HYRA: An Efficient Hybrid Reporting Method
for XG-PON Upstream Resource Allocation
Panagiotis Sarigiannidis
1
, Georgios Papadimitriou
2
, Petros Nicopolitidis
2
, Emmanouel Varvarigos
3
and Konstantinos Yiannopoulos
4
1
Department of Informatics and Telecommunications Engineering, University of Western Macedonia, Kozani, Greece
2
Department of Informatics, Aristotle University of Thessaloniki, Thessaloniki, Greece
3
Computer Technology Institute and Press ”Diophantus” N. Kazantzaki, University of Patras, Campus, Rio 26500, Greece
4
Department of Informatics and Telecommunications, University of Peloponnese, Terma Karaiskaki, Tripoli 22100, Greece
Keywords:
Dynamic Bandwidth Allocation, Learning Automata, Passive Optical Networks, XG-PON.
Abstract:
The dynamic bandwidth allocation (DBA) process in the modern passive optical networks (PONs) is crucial
since it greatly influences the whole network performance. Recently, the latest new generation PON (NG-
PON) standard, known as 10-gigabit-capable passive optical network (XG-PON), standardized by the inter-
national telecommunication union telecommunication standardization sector (ITU-T), emerges as one of the
most efficient access networking framework to cope with the demanding needs of the fiber to the x (FTTX)
paradigm, where x stands for home (FTTH), bulding (FTTB), or curve (FTTC). Motivated by the fact that
the ITU-T specifications leave the bandwidth allocation process open for development by both industry and
academia, we propose a novel DBA scheme for effectively delivering data in the upstream direction. Our
idea is based on a subtle suggestion induced by the XG-PON specifications; each developed DBA method
should combine both status reporting (SR) and traffic monitoring (TM) techniques. This means that a XG-
PON framework should be cognitive enough in order to be able either to request bandwidth reporting from the
connected users or estimate users’ bandwidth demands or both. In this article we cover this gap by propos-
ing a robust learning from experience method by utilizing a powerful yet simple tool, the learning automata
(LAs). By combining SR and TM methods, the proposed hybrid scheme, called hybrid reporting allocation
(HYRA), is capable of taking efficient decisions on deciding when SR or TM method should be employed so
as to maximize the efficacy of the bandwidth allocation process. Simulation results reveal the superiority of
our scheme in terms of average packet delay offering up to 33% improvement.
1 INTRODUCTION
The penetration of optical technology in the access
network domain is rapidly gaining ground. Passive
optical networks (PONs) are currently the most ef-
fective player of optical technology in the last mile
playground, since they offer a complete, efficient,
all-optical, and cost-effective solution to cope with
modern, demanding, and diverse services and appli-
cations. In essence, PONs interconnects users to the
backbone interface by means of optical fiber provid-
ing a full-optical path without the need of optical-to-
electrical conversion. This all-optical path creates a
transparent point to multi-point interconnection, of-
fering high data rates for both upstream and down-
stream direction. Nonetheless, the great potential of
PONs in terms of huge bandwidth provisioning has
not yet fully utilized due to diverse of users behaviors
and requirements.
In order to meet users requirements, quality of
service (QoS) guarantees should be ensured in PON
operation. However, static bandwidth allocations in-
duce low channel utilization and therefore limited ser-
vice provisioning. Having this in mind, the telecom-
munication standardization sector of the international
telecommunication union (ITU-T) dictates the usage
of dynamic bandwidth allocation (DBA) schemes. By
applying dynamic bandwidthdistribution methods the
bursty user traffic demands are effectively addressed.
The design of intelligent DBA algorithms advances
in a crucial challenge, especially in the upstream di-
rection where multiple users traffic streams have to
share common optical paths without negativelyaffect-
ing the network performance or violating QoS agree-
5
Sarigiannidis P., Papadimitriou G., Nicopolitidis P., Varvarigos E. and Yiannopoulos K..
HYRA: An Efficient Hybrid Reporting Method for XG-PON Upstream Resource Allocation.
DOI: 10.5220/0005048200050014
In Proceedings of the 5th International Conference on Optical Communication Systems (OPTICS-2014), pages 5-14
ISBN: 978-989-758-044-4
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
ments. By incorporating an effective DBA algorithm,
and therefore achieving a good network performance,
more subscribers could potentially join the network,
thus decreasing the network operations costs, and
even more standards could be reached on providing
cutting-edge applications to users.
The 10-gigabit-capable passive optical network
(XG-PON) is one of the most promising standards of
the next-generation PONs (NG-PONs). It comes with
several powerful assets such as enhanced cryptog-
raphy, compliance with older standards, higher data
rates for both directions, clear QoS-aware bandwidth
allocation processes, and energy-efficient support.
In this work, an adaptive, learning from experi-
ence, robust resource allocation scheme is proposed
in order to alleviate the impact of time-varying traf-
fic changes in the XG-PON systems. The so-called
hybrid reporting allocation (HYRA) utilizes a com-
bination of different, heterogeneous, yet allowed by
the standard, allocation policies in order to provide a
fully standard-compliant, efficient allocation method.
