Content uploaded by Xavier N Fernando
Author content
All content in this area was uploaded by Xavier N Fernando on Jan 05, 2014
Content may be subject to copyright.
EPON:
AN EXTENSIVE REVIEW FOR UP-TO-DA
TE DYNAMIC
BANDWIDTH ALLOCATION SCHEMES
Sattar Hussain and Xavier Fernando, Senior Member, IEEE
Dept. of Elect. and Comp. Engineering
Ryerson University, Toronto, Canada
ABSTRACT
Ethernet Passive Optical Network (EPON) has been widely
studied in literatures in the past few years. Researchers from
all around the world are investigating EPON main challenges
and Dynamic Bandwidth Allocation (DBA) problems. This
paper, reviews the most recent studies conducted in EPON
networks and presents the new proposed DBA schemes. The
reviewed schemes are classified according to the main chal-
lenge addressed by the investigator. A brief outline is given
for each one along with a discussion of its performance and
possible contribution to enhance EPON efficiency. Generally,
the main purpose of this article is to review EPON problems
and presents the up-to-date suggested solutions. Also to
indicate that further studies need to be carried out if a single
scheme that incorporates excellent bandwidth utilization with
effective QoS support and guaranteed fairness is required.
Keywords: EBON , DBA , ONU, OLT, QoS
1. INTRODUCTION
In the past decade, a great deal of effort have been done to
develop broadband access to the network in the area known as
the “last mile” . Based on its low cost passive components,
compatibility, scalability, and capability to deliver a bundled
service to subscribers, EPON became the attractive solution
to the bottleneck problem of the access network. Basically,
EPON consists of an optical line terminal (OLT) residing in
the central office, 1:N passive optical splitter/combiner, and
multiple optical network units (ONU) as shown in Fig.1. Each
ONU may serve one or more subscribers.
EPON carries the data encapsulated in ethernet frames,
which are compatible with IEEE 802.3 ethernet standards.
In down stream direction, EPON is a point-to-multipoint net-
work. The OLT broadcasts the data packets to all ONUs via
the common channel and each ONU extracts those packets
sent with its node ID (MAC address). In the upstream direc-
tion, EPON is a multipoint-to-point network where all ONUs
978-1-4244-1643-1/08/$25
c
° 2008
IEEE
Fig
. 1. Tree topology of EPON
network
have to share the common transmission media. Therefore, an
appropriate access mechanism is required to avoid collisions.
The channel data rate from ONU to OLT may not necessary
equal to the user access rate from user to an ONU. Usually,
the channel data rate is higher than user access rate.
Most of the reviewed studies carried out to improve EPON
efficiency in term of bandwidth utilization, minimizing packet
delay, increasing network throughput, and providing better
fairness and QoS support to traffic. Obviously, these efforts
have led to the wide adoption of the 1 Gb/s access network to
support the emerging HDTV and other broadband services.
Meanwhile, the increasing demands for band-sensitive appli-
cations with the need to support multi-dwelling units and next
generation wireless backhaul, prompt the IEEE 802.3 work-
ing group to intiate a study of the 10 Gb/s EPON architec-
tures [1]. Therefore, it is expected that this high speed access
network will be the best candidate replacement for the next
generation CATV networks.
2. BANDWIDTH UTILIZATION
In EPON, channel access is a crucial issue as all ONUs have
to share the upstream bandwidth. Thus, an efficient band-
width allocation algorithm is essential in providing a guaran-
teed network service to end-users. Generally, all bandwidth
allocation schemes can be classified according to the criteria
shown in Fig.2. Typically, bandwidth allocation can be ei-
ther static or dynamic. in static allocation [2], a fixed-size
000511
Fig
. 2. Classification of DBA
schemes
time-slot is allocated to each ONU regardless of the instanta-
neous bandwidth demand of this ONU. This approach is pre-
ferred method of channel sharing as it provides a simple and
cost-effective solution. However, it doesn’t have the ability
to adapt to the bursty nature of network traffic. This bursti-
ness results in some slots overflow even under very light load,
resulting in packets being delayed for several time-slot peri-
ods [3]. It is also true that some slots remain not filled com-
pletely even at high traffic load resulted in upstream band-
width being under-utilized.
