 3000 bytes. Request bytes of three

TCONTs of ONU1, ONU2, and ONU3 are 9000, 6000,

and 3000 bytes, respectively. The PARP table is ob-

served within three frame bytes. In this example, guard

bytes are ignored, pre-assured bandwidth of TCONT2

and 3 is 3000 bytes, and β is 1.0.

max

TCONT

max

TCONT

In the first frame bytes, firstly, max(9000, 6000, 3000 )

= 9000 bytes so that TCONT2 of ONU1 is chosen. In the

first frame bytes, request bytes of TCONT2 of ONU1

decrease from 9000 bytes to 3000 bytes due to min(9000,

6000) = 6000 bytes for the assured bandwidth allocation.

Remainders of frame bytes decrease from 15625 bytes to

9625 bytes.

Figure 5. DBA with PARP.

Secondly, max(9000, 6000 , 3000) = 9000 bytes so that

TCONT3 of ONU1 is chosen. Request bytes of TCONT3

of ONU1 decrease from 9000 bytes to 6000 bytes due to

min(9000, 3000) = 3000 bytes for the assured bandwidth

allocation. Remainders of frame bytes decrease from

9625 bytes to 6625 bytes. Request bytes of TCONT3 of

ONU1 decrease from 6000 bytes to 0 byte from (15625

× 3 - 6000 - 3000 × 5) × 6000 / (6000 + 6000 + 3000) =

10350 bytes and min(6000, 10350, 6625) = 6000 bytes

for the non-assured bandwidth allocation. Remainders of

frame bytes decrease from 6625 bytes to 625 bytes.

Thirdly, max(9000, 6000, 3000) = 9000 bytes so that

TCONT4 of ONU1 is chosen. Request bytes of TCONT4

of ONU1 decrease from 9000 bytes to 8375 bytes due to

min(9000, 625) = 625 bytes for the best-effort bandwidth

allocation.

In the second frame bytes, firstly, TCONT2 of ONU2

is chosen due to max(3000, 6000, 3000) = 6000 bytes

which turns to TCONT2 of ONU2. Firstly, in PARP ta-

ble of the second frame bytes, request bytes of TCONT2

of ONU2 decrease to 0 bytes for the assured bandwidth

allocation due to min(6000, 6000) = 6000 bytes. Re-

mainders of frame bytes decrease from 15625 bytes to

9625 bytes.

Secondly, max(0, 6000, 3000) = 6000 bytes so that

TCONT3 of ONU2 is chosen. Request bytes of TCONT3

of ONU2 decrease from 6000 bytes to 3000 bytes due to

min(6000, 3000) = 3000 bytes for the assured bandwidth

allocation. Then, request bytes of TCONT3 of ONU1

decrease from 3000 bytes to 0 bytes from (15625 × 3 -

6000 - 3000 × 5) × 6000 / (0 + 6000 + 3000) = 17250

bytes and min(3000, 17250, 6625) = 3000 bytes for the

non-assured bandwidth allocation. Remainders of frame

bytes decrease from 6625 bytes to 3625 bytes.

Thirdly, TCONT4 of ONU1 is chosen due to

max(8375, 6000, 3000) = 8375 bytes. Request bytes of

TCONT4 of ONU1 decreases from 8375 bytes to 4750

bytes due to min(8375, 3625) = 3625 bytes for the best

effort bandwidth allocation. Remainders of frame bytes

decrease from 3625 bytes to 0 byte.

In the third frame bytes, firstly, due to max(3000, 0,

3000) = 3000 bytes, it turns to TCONT2 of ONU1. In

PARP table of the third frame bytes, request bytes of

TCONT2 of ONU1 decrease from 3000 bytes to 0 byte

due to min(3000, 6000) = 3000 bytes for assured band-

width allocation. Remainders of frame bytes decrease

from 15625 bytes to 12625 bytes.

Secondly, due to max(0, 0, 3000) = 3000 bytes,

TCONT3 of ONU3 is chosen. Request bytes of TCONT3

of ONU3 decrease from 3000 bytes to 0 byte. Remain-

ders of frame bytes decrease from 12625 bytes to 9625

bytes due to min(3000, 3000) = 3000 bytes for the as-

sured bandwidth allocation.

Thirdly, due to max(4750, 6000, 3000) = 6000 bytes,

C.-T. CHIU, Y.-C. WANG 169

TCONT4 of ONU2 is allocated best-effort bandwidth.

Request bytes of TCONT4 of ONU2 decrease from 6000

bytes to 0 byte due to min(6000, 9625) = 6000 bytes.

Remainders of frame bytes decrease from 9625 bytes to

3625 bytes. PARP allocates high request bytes.

3. Performance Evaluation

3.1. Parameters

GPON is simulated in a C program to evaluate perform-

ance of PARP with 1 OLT and 8 ONUs. Mention to

simulated parameters, bandwidth of GPON is set as

1.24416 Gbps and network capacity is set as 1 G. The

frame duration is set as 125 us. Propagation delay be-

tween OLT and ONUs is set as 200 us. Guard bytes are

set as 20 bytes. The size of TCONTs is set as 10 Mbytes.