HYRA exploits the capabilities of traffic monitoring
technique so as to effectively re-distributes the sur-
plus bandwidth gained by isolating the underutilized
ONUs. Demanding users are favored and therefore
more bandwidth is allocated to active users without
overshadowing the network operation. The suggested
scheme seems to be capable of adapting to various
network changes offering thus notable improvements,
in terms of upstream packet delay, as indicated by sev-
eral simulation results based on real multimedia traf-
fic traces.
The remainder of the paper is organized as fol-
lows. Section 2 introduces several features of the un-
derlying allocation policies in order to providea better
understanding of the XG-PON sub-layers. In Section
3 existing research efforts towards resource allocation
in XG-PON are outlined. A detailed description of the
proposed scheme is provided in Section 4. Section 5
illustrates the obtained results, followed by detailed
reports. Finally, conclusions are given in Section 6.
2 BACKGROUND
The XG-PON framework defines a point-to-
multipoint optical access infrastructure providing
(nominal) 10 Gbps data rate in at least one direc-
tion. One of its most determinant layers is the
XG-PON transmission convergence (XGTC) layer,
in which the functional protocols and procedures
including the way of performing resource allocation
and provisioning QoS between the upper layers
and the physical layer, are thoroughly described.
According to the standard specifications, the XGTC
layer is structured in three sub-layers, namely the
service adaptation, the framing, and the physical
(PHY) sub-layer. The service adaptation sub-layer
performs service data unit (SDU) encapsulation and
multiplexing and creates XG-PON encapsulation
method (XGEM) frames. The framing sub-layer
receives the constructed XGEM frame and forms
the downstream XGTC frame. The downstream
frame encloses multiple XGTC payloads which are
distinguished based on their Alloc-ID. The Alloc-ID
field identifies the recipient of the allocation within
the ONU. Lastly, the PHY sub-layer applies bit
error correction algorithms, it performs scrambling
to the content, and it synchronizes the frames. It
is worth mentioning that the XGTC layer holds for
both upstream and downstream directions, hence
the aforementioned procedures reversely hold in the
upstream direction.
In the downstream direction the XGTC layer
is responsible of receiving SDUs from the upper
layers and producing an uninterrupted bitstream at
the nominal interface, which in the downstream di-
rection supports 9.95328 Gbps divided into 125 µsec
downstream frames. The duration of the downstream
frame, in accordance with the given downstream
rate, corresponds to 155520 Bytes. However, this
size includes coding and control information. The
physical synchronization block field comes first
in downstream flow (PSBd), which includes a
synchronization bitstream, the PON identification
number, counters, and other control information.
An important control field, known as BWmap,
which is associated with the bandwidth allocation
process, is enclosed in the XGTC header. It is used
by the OLT to inform the ONUs about the granted
transmission opportunities; it defines the start time
of the transmission opportunity and the grant size
per Alloc-ID for each ONU. In essence, the OLT
continuously broadcasts data to ONUs, including
requested data delivery, messages, and bandwidth
allocation information.
The XG-PON standard implicitly assumes syn-
chronization between downstream and upstream
frames. This means that the i-th downstream frame
is associated to the i-th upstream frame, even though
the i-th upstream frame could reach the OLT late
due to propagation time. Nonetheless, the allocation
information included in the i-th downstream frame
corresponds to the i-th upstream frame. To be syn-
chronized, both frames have the same length, thus the
duration of the upstream frame is 125 µsec. However,
it accounts for 38880 Bytes due to the fact that the
(nominal) upstream data rate is 2.48832 Gbps. The
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
6
PSB of the upstream frame (PSBu) contains the
preamble and the delimiter fields. Then, the XGTC
burst follows, which includes a control field in the
front (XGTC header) and a trailer (XGTC trailer).
The existence of the inner header, which is called
dynamic bandwidth report (DBRu), determines the
adopted resource allocation method. Two options
are allowed by the standard, namely a) the status
reporting (SR) method, in which each allocation
encloses the DBRu header and reports the OLT its
buffer status, and b) the traffic monitoring (TR), in
which the OLT monitors the idle upstream frames
to perceive the bandwidth pattern of each Alloc-ID.
According to the specifications, the XG-PON OLT
should support both techniques in a separate way or
even combined. More information about XG-PON
could be found in (Effenberger, 2010).
The resource allocation obeys to specific down-
stream and upstream principles. Each ONU receives
a guaranteed bandwidth portion including three al-
location parameters: a) the fixed bandwidth, R
f
, is
given regardless of ONU’s traffic demands, b) the as-
sured bandwidth, R
a
, is given as long as the ONU
has unsatisfied traffic demands, and c) the maximum
bandwidth, R
m
, represents the upper limit on the to-
tal (guaranteed) bandwidth. Beyond the guaranteed
bandwidth, the surplus bandwidth is shared to ONUs
still having unmet bandwidth requests.