On the other hand, in dynamic bandwidth allocation a vari-
able size time-slot is allocated dynamically to each ONU
based on its instantaneous bandwidth demand. Here, the ex-
cessive bandwidth of lightly loaded ONUs can be used to
meet the bandwidth demands of heavy loaded ONUs. Thus,
DBA algorithm can adapt to the bursty nature of the network
traffic and achieve better bandwidth utilization. For an effi-
cient scheme, the DBA algorithm should support the priority
queues in various traffic classes and fairness among ONUs.
EPONs rely on Multi-Point Control Protocol (MPCP) to
support time-slot allocation and arbitrate the access of multi-
ple ONUs sharing a common channel. This protocol which
is being developed by IEEE 802.03ah task force, supports
a wide range of DBA schemes. To accomplish the band-
width allocation process, MPCP protocol used two basic con-
trol messages: GATE and REPORT. The GATE messages is
used by the OLT to allocate (grant) transmission window to
an ONU. while, the REPORT message is used by an ONU
to report its bandwidth request. Upon receiving a REPORT
message, the OLT passes it to the DBA algorithm to calcu-
late the transmission schedule for the next transmission cy-
cle. After executing the DBA algorithm, the OLT sends a
GATE message to issue a bandwidth grant to the ONU. The
OLT also calculates the Round-Trip Time (RTT) to each ONU
from the reported messages. The REPORT and GATE mes-
sages should only contain ONU’s node identification and re-
quested/granted window size.
To ensure an efficient transmission, a polling protocol must
be used to schedule multiple REPORT/GATE transmission
cycles. In [4] Kramer et al. proposed the very well known
cyclic polling protocol, the Interleaved Polling with Adaptive
Cyclic Time (IPACT) algorithm. Similar to the hub polling,
IPACT requires the OLT to poll ONUs in a round-robin fash-
ion and dynamically granted a time-slot to each one. The
granted time-slot is not fixed as the case in static allocation,
but it is varying according to ONU’s buffer occupancy sta-
tus. IPACT uses the interleaved polling scheme where the
next ONU is polled before the transmission from the previous
one has arrived. Such kind of polling helps in minimizing up-
stream under-utilization due to walk-time (switch over time)
caused by the propagation delay. Moreover, IPACT does not
need to synchronize the ONUs to a common reference clock
or to perform a ranging process (arrange all the ONUs to ap-
pear equidistance from OLT) as it required by the traditional
TDMA schemes.
The entire scheduling of IPACT is located in the OLT at the
center office. Therefore, it is easy to change the scheduling at
run-time adaptively based on some network conditions with-
out any negotiation with the ONUs. Generally, IPACT is an
efficient scheme for bandwidth allocation but it is not taking
into account the class of service (CoS) in the allocation pro-
cess. Thus, it is not supporting QoS to traffic as it consider all
traffic as the same.
Typically, the DBA schemes can be classified into non-
prediction schemes such as the limited and gated schemes and
the prediction-oriented schemes such as the constant credit,
linear credit, and the elastic scheme. Refer to Fig. 2 for
this classification. In limited scheme, the OLT grants the re-
quested bandwidth but no more than a maximum transmis-
sion window (T W
max
).While in gated scheme, the ONU al-
ways granted as much as the bandwidth it has requested but
its buffer queue limited to Q bytes. Thus the ONU can not re-
quest more than Q bytes. It is clear that these schemes are not
considering the possible packets arriving just after the ONU
000512
sends its REPORT message. These packets
can only wait the
next grant cycle to be transmitted which causes a consider-
able delay. On the other side, the constant credit scheme adds
a constant credit to the requested bandwidth to accommodate
the early arriving bytes or those bytes that arrive just after
the ONU sends its REPORT message. In the linear credit
scheme, the size of this extra bytes is not constant but propor-
tional to the requested window . Finally, in the elastic scheme,
T W
max
has no limit and it is calculated in such a way that the
accumulated size of the last N grants (including the one be-
ing granted) dose not exceed N × TW
max
, where, N is the
number of ONUs. The basis of these schemes is that network
traffic usually has a certain degree of predictability [5].