TCONT types are one TCONT2, one TCONT3, and one

TCONT4 in each ONU. The network traffic is generated

by exponential inter-arrival time and packet sizes are 64

bytes, 500 bytes, and 1500 bytes with the probability 0.6,

0.2, and 0.2, respectively. Parameters of O fferedLoadTCONT2

and OfferedLoad TCONT3 are 9500 bytes and 3000 bytes,

respectively. In the scenario1 of queueing delay with

sharing uniformly total offered loads, total offered loads

are shared 1/24 to each of 24 TCONTs in 8 ONUs. In the

scenario2 of queueing delay without sharing uniformly

total offered loads, total offered loads are shared 1/16 to

each of 12 TCONTs in 4 ONUs and 1/48 to each of 12

TCONTs in 4 ONUs.

3.2. Scenario1

Figure 6 is the comparison of queueing delay when total

offered loads are shared uniformly to TCONTs. In the

1.7 Gbps total offered load, four kinds of queueing delay

of TCONT3 of PAWRR with setting



1.0, PAWRR

with setting



1.0, PARP, and PWRR are 3.97 ms,

3.70 ms, 3.62 ms, and 3.80 ms, respectively.

Compared with 3.97 ms and 3.80 ms queueing delay,

of TCONT3 of PAWRR with setting

min

TCONT



1.0 is

low since heavy loads make

urplus

3TCONT

3NT

low and

fixed. Fixed makes fixed. of

TCONT3 of PWRR is set only in proportion to

which is a critical value of assured bandwidth. Therefore,

of TCONT3 of PAWRR is lower than

of TCONT3 of PWRR. non assured is allocated more to

TCONT3 of PWRR than TCONT3 of PAWRR setting

3TCONT

assured

max

TCONT

min

TCONT

min

TCO

3TCONT

assured

min

TCONT

BTCO

B



1.0.

Compared with 3.80 ms and 3.70 ms queueing delay,

of TCONT3 of PAWRR with setting

min

TCONT



1.0 is

high to high request bandwidth TCONT3 and low to low

request bandwidth TCONT3 since the weights are set in

proportion to . Even of TCONT3 of

PWRR is high, low is chosen by the minimum

function and high needs more bandwidth than

. Therefore, assured is allocated more to

TCONT3 of PAWRR with setting

3TCONT

request

BTC

min

TCONT

3ONT

request

3TCONT

request

BTCON

non

B

min

TCONT

B3T



1.0 than TCONT3

of PWRR.

Compared with 3.70 ms and 3.62 ms queueing delay,

WRR of PAWRR follows a fixed polling order. When

request bandwidth is low, the request bandwidth is still

polled. This results in low bandwidth allocation from a

minimum function. PARP chooses highest request band-

width to be polled. Therefore, more bandwidth is allo-

cated due to request bandwidth is high.

3.3. Scenario 2

Figure 7 is the comparison of queueing delay when total

offered loads are not shared uniformly to TCONTs. In

the 1 Gbps total offered load, four kinds of queueing

delay of TCONT3 of PAWRR with setting



1.0,

PAWRR with setting



1.0, PARP, and PWRR are

5.18 ms, 5.12 ms, 1.82 ms, and 8.17 ms, respectively.

Compared with 5.18 ms and 8.17 ms queueing delay,

urplus is high since 1 Gbps is light loads for TCONT3.

Therefore, min of TCONT3 of PAWRR is higher

than minof TCONT3 of PWRR. is al-

located more to TCONT3 of PAWRR with setting

3TCONT

3NTO3TCONT

non assured

B



1.0 than TCONT3 of PWRR. Compared with 5.18 ms

and 5.12 ms queueing delay, of TCONT3 of

PAWRR with setting

min

TCONT



1.0 is high to high request

bandwidth TCONT3 and low to low request bandwidth

TCONT3 since the weights are set in proportion to

urplus . Therefore, assured is allocated more to

TCONT3 of PAWRR with setting

B3TCONT

non

B



1.0 than TCONT3

of PAWRR with setting



1.0.

Figure 6. Performance of queueing delay in scenario1.

Figure 7. Performance of queueing delay in scenario2.

C.-T. CHIU, Y.-C. WANG

170

Compared with 5.12 ms and 1.82 ms queueing delay,

WRR in PAWRR follows a fixed polling order. Times to

serve each TCONT is equal. However, TCONTs with

high and less loads need more and less polling times,

respectively. PARP gives more and less polling times to

TCONTs with high and less loads, respectively. There-

fore, queueing delay of PARP is lower than queueing

delay of PAWRR with setting



1.0.

By different polling times, besides TCONT3, PARP

improves queueing delay of TCONT2 and 4. In the 1.3

Gbps total offered load, four kinds of queueing delay of

TCONT2 of PAWRR with setting



1.0, PAWRR

with setting



1.0, PARP, and PWRR are 8.76 ms,

8.76 ms, 2.10 ms, and 8.76 ms, respectively. In the 0.6

Gbps total offered load, four kinds of queueing delay of

TCONT4 of PAWRR with setting



1.0, PAWRR

with setting



1.0, PARP, and PWRR are 2.87 ms,

2.80 ms, 1.27 ms, and 2.81 ms, respectively.