3 RELATED WORK
In general, the development of novel, efficient, and
effective DBA schemes in PONs have received a lot
of attention. The authors in (Kanonakis and Tomkos,
2009) introduced the offset-based scheduling with
flexible intervals concept for gigabit PONs (GPONs).
The rationale behind this concept stands on apply-
ing flexible scheduling intervals. In essence, the au-
thors proposed lower scheduling intervals regarding
the polling policy between the ONUs and the OLT;
however they keep the SR method as the main report-
ing method. The scheme presents improvements in
terms of network throughput and average packet de-
lay. In (Han, 2014) a high-utilization scheme was
presented. A common available byte counter and a
common down counter for multiple queues of a ser-
vice class are utilized in order to effectively share the
surplus bandwidth to demanding users. Our previous
efforts in (Sarigiannidis et al., 2013b) deals with the
fairness provisioning, by intending to resolve unequal
resource allocation in the downstream data delivery.
In particular, a fair bandwidth assignment scheme is
devised and evaluated. The Max-Min fairness con-
cept is applied in order to ensure a fair downstream
broadcast schedule between multiple ONUs.
Beyond bandwidth allocation development, in
(Yoshimoto et al., 2013), flexible speed upgrades
for NG-PONs, such as the XG-PON, are discussed.
The authors investigate performance and cost issues
while they face reach extensions matters. Efforts in
(Mullerova et al., 2012) focus on the usage of specific
wavelength blocking filters so as to restrain the unde-
sirable interference when GPONs and XG-PONs co-
exist. In (Lee et al., 2013) an ONU fast management
that reduces the time required to update a remotely-
located user terminal’s software was inaugurated. Fi-
nally, features of the first XG-PON testbed could be
found in (Jain et al., 2011).
By examining the efforts presented in the liter-
ature we can easily infer that a) the research field
of providing effective bandwidth allocation in XG-
PON remains open and challenging and b) all DBA
schemes presented in literature assume the SR method
as the core scheduling policy. In this article, we step
beyond the pure usage of the SR method by inaugurat-
ing a hybrid reporting method that takes into account
the existing bandwidth pattern revealed by function-
ing the TR method.
4 HYRA
The proposed hybrid reporting allocation scheme is
described in detail in this Section.
4.1 Problem Definition
The transmission of an idle XGEM by an ONU sig-
nals either upstream traffic absence or transmission
restrictions, e.g., fragmentation violations. In any
case, the OLT perceives that an ONU has no traf-
fic to send when an idle XGEM frame is received
by this ONU. This phenomenon could induce band-
width wastage if the OLT neglects the reception of
idle XGEM(s) by a single or more ONUs. For ex-
ample, the OLT is responsible of distributing a min-
imum bandwidth fraction to all ONUs in accordance
to ITU-T specifications. As in Section 2 mentioned,
a fixed bandwidth rate, equal to R
f
is given to all
ONUs independently of their bandwidth reports. To
this end, bandwidth losses are caused when the OLT
shares bandwidth to ONUs that consecutively or spo-
radically return idle XGEM frames back to the OLT.
Accordingly, bandwidth opportunities that remain un-
derutilized overshadow the packet delivery quality of
demanding ONUs in terms of data packet delay. This
HYRA:AnEfficientHybridReportingMethodforXG-PONUpstreamResourceAllocation
7
problem could be addressed by applying a more so-
phisticated method to effectively offer bandwidth to
the connected ONUs. Nevertheless, the ITU-T speci-
fications clearly allows the usage of monitoring tech-
niques in order to deal with underutilized ONUs. This
gap is efficiently faced in this article by proposing a
robust, hybrid scheduling policy that effectively com-
bines both reporting methods. To this purpose, a sim-
ple yet powerful adaptive mechanism is employed;
the learning automata (LAs).
4.2 Learning Automata
In general, many networks operate in environments
with unknown and time-varying features. In access
networks especially the time-varying parameters are
often quite radical and might dramatically affect the
network performance. Examples of such parameters
are the burstiness, the traffic heterogeneity, and the
user traffic activity. The changing nature of such char-
acteristics entails careful and sophisticated design of
efficient networking protocols and as a consequence,
adaptivity arises as one of the most important proper-
ties of such protocols.
In order to meet the aforementioned requirements,
LAs are adopted as an enhancing, adaptive mecha-
nism to the OLT’s decision process. LAs are artificial
intelligence tools that can provide adaptation to sys-
tems operating in changing and/or unknown environ-
ments (Misra et al., 2013). A LA is a finite state ma-
chine that interacts with a stochastic environment and
tries to learn the optimal action offered by the envi-
ronment via a learning process. In this work, a LA is
encompassed in the OLT to interact with the environ-
ment. The environment included the ONUs, the net-
work characteristics, such as the ONUs’ requests, and
the network configuration, e.g., bandwidth allocation
rules and restrictions. Being the thinking tank, the
OLT, enhanced with the LA, exchanges information
with the environment. For example, the OLT decides
about the schedule and informs the ONUs about it. On
the contrary, each ONU reports to the OLT by send-
ing bandwidth requests with regard to users needs. In
accordance to the bandwidth allocation specifications,
the OLT is able to make specific decisions. The set of
possible decisions an OLT could make constitutes the
action poll of the LA. Moreover, a feedback is gener-
ated each time the OLT, by the aim of the LA, makes
a decision. The feedback is originated by the environ-
ment, e.g., an idle XGEM frame is a feedback, and the
LA receives it, updates the significance of each action
in its pool, and prepares the next action. The process
is repeated and finally leads the LA to select the best
possible action from the pool of possible ones.