3. AVERAGE PACKETS DELAY
As an example of non-prediction mechanism, DBA algorithm
with limited bandwidth allocation proposed in [6]. The trans-
mission window of each ONU is upper bounded by T W
max
which could be specified by the Service Level Agreement
(SLA). When the requested bandwidth is less than T W
max
,
the OLT grants the requested time slot. Otherwise, T W
max
is granted. Compared to the static scheduling, the scheme
shows a very little delay at low loads. However, the high net-
work loads result in high delay as specified by the reported
simulation results. The drawback of this scheme being in no
more than T W
max
can be granted to heavily loaded ONUs
no matter how long its buffer queue, which results in some
packets delayed at high load.
To address this problem, Assi et al., proposed in [7]
an allocation scheme that makes use of the total exces-
sive bandwidth of lightly loaded ONUs to fill out the band-
width demands of the heavily loaded ONUs. To explain
that, let M denotes the number of the lightly loaded ONUs
and BW
i
, BW
MIN
i
denote ONU
i
requested and mini-
mum guaranteed bandwidth respectively, then the total ex-
cessive bandwidth (BW
total
excess
=
P
M
i
(BW
MIN
i
− BW
i
)),
can be allocated fairly among the K heavily loaded ONUs
in proportion to the bandwidth demand of each one
(BW
excess
i
=
BW
total
excess
×BW
i
P
∀k∈K
BW
k
), where, BW
excess
i
is the ex-
cessi
ve bandwidth share of ONU
i
. To implement this allo-
cation scheme, the OLT must collect the REPORT messages
from all ONUs before it starts computations of bandwidth al-
location. Thus, a waiting time (idle period) equal to the sum
of DBA computation time (T
ct
) and RTT resulted where, the
upstream channel is unutilized as shown in Fig. 3.
To overcome the idle time, the authors used an early al-
location mechanism which schedules a lightly loaded ONUs
immediately, while the heavy loaded ones have to wait un-
til the OLT finished collecting the REPORT messages.This
approach improve the bandwidth utilization and network
throughput under light and medium loads but makes no dif-
ference in high load as the number of heavily loaded ONUs
increases.
To further improve the idea of early allocation, Zheng in
[2], suggested an algorithm that saves the ending time of the
last time slot (T
end
) in a tracker to be used in the following
ONU scheduling. Then if the requested bandwidth is smaller
than the minimum guaranteed bandwidth, OLT grants this
bandwidth immediately. Otherwise, the OLT will defer the
transmission grant if the tracker value is smaller than the idle
time and the time next REPORT message arrives is earlier
than (T
end
− RR T /2). The reader can refer to [2] for more
details about this scheme. The reported simulation results
show an efficient utilization of the idle time. However, it is
not eliminated totally.
Fig
. 3. Illustration of waiting time
in bandwidth allocation
process
In [8], a new scheme called Per-Slot DBA (PSDBA) pro-
posed to eliminate completely the idle time. In this algorithm,
if an ONU sends a REPORT message at time say T
1
and the
requested bandwidth is greater the the minimum guaranteed
one, the OLT will not sends the grant at T
1
but at T
2
where
T
2
is the latest time for OLT to send the GATE message with-
out causing any bandwidth waste. During the time interval
T
2
− T
1
there might be some REPORT messages arrive with
bandwidth request less than the minimum guaranteed band-
width, then the excessive bandwidth can be assigned to the
heavy loaded one. The scheme gives considerable improve-
ment in average packets delay and network throughput as in-
dicated by the reported simulation results.