4. Conclusions

We propose request-based DBA called PARP in GPON.

PARP allocates min-max bandwidth with weights for

critical values. When weights are assigned in proportion

to request bandwidth, non-assured bandwidth is allocated

more and less to high and less request bandwidth

TCONT3, respectively. Non-assured bandwidth is adapted

so that it adapts varying request bandwidth. Beside the

bandwidth allocation, request-based polling is used to

poll the highest request bandwidth TCONTs in the same

type. PARP allocates more bandwidth to TCONTs of

high request bandwidth. By a C program, simulations are

evaluated when total offered loads are or are not uni-

formly shared to TCONTs. Simulative results indicate

queueing delay in proportion to request bandwidth is

better than one in proportion to guaranteed bandwidth

and critical values. Simulative results also indicate

queueing delay is improved when the polling order is

chosen high request bandwidth TCONTs.

REFERENCES

[1] ITU Rec. G.984.1, “Gigabit-Capable Passive Optical

Networks (G-PON): General Characteristics,” Geneva,

Switzerland, 2003.

[2] M. Mcgarry, M. Maier and M. Reisslein, “Ethernet Pas-

sive Optical Network (EPON): Building A

Next-Generation Optical Access Network,” IEEE Com-

munications Magazine, Vol. 40, No. 2, 2002, pp. 66-73.

doi:10.1109/35.983910

[3] D. Angelopoulos, S. Venieris and I. Stassinopoulos, “A

TDMA Based Access Control Scheme for APON's,”

Journal of Lightwave Technology, Vol. 11, No. 5, 1993,

pp. 1095-1103. doi:10.1109/50.233273

[4] ITU Rec. G.983.4, “A Broadband Optical Access System

with Increased Service Capability Using Dynamic Band-

Width Assignment,” Geneva, Switzerland, 2001.

[5] J. Angelopoulos, H. Leligou, T. Argyriou and S. Zontos,

“Efficient Transport of Packets with QoS in An

FSAN-Aligned GPON,” IEEE Communications Maga-

zine, Vol. 42, No. 2, 2004, pp. 92-98.

doi:10.1109/MCOM.2003.1267106

[6] G. Kramer, B. Mukherjee and G. Pesavento, “IPACT: A

Dynamic Protocol for An Ethernet PON,” IEEE Commu-

nications Magazine, Vol. 40, No. 2, 2002, pp. 74-80.

doi:10.1109/35.983911

[7] M. Ma, Y. Zhu and T. Cheng, “A Bandwidth Guaranteed

Polling MAC Protocol for Ethernet Passive Optical Net-

works,” IEEE INFOCOM, Vol. 1, No. 30, 2003, pp.

22-31.

[8] Y. Zhu, M. Ma and T. Cheng, “A Novel Multiple Access

Scheme for Ethernet Passive Optical Networks,” IEEE

GLOBECOM, Vol. 22, No. 1, 2003, pp. 2649-2653.

[9] S. Choi and J. Huh, “Dynamic Bandwidth Allocation

Algorithm for Multimedia Services Over Ethernet

PONs,” IEEE INFOCOM, Vol. 1, No. 30, 2003, pp.

22-31.

[10] C. Assi, Y. Ye and S. Dixit, “Support of QoS in

IP-based”

[11] Ethernet PON, IEEE GLOBECOM, Vol. 7, No. 1, 2003,

pp. 3737-3741. doi:10.1080/028418501127346846

[12] J. Xie, S. Jiang and Y. Jiang, “A Dynamic Bandwidth

Allocation Scheme for Differentiated Services in

EPONs,” IEEE Communications Magazine, Vol. 42, No.

8, 2004, pp. 32-39. doi:10.1109/MCOM.2004.1321385

[13] B. Chen, J. Chen and S. He, “Efficient and Fine Schedul-

ing Algorithm for Bandwidth Allocation in Ethernet Pas-

sive Optical Networks,” IEEE Journal of Selected Topics

in Quanium Electronicsxxcta Radiologica, Vol. 12, No. 4,

2006, pp. 653-660. doi:10.1109/JSTQE.2006.876607

[14] H. Leligou, C. Linardakis, K. Kanonakis, J. Angelopoulos,

and T. Orphanoudadkis, “Efficient Medium Arbitration of

FSAN-Compliant GPONs,” Journal of Communication

Systems, Vol. 19, No. 5, 2006, pp. 603-607.

doi:10.1002/dac.761

[15] C. Su-il and P. Jaehyung, “SLA-Aware Dynamic Band-

Width Allocation for QoS in EPONs,” IEEE/OSA Journal

of Optical Communications and Networking, Vol. 2, No.

9, 2010, pp. 773-781.doi:10.1364/JOCN.2.00077