LAs are widely used in networking involving all
layers. In the physical layer, LAs provide adaptivebe-
havior towards channel characteristics (Nicopolitidis
et al., 2011), framing determination (Sarigiannidis
et al., 2011), and signal processing (Huang, 2008). In
the data link layer, medium access decisions are gov-
erned by LAs towards transmission power determi-
nation (Joshi et al., 2008), packet collision avoidance
(Eraghi et al., 2011), and spectrum sensing in cogni-
tive networks (Sarigiannidis et al., 2013a). Routing
enhancement is provided by LAs in (Economides and
Silvester, 1988), while efforts in (Navid, 2010) sup-
port energy-aware adaptive LAs-based protocols. In
transport layer, the congestion control mechanism is
empowered with a LA to adapt to network changes
(Venkata Ramana et al., 2005). Lastly, LAs play a
critical role to provide reliable data transmission ve-
hicular ad hoc networks (VANETs) (Kumar et al.,
2013) considering the application layer, while the au-
thors in (Bozicevic et al., 2004) demonstrated how
modern applications can become more efficient using
LAs.
The rationale behind the adoption of LAs as the
core learning from experience mechanism in this
work can be summarized as follows:
1. Low complexity; the calculations the LA engages
are linear and quite simple.
2. Rapid convergence;the convergencespeed is con-
trolled by specific parameters, hence LA are ca-
pable of rapidly converging under normal condi-
tions.
3. Flexibility; a LA is able to cooperate with any
thinking module, so the enhancement of the OLT
with a LA is a straightforward and cost-effective
task.
4. Efficiency; LAs are proved to be efficient and
widely utilized due their ability to support accu-
rate estimations.
4.3 Formulation
Figure 1 illustrates how HYRA operates to provide
an efficient solution to the upstream bandwidth allo-
cation. This state diagram holds in the OLT side. It
determines how the OLT behaves depending on each
ONU reaction/feedback. The upper rectangular en-
closes the SR operation, denoted by the Status Re-
porting state, where the OLT includes an ONU in
the upstream bandwidth allocation process as long as
the ONU reports an active XGEM. Upon the recep-
tion of an empty XGEM by an ONU, the OLT as-
sumes that this ONU experiences a period of inactiv-
ity, hence it isolates it from bandwidth distribution.
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
8
6WDWXV
5HSRUWLQJ
$FWLYH
;*(0
7UDIILF
0RQLWRULQJ
1R
%DQGZLGWK
(
P
S
W
\
;
*
(
0
8
S
R
Q
W
K
H
&
R
P
S
O
H
W
L
R
Q
R
I
W
K
H
,
V
R
OD
W
L
R
Q
3
H
U
L
R
G
,VRODWLRQIRUD3HULRG
'HWHUPLQHGE\WKH/$
653ROLF\
703ROLF\
Figure 1: HYRA structure.
Thus, this ONU enters to TM session, where neither
upstream opportunities are included to the forthcom-
ing downstream frame(s) to this ONU nor upstream
bursts are accepted from this ONU. In other words,
the assigned guaranteed bandwidth destined to this
ONU is shared among the active ONUs. In such a
way, upstream bandwidth is saved in favor of the de-
manding users. Moreover, active users can effectively
experience even high degrees of QoS. Yet, the con-
nectivity ratio of the PON can be expanded including
more users and ONUs. The isolation period is deter-
mined by the automaton.
The duration of the isolation period is determined
by the LA. Setting an ONU in TM status, the OLT
makes a decision about the isolation interval based on
the past traffic activity of this ONU. To this end, the
automaton maintains a pool of actions where each ac-
tion denotes a specific isolation interval. Of course,
due to the periodicity of the downstreamtransmission,
each action is expressed as a multiple of 125 µsec.