In [9] authors proposed IPACT with Smallest Available Re-
port First (SARF) scheme. OLT In this scheme, updates a
list with a queue length of each ONU every time it receives
a REPORT message and marks the not scheduled ONU as
“pending”. The SARF algorithm arranges all the pending
ONUs in ascending order according to their queue length and
the one with the smallest length served first. But the OLT
will defer the GATE message to the latest possible time that
causes no bandwidth waste. This process continues until the
entire pending ONUs served. By this way, the average cy-
cle time can be reduced and consequently reducing average
packets delay. However, that is true under some load values
but makes no difference at light loads since the channel al-
ready under-utilized. Also granting more small transmission
widows at low loads, results in more bandwidth waste due to
guard times and REPORT messages.
The OLT in the algorithm Dynamic Polling Order Arrange-
000513
ment (DPOA) proposed in [10] also arranges
the ONUs ac-
cording to the queue length but in descending order. This
scheme shows a better performance than the basic IPACT un-
der low and medium load, but the polling order makes no dif-
ference at high load as the number of heavily loaded ONU
increases.
To accommodate the early arriving packets,the authors in
[11] proposed a DBA with Multiple services (DBAM) that
executes a linear prediction scheme. The calculation of lin-
ear estimated credit for each ONU depends on the ratio of
the ONU waiting time to the entire polling cycle time. The
scheme enhances the multiple services among ONUs with di-
verse bandwidth request and results in packet delay improve-
ment of all traffic classes as the reported simulation results
can show. However, the inaccuracy of the prediction under
non-uniform load may deteriorate the performance of this al-
gorithm.
Another prediction based scheme called Early DBA with
Prediction-based Fair Excessive Bandwidth Reallocation
(PFEBER) proposed in [12]. The algorithm used a historical
traffic to calculate a variance for each ONU and obtain a list
termed unstable degree list from those ONUs of higher vari-
ance. The scheme lowers the prediction inaccuracy by assign-
ing different weights to the linear credit depending wether the
ONU belongs to the unstable degree list or not. The scheme
results in more stability at non-uniform loads. However, an
optimal number of REPORTS messages necessary to obtain
the high variance ONUs list must be considered for good per-
formance.
4. QOS SUPPORTING SCHEMES
With the convergence of multiple applications over one
shared link , the need for EPON service differentiation and
quality of service has become a major issue. To support these
applications, the traffic is separated into a number of classes
with different priorities [13]. The high priority class is the
Expedited forward (EF) traffic, which is delay sensitive and
requires bandwidth guarantees. The medium priority class
is the Assured Forward (AF) traffic, which is not delay sensi-
tive but requires bandwidth guarantees. The low priority class
is the Best Effort (BE) traffic, which is neither delay sensi-
tive nor requires a bandwidth guarantees. To support QoS,
DBA algorithm has to allocate the upstream bandwidth fairly
among these different classes.
Basically, there are two types of scheduling that support
QoS: inter-ONU and intra-ONU. The first one arbitrates the
transmission of the different ONU while the second one arbi-
trates the transmission of the different priority queues in each
ONU. Some DBA scheme designed so that the OLT performs
both the inter-ONU and the intra-ONU scheduling (central-
ized EPON). However, other DBA schemes allow the OLT
to perform only the inter-ONU while the ONU performs the
intra-ONU scheduling as the case in the hierarchical EPON.
Moreover, there are two types of the intra-ONU scheduling:
strict priority and non-strict priority. In strict priority schedul-
ing, a lower priority queue is scheduled only if all queues with
higher priorities are empty [14] which may result in a starva-
tion for low priority traffic. On the other side, the non-strict
priority, scheduling allow to transmit first the reported pack-
ets regardless of their priority within the allocated transmis-
sion window. But the priority of each class determines its
transmission order.