Furthermore, according to the ITU-T there is a maxi-
mum limitation on setting an ONU in idle condition,
that is 50 msec. Thus, a maximum of 50000/125 =
400 possible isolation periods are defined. Bearing
these in mind, the action pool is defined as follows:
A = a
0
,a
1
,a
2
,...,a
400
(1)
The state a
0
means no isolation, i.e., the ONU re-
turns to SR status. The automaton maintains a proba-
bility for each action. This probability is determined
based on the past traffic pattern of each ONU and it
dictates how possible is the state to be the optimal
one. The (action) probability vector at downstream
frame f is given below:
P
i
( f) = p
i
0
( f), p
i
1
( f), p
i
2
( f),..., p
i
400
( f) (2)
where i stands for the ONU identity. Obviously, it
holds that:
400
∑
j=0
p
i
j
( f) = 1 ∀ i (3)
Initially, all probabilities are equally set:
p
i
j
( f) = 1/401, 0 ≤ j ≤ 400, ∀ i (4)
The feedback received by each ONU is translated
based on Eq. (1). When the OLT receives an empty
XGEM from an ONU, it records the elapsed time
passed this ONU remained idle, i.e., between this in-
stance and the reception of an active XGEM. This
time period signals the LA’s feedback. If T
1
stands for
the time instance the OLT received an empty XGEM
and T
2
denotes the time an active XGEM received by
the OLT, the time period is transformed to an action:
a
k
= ⌊
T2− T1
125
⌋ 0 ≤ k ≤ 401 (5)
In essence, the automaton takes this time period and
associates it with an action from the pool. For ex-
ample, let T
1
= 1200 and T
2
= 1633. The associated
action is a
4
= 3· 125µsec.
The adaptivity of the present model lies in the
incorporated learning mechanism, which enables the
adjustment of the action probability vector based on
past experience. Upon the reception of the feedback,
at frame f, the automaton updates its action probabil-
ity distribution. First, the action corresponding to the
feedback is awarded:
p
i
k
( f + 1) = p
i
k
( f) +
400
∑
j=0, j6=k
L(p
i
j
( f) − a) (6)
In the above equation, i denotes the ONU, k denotes
the feedback action, L stands for the convergence
speed (the larger L the faster convergence), and a
symbolizes a quite small number used for avoiding
the probabilities taking zero values. Of course, the
award given to the actual action k stems from summa-
rizing a small fraction from all others j 6= k probabil-
ities. Accordingly, the probability of all other actions
is slightly reduced:
p
i
j
( f +1) = p
i
j
( f)−L(p
i
j
( f)−a) ∀ j 6= k, 0 ≤ j ≤ 400
(7)
In the light of the aforementioned aspects, the op-
eration of the LA lies in the Traffic Monitoring state
of Figure 1. In order to decide about the duration
of the isolation period the LA selects the most prob-
able action. Hence, given that the feedback recep-
tions increase the probability of the action that ap-
pears most, known as optimal action, the LA is able
to determine the traffic pattern of each ONU. In this
HYRA:AnEfficientHybridReportingMethodforXG-PONUpstreamResourceAllocation
9
way, the OLT effectively exploits the combination of
SR and TM policies in order to provide adequate up-
stream scheduling.
Finally, the isolated ONU returns to SR state. This
triggers the OLT to include this ONU to the next
downstream frame. If its traffic inactivity sustains,
i.e., it sends an empty XGEM again, the operation of
the ONU is switched to the Traffic Monitoring state.
Otherwise, it normally continues using the SR policy.
4.4 Operation
The complete operation of the enhanced OLT is de-
scribed in Algorithm1. The update procedure is
shown in Algorithm2. It is worth mentioning that a
probability vector update holds when a newer feed-
back is received by the OLT. This fact appears when
the propagation delay of an ONU is large, hence up-
stream bursts of this ONU arrive at OLT late, i.e., af-
ter the transmission of the next downstream frame. In
this case, the isolation of the ONU is canceled, if the
previous upstream burst was an empty XGEM. In ad-
dition, the feedback is equal to a
0
, meaning that the
received actual idle time of this ONU was less than
125 µsec.
5 PERFORMANCE EVALUATION
This section is devoted in presenting the evaluation
results of the proposed scheme.
Algorithm 1 : Bandwidth allocation process
Initialize the probability vector
for each 125 µsec do
for each ONU do
if ONU is isolated then
Exclude the ONU from the downstream
frame
end if
if ONU sent empty XGEM then
The LA decide about the isolation period
calculated in terms of 125 µsec; let it be T
Exclude the ONU from the downstream
frame for the next T frames
end if
if ONU sent active XGEM then
Apply SR policy to grant bandwidth
end if
end for
end for
Algorithm 2 : Update procedure
Initialize the probability vector
for each received upstream burst by ONU j do
Calculate the feedback
Associate the feedback to an action from the
pool
Update the ONU’s probability vector
if a newer upstream burst received then
if the burst included empty XGEM then
Set the feedback equal to a
0
Update the ONU’s probability vector
Cancel the isolation of the ONU (if any)
end if
end if
end for
5.1 Environment
A simulation environment was implemented in Mat-
lab in order to assess the proposed hybrid scheme. In
particular, the upstream process of a XG-PON is ex-
amined in terms of average packet delay. The intro-
duced scheme is compared with the pure SR policy
keeping the same network features and parameters.
The pure SR policy operates including all ONUs in
the downstream frame interdependentlyof their traffic
activity. Both schemes follow the same resource allo-
cation and QoS provisioning guidelines as described
in ITU-T G987.3 specifications.