In [15], a DBA algorithm supporting multiple priority
queues and fairness proposed. Three priority levels used to
classify services. Accordingly, the total available bandwidth
is divided into fixed bandwidth for high priority traffic, as-
sured bandwidth for medium priority traffic, and best effort
bandwidth for low priority traffics. In each transmission cy-
cle a fixed portion of the available bandwidth allocated to the
higher priority traffic regardless of whether or not there are
frames to send. The remaining part of the available band-
width is weighted depending on the ratio of the total band-
width requested for the medium traffic to the sum of the total
bandwidth requested for medium and low level priority. The
weighting mechanism intended to guarantee that the lower
priority traffic will have some access to the upstream chan-
nel. The reported simulation results show a good level of QoS
support for the different classes but fairness is not guaranteed
for low level priority class.
Unlike the conventional slot-size based DBA scheme
where traffic of all classes from single ONU is mixed together,
the scheme in [16], divides the transmission cycle (frame) to
three parts and allocates each part to different class of service.
Consequently, the frame will have a fixed part of N time-slots
for the high priority traffic corresponding to N ONUs then
quasi-dynamic part with number of time-slot equals to the
number of ONUs with medium priority traffic and finally a
dynamic one time-slot part shared by all the ONUs for low
priority traffic. Under light loads all the ONUs sharing the
dynamic part may have a transmission grant. But at heavy
loads the OLT may dedicate the whole dynamic part to a sin-
gle ONU in one transmission cycle and to another ONU in
other transmission cycle. And in each N frame ONU may be
authorized to transmit the low priority traffic. Although, this
scheme meets QoS requirements and save overhead problem
by skipping frames for low priority traffic, the packets delay
of low priority service class is significantly prolonged.
The hierarchical DBA algorithms proposed in [17] [18],
perform the inter-ONU and intra-ONU scheduling. The pro-
posed schemes are composed of a low-level scheduler in the
ONU and a high-level scheduler in the OLT. The high-level
scheduler allocates bandwidth to the ONUs, while the low-
level scheduler distributes the allocated bandwidth among the
different priorities queues. There is no need to send a separate
GATE messages for each queue (as the case in the OLT-based
schemes), sine the OLT grants an aggregate bandwidth per
ONU. This will help in saving the overhead problem as all
000514
of the queues in one ONU are serv
ed consecutively with no
guard times between their transmissions.
The OLT in [17], allocates bandwidth to the ONUs in
proportion to weights associated with their class and queue
length, while the ONU preferentially allocates its bandwidth
to queue with static priority order.
In [18], the Integrated Two Classes DBA (ITCDBA) algo-
rithm is located in both OLT and ONU. The OLT scheduler
allocates bandwidth according to the requested size of ONUs
bounded by upper limit to avoid the ONU with lower priority
requests too much bandwidth. Then ONU scheduler allocates
the granted bandwidth to each queue according to its priority
upper bounded by queue maximum requested bandwidth to
avoid the prolonged dealy of lower priority queue.
5. FAIRNESS SUPPORTING SCHEMES
To address the fairness problem in EPON, the authors in [19]
[20], proposed new DBA schemes that maintain the fairness
in DBA operation. Basically, an ONU has the right to share
the available resources according to its pre-assigned weight
defined by the service level agreement before its bandwidth
request is fully satisfied. Thus, any ONU can not affect oth-
ers by overloading the network. However, it can be easily
seen that most of the proposed DBA are focusing in satis-
fying ONU’s bandwidth request over maintaining better fair-
ness. In the other words, the instant bandwidth request of each
ONU may effect the fairness policy of the OLT’s DBA deci-
sion. To address this issue, a Weight-based DBA (WDBA)
that constantly maintains a maximum value of fairness index
(f) through the DBA operation is proposed in [19]. The au-
thors suggested that a fairness index can be maintained at the
highest value (guaranteed fairness or f=1) by proper assign-
ing (BW
excess
i
) which is the ON U
i
share of the total ex-
cessive bandwidth saved by under-load ONUs (BW
excess
total
).
The initial value of BW
excess
i
defines an initial threshold
for the corresponding ONU based on the weighted Max-Min
fairness principle. Then the algorithm iteratively computes
(BW
excess
i
=
w
i
P
n
i=1
w
i
× B
W
excess
total
) to find the optimum
value of B
W
excess
i
. Where w
i
is ONU
i
weight.