Table 1: Simulation Parameters.
Upstream Rate 2.48832 Gbps
ONU Buffer Size 100 MB
Assured Bandwidth 500 Bytes
Maximum Bandwidth 750 Bytes
Downstream Frame Period 125 µsec
Learning Period 100 Downstream Frames
Guard Time 64 bits
L 0.1
a 10
−5
Number of actions 401
Simulation Time 100 sec
In order to provide realistic results, real multime-
dia packet traces have been used. The captured traf-
fic corresponds to the upstream direction, i.e., it con-
tains traffic flows from users to server (OLT). The
captured traffic streams were obtained using the Wire-
shark tool. Three traffic streams have been utilized: a)
voice over IP (VoIP) session using the user datagram
protocol (UDP) and the Skype application, b) real me-
dia streaming application based on transmission con-
trol protocol (TCP), and c) live stream session. Fur-
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
10
thermore, a constant bit rate (CBR) backgroundtraffic
was assumed for each ONU.
Concerning the VoIP application, each ONU pro-
duces an average traffic of 0.038 Mbps, while the av-
erage packet size is equal to 1372 Bytes. The session
includes a total number of 297 data packets. The real
media streaming application generate an average traf-
fic of 0.04 Mbps (per ONU) having an average packet
size of 121 Bytes. The session engages 1277 data
packets. The live stream (per ONU) contains 1458
data packets with average traffic equal to 0.05 Mbps
and an average packet size of 1430 Bytes. The back-
ground traffic generates an average load of 0.01 Mbps
for each individual ONU. The average summarized
traffic load per ONU reaches 0.138 Mbps.
Considering the propagation differentiation of the
ONUs the following formula was employed:
Distance
i
= RTT factor+ i (8)
The above formula expresses the distance of the
i-th ONU (Distance
i
) from the OLT in terms of Km.
The default value of the RTT factor is 30. Thus, if
10 ONUs are assumed, the last, i.e., the 10-th, ONU
is considered 40Km far from the OLT.
The (upstream) resource allocation incorporates a
traffic descriptor in accordance to the standard spec-
ifications. To be more specific, the default values of
each bandwidth flow were: a) the fixed bandwidth,
R
f
was set to 250 Bytes, b) the assured bandwidth,
R
a
, was set to 500 Bytes, and c) the maximum band-
width, R
m
, was set to 750 Bytes.
Each ONU possesses a large enough buffer so as
to exclude packet drops, e.g., 100 MBytes. The aver-
age packet delay is defined as the time elapsed from
the packet arrival to the final packet delivery to the
OLT.
Considering the LA operation, the convergence
speed parameter L was set to 0.1, since this value was
the most effective one based on the conducted experi-
ments. The value of protecting parameter a was set to
10
−5
.
To prevent upstream transmissions from colliding
and jamming each other, the OLT keeps a guard time
between upstream allocations equal to 64 bits.
For each experiment conducted the simulation
time was set to 100 sec. A learning initial period
for the LA was considered where only the probability
vector update takes place without engaging decisions.
For this learning period, the OLT utilizes the pure SR
policy. The duration of the learning period was 100
downstream frames, i.e., 100· 125 µsec = 12.5 msec.
Table 1 summarizes the main simulation parame-
ters.
5.2 Results and Discussion
5 10 15 20 25 30
0.35
0.4
0.45
0.5
Number of ONUs
Average upstream delay (ms)
HYRA
Pure Status Reporting
Figure 2: Average Upstream Delay vs. Number of ONUs.
5 10 15 20 25 30
10
15
20
25
30
35
Number of ONUs
Average upstream delay reduction (%)
Figure 3: Average Percentage of Upstream Delay Reduc-
tion vs. Number of ONUs.
First, the impact of the proposed scheme pertinent to
the population of the XG-PON is investigated. Fig-
ure 2 depicts the performance of both policies as the
number of ONUs increases. Four remarks are raised
by observing the results of this figure:
1. The average upstream delay increases; this is at-
tached to the fact that the population growth in-
duces more bandwidth requests for the upstream
direction leading a considerable number of data
packets to be delayed in the ONU queue due to
resource constraints.
2. HYRA presents much lower average upstream de-
lay; the superiority of the proposed scheme is un-
doubted. The suggested adaptive technique has
managed to adequately reform the way of con-
structing the upstream schedule. Underutilized
ONUs are temporarily isolated, giving the chance
to demanding ONUs to faster forwarding their
data to the OLT.
HYRA:AnEfficientHybridReportingMethodforXG-PONUpstreamResourceAllocation
11
3. The operation of the LA is accurate; This is
proved by the observation that the function of the
learning from experience tool is beneficial to the
network performance in terms of packet delay. If
the LA process was inaccurate then the operation
of HYRA would lead to higher upstream delays.