The scheme grants as much as requested bandwidth to what
is called satisfiable ONUs that are requesting no more re-
source than their corresponding threshold and the threshold
value updated each time a satisfiable ONU granted band-
width. to reach a final decision, the OLT has to repeatedly
compare the requested bandwidth of the remaining ONUs
with continuously updated threshold.
Similar procedure applied in Fairness-Guaranteed (FG)
scheme proposed in [20] except that the authors suggested
the use of global dynamic variable α and make α ∗ b
min
k
the
threshold value of the granted bandwidth to ON U
k
. Since the
requested bandwidth value b
request
k
varies in different cycles,
α should be adjusted iteratively in every cycle to make B
g rant
approach B
UP
, where B
grant
is the total granted bandwidth
and B
UP
is the size of total upstream transmission.
The reported simulation results of these two iterative
schemes show that the fairness can be maintained effectively
when the bandwidth allocation based on updating threshold
of the maximum granted bandwidth rather than the sole de-
pendence on priority levels.
In [13], authors proposed a nother fairness support algo-
rithm called Fast Class-of-service Oriented Packet Schedul-
ing (FCOPS). The algorithm employs a credit pooling tech-
nique to partition the network bandwidth among different
classes of service. The scheme maintains N credit pools for N
class of service and another M credit pools for M ONU. The
first pools, used to enforce a long term average rate of each
class traffic transmitted from all ONUs. While the second
pools used to control the bandwidth used by each ONU. A
sufficient credits in both pools must exist before OLT granted
a transmission to a request from any ONU for a given class
of service. A weighting mechanism enforced by using a
Weighted Round Robin (WRR) arbiter placed before each
CoS credit pool is used in this scheme to ensure that the ONUs
of certain CoS will receive a fair share of the CoS credit pool
based on their weights.
To eliminate the idle time, the investigators, suggested that
the beginning of each cycle to be coincide with the arrival
of a transmission slot from ONU1. This arrival will trigger
the OLT to start processing grant for this ONU and marks the
beginning of new transmission cycle. Each ONU request will
be processed instantaneously in such a way that while OLT
computes grants for one ONU, the upstream channel can be
used by another ONU. Beside efficient bandwidth utilization,
the scheme can maintain a high degree of fairness among all
end-users.
6. CONCLUSION
From the reviewed articles, it can be easily figure out that
great efforts heve been spent in the last few years to increase
EPON efficiency in utilizing the upstream channel and en-
hancing QoS support as well as maintaining the fairness in
DBA operation. In most cases, the new schemes show great
performances in one or two of EPON meterics and fair per-
formance in others. Also, their performance may change from
certain load values to another. It is expected that most of the
undergoing and future researches will further scale up EPON
speed and efficiency as a result of increasing demands for
more bandwidth-sensitive applications .
7. REFERENCES
[1] Glen Kramer, “What is Next for Ethernet PON,” Int.
Conference on Optical Internet, COIN 2006, pp. 49–54,
July 2006.
000515
[2] J. Zheng, “Efficient Bandwidth Allocation Algorithm
for Ethernet
Passive Optical Networks,” IEE Proc. on
Commun., vol. 153, no. 3, pp. 464–468, June 2006.
[3] Abdallah Shami, Xiaofeng Bai, Chadi M. Assi, and
Nasir Ghani, “Jitter Performance in Ethernet Passive
Optical Networks,” Jornal of Lightwave technology, vol.
23, no. 4, pp. 1745–1753, April 2005.
[4] Glen Kramer, Biswanath Mukhherjee, and Gerry Pe-
savento, “Interleaved Polling with Adaptive Cycle Time
(IPACT): A Dynamic Bandwidth Distribution Scheme
in an Optical Access Network,” Tech. Representation
CSE-2001-4, Dept. of Comp. Sci., UC, August 2001.
[5] Jun Zheng and Hussain T. Muftah, “Media Access
Control for Ethernet Passive Optical Networks: An
Overview,” IEEE Commun Magazine, vol. 43, no. 2,
pp. 145–150, February 2005.