4. The extend of beneficial offering by HYRA is no-
table; According to Figure 3, the improvement
offered by HYRA is at least 8% and at most
33.5%. The mean improvement advanced by
HYRA reaches almost 22%, while it gains ground
as the population of the XG-PON becomes dense.
150 170 190 210 230 250
0.34
0.36
0.38
0.4
0.42
Fixed bandwidth (Bytes)
Average upstream delay (ms)
HYRA
Pure Status Reporting
Figure 4: Average Upstream Delay vs. Fixed Bandwidth.
150 170 190 210 230 250
20
22
24
26
Fixed bandwidth (Bytes)
Average upstream delay reduction (%)
Figure 5: Average Percentage of Upstream Delay Reduc-
tion vs. Fixed Bandwidth.
Second, the effect of the guaranteed bandwidth
rate is examined. In particular, Figure 4 draws the
performance of both schemes in terms of average
upstream packet delay when the fixed bandwidth
changes. By altering the level of guaranteed band-
width we endeavor to demonstrate how the XG-PON
operation is effected by applying an adaptive policy
such as HYRA. The fixed bandwidth, R
f
, alters from
150 to 250 Bytes. Concurrently, Figure 5 presents the
percentage of improvements compared to pure SR.
The observed aspects are summarized as follows:
1. The average upstream delay remains stable when
the pure SR is applied; this is caused due to the
fact that the SR policy provides the same fixed
bandwidth to all ONUs irregularly to the traffic
condition of each ONU.
2. HYRA performance excels; HYRA achieves not
only to support faster upstream communication,
yet it manages to accomplish even less upstream
delay as the fixed bandwidth increases. This phe-
nomenon is attached to the fact that the impact
of idle ONUs is better exploited by re-distributing
the bandwidth of inactive ONUs to the demanding
ONUs. Hence, the more the surplus bandwidth
the faster upstream data delivery.
3. The average improvements are guaranteed and
considerable; The minimum observed improve-
ment is achieved when the fixed bandwidth re-
ceived its minimum value. The difference be-
tween pure SR and HYRA peaks when the guar-
anteed bandwidth takes its maximum value.
25 26 27 28 29 30 31 32 33 34
0.32
0.34
0.36
0.38
0.4
0.42
0.44
RTT factor
Average upstream delay (ms)
HYRA
Pure Status Reporting
Figure 6: Average Upstream Delay vs. RTT Factor.
Third, the impact of the propagation delay is in-
spected in Figure 6. Once more, the superiority of
the proposed scheme is revealed. The RTT factor
changes from 25 to 34 shedding light to the impact
of the distance differentiation in the network perfor-
mance. It is obvious that the distance change keeps
the difference between the two schemes stable and
equal to 22.5% approximately. As expected, the prop-
agation impact induces a marginal effect to the perfor-
mance of HYRA; however the average delay keeps
increasing as the RTTs become larger. Nonetheless,
HYRA has managed to offer lower delay yet in wider
XG-PON systems.
Overall, the proposed scheme succeeds to offer
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
12
26 27 28 29 30 31 32 33 34
22.6
22.8
23
23.2
RTT factor
Average upstream delay reduction (%)
Figure 7: Average Percentage of Upstream Delay Reduc-
tion vs. RTT Factor.
considerable improvements to network performance
in terms of average upstream packet delay. The role
of LAs is deemed as quite beneficial providing accu-
rate and efficient decisions to the OLT. The simula-
tion results present a realistic picture since they based
on real traffic sessions, yet incorporating multimedia
traces.
6 CONCLUSIONS
An adaptive, efficient, and robust allocation scheme,
known as HYRA, was presented in this paper for
XG-PON systems. The scheme incorporates a LA
to the OLT in order to strengthen the efficacy of the
upstream resource allocation decisions. The learn-
ing from experience technique exploits both SR and
TM policies in order to benefit the demanding ONUs.
This is accomplished by re-distributing the surplus
bandwidth portion, gained by the idle ONUs, to the
users that really need more bandwidth. The re-
distribution is signaled by the existence of empty
XGEMs. Extensive simulation results using real mul-
timedia packet traces indicate the superiority of the
proposed scheme when compared to the pure SR pol-
icy. The improvements observed by the obtained re-
sults show that HYRA has managed an average im-
provement percentage of 22.5%, while it achieves to
reduce the delay levels by 33% in high populated XG-
PONs.
ACKNOWLEDGEMENTS
This work has been funded by the NSRF (2007-2013)
SynergasiaII/EPAN-II Program ”Asymmetric Passive
Optical Network for xDSL and FTTH Access,” Gen-
eral Secretariat for Research and Technology, Min-
istry of Education, Religious Affairs, Culture and
Sports (contract no. 09SYN-71-839).
REFERENCES
Bozicevic, G., Sarjanovic, E., and Tkalic, M. (2004). Dis-
tributed internet applications using learning automata.
In Electrotechnical Conference, 2004. MELECON
2004. Proceedings of the 12th IEEE Mediterranean,
volume 2, pages 599–602 Vol.2.