[6] Glen Kramer and Biswanath Mukhherjee, “IPACT: A
Dynamic Protocol for an Ethernet PON (EPON),” IEEE
Commun Magazine, vol. 40, no. 2, pp. 74–80, February
2002.
[7] Chadi M. Assi, Yinghua Ye, Sudhir Dixit, and Mo-
hamed A. Ali, “Dynamic Bandwidth Allocation for
Quality-of-Service Over Ethernet PONs),” IEEE Jornal
On Selected Area In commun, vol. 21, no. 9, pp. 1467–
1477, November 2003.
[8] T. Fan and H. T. Mouftah, “A QoS-based Dynamic
Bandwidth Allocation algorithm for EPONs,” 23rd
Biennial Symposium. on Commun, pp. 140–143, June
2006.
[9] Swapnil Bhatia and Radim Bartos, “IPACT with Small-
est Available Report First: A New DBA Algorithm for
EPON,” in IEEE Int. Conference on Commun., ICC’07,
June 2007, pp. 2168–2173.
[10] M. Ma, L. Liu, and T. H. cheng, “Adaptive Scheduling
for differentiated Services in An Ethernet Passive Net-
work),” Jornal of OPtical Networking, vol. 4, no. 10,
pp. 661–670, October 2005.
[11] Yuanqiu Luo and Nirwan Ansari, “Bandwidth Alloca-
tion for Multiservice Access on EPON),” IEEE Optical
Commun, vol. 43, no. 2, pp. S16–S21, February 2005.
[12] I. Shyan Hwang, Zen-Der Shyu, Liang-Yu Ke, and
Chun-Che Chang, “A Novel DBA Mechanism with
Prediction-based Fair Excessive Bandwidth Realloca-
tion Scheme in EPON,” IEEE International Conference
on Networking ICN’07, April 2007.
[13] Hassan Naser and Hussain T. Muftah, “A Fast Class-of-
Service Oriented Packet Scheduling Scheme for EPON
Access Networks),” IEEE Commun. Letters, vol. 10, no.
5, pp. 396–398, May 2006.
[14] Ahmad R. Dhaini, Chadi M. Assi, and Nasir Ghani,
“Adaptive Fairness Through intra-ONU scheduling for
Ethernet Passive Optical Networks,” in IEEE Int. Con-
ference on Commun., 2006, June 2006, vol. 6, pp. 2687–
2692.
[15] Seong-Ho, Jin-Man Kim, and Jong-Wook Jang, “A new
dba algorithm supporting priority queues and fairness
for epon,” in Proceeding of SPIE, 2005, vol. 5626, pp.
1116–1122.
[16] Qing Wang and Yan He, “Analysis and improvement of
dynamic bandwidth allocation in epon,” in Proceeding
of SPIE, 2005, vol. 5626.
[17] Key-Hyun, Kyeong-Eun Han, and Young-Chon Kim,
“Herarchical Dynamic Bandwidth Allocation Algo-
rithm for Multimedia Service over Ethernet PONs),”
ETRI Jornal, vol. 26, no. 4, pp. 321–331, August 2004.
[18] Hong Chen, Depeng Jin, and Lieguang Zeng, “A novel
dynamic bandwidth allocation scheme in epons,” in Pro-
ceeding of SPIE, 2005, vol. 6022, pp. 60220U–1.
[19] Xiaofeng Bai, Abdallah Shami, and Chadi Assi, “On the
Fairness of Dynamic Bandwidth Allocation Schemes in
Ethernet Passive Optical Networks),” Computer Com-
munications, vol. 29, no. 11, pp. 2123–2135, July 2006.
[20] Wu Liu, Deming Liu, Guangxi Zhu, and Baomin Hu,
“Fairness guaranteed scheme: A novel algorithm for dy-
namic bandwidth allocation in epon,” in Proceeding of
SPIE, 2006, vol. 63542U, p. p63542U.
000516