Economides, A. and Silvester, J. (1988). Optimal routing
in a network with unreliable links. In Computer Net-
working Symposium, 1988., Proceedings of the, pages
288–297.
Effenberger, F. (2010). Tutorial: Xg-pon. In Opti-
cal Fiber Communication (OFC), collocated National
Fiber Optic Engineers Conference, 2010 Conference
on (OFC/NFOEC), pages 1–37.
Eraghi, A. E., Torkestani, J. A., Meybodi, M. R., and Navid,
A. H. F. (2011). Cellular learning automata-based
channel assignment algorithms for wireless mobile ad
hoc networks. In Int Conf Mach Learn Comput IPC-
SIT, volume 3, pages 173–177.
Han, M.-S. (2014). Dynamic bandwidth allocation with
high utilization for xg-pon. In Advanced Communi-
cation Technology (ICACT), 2014 16th International
Conference on, pages 994–997.
Huang, J. (2008). Robotics, signal processing, decision
making and self-reproducing machine learning. In
Granular Computing, 2008. GrC 2008. IEEE Inter-
national Conference on, pages 306–311.
Jain, S., Effenberger, F., Szabo, A., Feng, Z., Forcucci, A.,
Guo, W., Luo, Y., Mapes, R., Zhang, Y., and O’Byrne,
V. (2011). World’s first xg-pon field trial. Lightwave
Technology, Journal of, 29(4):524–528.
Joshi, T., Ahuja, D., Singh, D., and Agrawal, D. (2008).
Sara: Stochastic automata rate adaptation for ieee
802.11 networks. Parallel and Distributed Systems,
IEEE Transactions on, 19(11):1579–1590.
Kanonakis, K. and Tomkos, I. (2009). Offset-based
scheduling with flexible intervals for evolving gpon
networks. Lightwave Technology, Journal of,
27(15):3259–3268.
Kumar, N., Iqbal, R., James, A., and Dua, A. (2013). Learn-
ing automata based contention aware data forward-
ing scheme for safety applications in vehicular ad hoc
networks. In e-Business Engineering (ICEBE), 2013
IEEE 10th International Conference on, pages 385–
392.
Lee, Y., Lee, D., Yoo, H., and Kim, Y. (2013). Fast manage-
ment of onus based on broadcast control channel for
HYRA:AnEfficientHybridReportingMethodforXG-PONUpstreamResourceAllocation
13
a 10-gigabit-capable passive optical network (xg-pon)
system. Communications and Networks, Journal of,
15(5):538–542.
Misra, S., Krishna, P., Saritha, V., and Obaidat, M. (2013).
Learning automata as a utility for power management
in smart grids. Communications Magazine, IEEE,
51(1):98–104.
Mullerova, J., Korcek, D., and Dado, M. (2012). On wave-
length blocking for xg-pon coexistence with gpon
and wdm-pon networks. In Transparent Optical Net-
works (ICTON), 2012 14th International Conference
on, pages 1–4.
Navid, A. (2010). Selarp: Scalable and energy-aware learn-
ing automata-based routing protocols for wireless sen-
sor networks. In Sensor Technologies and Appli-
cations (SENSORCOMM), 2010 Fourth International
Conference on, pages 570–576.
Nicopolitidis, P., Papadimitriou, G. I., Pomportsis, A. S.,
Sarigiannidis, P., and Obaidat, M. (2011). Adaptive
wireless networks using learning automata. Wireless
Communications, IEEE, 18(2):75–81.
Sarigiannidis, A., Nicopolitidis, P., Papadimitriou, G., Sa-
rigiannidis, P., and Louta, M. (2011). Using learn-
ing automata for adaptively adjusting the downlink-
to-uplink ratio in ieee 802.16e wireless networks. In
Computers and Communications (ISCC), 2011 IEEE
Symposium on, pages 353–358.
Sarigiannidis, P., Louta, M., Balasa, E., and Lagkas, T.
(2013a). Adaptive sensing policies for cognitive wire-
less networks using learning automata. In Computers
and Communications (ISCC), 2013 IEEE Symposium
on, pages 000470–000475.
Sarigiannidis, P., Papadimitriou, G., Nicopolitidis, P., and
Varvarigos, E. (2013b). Ensuring fair downlink allo-
cation in modern access networks: The xg-pon frame-
work. In Communications and Vehicular Technology
in the Benelux (SCVT), 2013 IEEE 20th Symposium
on, pages 1–5.
Venkata Ramana, B., Manoj, B. S., and Murthy, C. (2005).
Learning-tcp: a novel learning automata based reli-
able transport protocol for ad hoc wireless networks.
In Broadband Networks, 2005. BroadNets 2005. 2nd
International Conference on, pages 484–493 Vol. 1.
Yoshimoto, N., Kani, J., Kim, S.-Y., Iiyama, N., and Terada,
J. (2013). Dsp-based optical access approaches for
enhancing ng-pon2 systems. Communications Maga-
zine, IEEE, 51(3):58–64.
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
14
