Sie sind auf Seite 1von 6

PAS: A Fair Game-Driven DBA Scheme for

XG-PON Systems
Christos Dalamagkas, Ioannis Moscholios, Member, IEEE
Panagiotis Sarigiannidis, Member, IEEE, Department of Informatics and Telecommunications,
Department of Informatics and Telecommunications University of Peloponnese,
Engineering, 22100, Tripolis, Greece,
University of Western Macedonia, idm@uop.gr
Karamanli & Ligeris Street, 50100,
Kozani, Greece
cdalamagkas@ieee.org,
psarigiannidis@uowm.gr,

Thomas D. Lagkas, Member, IEEE Mohammad Obaidat, Fellow, IEEE


Computer Science Department, The University of Sheffield, King Abdullah II School of IT, University of Jordan, Jordan
International Faculty, CITY College, m.s.obaidat@ieee.org
Thessaloniki, Greece,
T.Lagkas@sheffield.ac.uk

Abstract—New generation passive optical networks (NG- (NG-PON2) standard that allows 4-8 time and wavelength
PONs) are emerged as promising technologies for delivering division multiplexing TWDM channel pairs.
bandwidth-demanding services in modern access networks. The
PON technologies offer significant capabilities. They allow
10-gigabit-capable passive optical network (XG-PON) standard
has been proposed as a high-throughput architecture that optical lightpaths between the user premises (curb, building
realizes the Fiber-To-The-x (FTTx) paradigm, where x stands and home) and the Central Office (CO) in a very cost-
for curb (FTTC), building (FTTB) or home (FTTH). XG-PON effective way. The main network entities of a PON are
is a powerful technology, capable of supporting 10 Gbit/s in the the Optical Network Units (ONUs), which represent the
downstream and 2.5 Gbit/s in the upstream direction. However,
connection interfaces where the end users access the PON
in order to fully utilize the huge bandwidth in such technology,
an efficient and fair bandwidth allocation scheme is required architecture and the Optical Line Terminal (OLT), which
for adequately coordinating the transmission opportunities of constitutes the gateway of the architecture to the backbone.
the involved network entities based on their real bandwidth Fig. 1 illustrates a typical architecture of a PON system in a
demands. A multitude of techniques and schemes have been tree topology.
presented over the last few years aiming at providing dy-
namic bandwidth distribution in PONs and NG-PONs systems. XG-PON system is emerged as one of the most promising
However, the vast majority of those techniques are based on candidates for representing NG-PON paradigm. It incorpo-
heuristics. This paper develops a novel bandwidth distribution rates a set of strong definitions and tools, such as a well-
algorithm which is based on the tragedy of the commons defined Quality of Service (QoS) allocation policy. However,
theory, aiming at ensuring a fair bandwidth distribution among an efficient Dynamic Bandwidth Allocation (DBA) scheme
the Optical Network Units (ONUs) in a deployed XG-PON
system. The algorithm is able to apply a balanced bandwidth is needed to leverage the huge potential of XG-PON systems.
distribution schedule amongst the connected ONUs, formed as The bandwidth distribution process in modern PON sys-
players, aiming at ensuring a fair transmission opportunity tems can be seen as a situation where many participants
arrangement in the upstream direction. interact with each other. In many cases, the participants
do not acting fairly. For example, a participant may try to
I. I NTRODUCTION
monopolize the offered transmit opportunity in the upstream
Optical networks present very promising characteristics direction of a PON system. As a result an unfair bandwidth
such as high throughput, flexibility, scalability and energy- transmission program may occurred, if the OLT, which is
efficiency. Next Generation PONs (NG-PONs) allow 10 responsible of creating the (transmission) program, does not
Gbit/s nominal speed in the downstream direction or even prevent such a behaviour. Game theory may be the solution
symmetrically. For example, the 10-gigabit-capable PON in such situations, where the participants could be modeled
(XG-PON) system supports up to 10 Gbit/s downstream as players that interact with each other. The tragedy of the
rate [1]. Recently, the Telecommunication standardization commons concept can be applied in games where n players
sector of the International Telecommunication Union (ITU- have to share bandwidth. This game is classified based on
T) presented the 40-Gigabit-capable passive optical networks two main cases: a) if the bandwidth to be shared is enough
C. Dalamagkas, P. G. Sarigiannidis, I. D. Moscholios, T. D. Lagkas, and M. S. Obaidat, “PAS: A Fair
Game-Driven DBA Scheme for XG-PON Systems,” Proceedings of the 11th International Symposium
on Communication Systems, Networks, and Digital Signal Processing (CSNDSP), Budapest, Hungary,
July 2018.

The published version is available at IEEE Xplore: http://dx.doi.org/10.1109/CSNDSP.2018.8471787


allocation decisions based on the outcome of an adjustable
multistage optimization problem.
In the light of the aforementioned remarks, this paper is
focused on providing a fair bandwidth distribution schedule
IP
by applying a game stemming from the tragedy of the
commons concept. To this end, we contribute to devising
VoIP a mathematically-based bandwidth allocation scheme, which
Passive
Central
Office
Optical Line
Terminal
Splitter/
Combiner
is driven by a game with a Nash Equilibrium. The game is
IPTV demonstrated in a XG-PON system, assuming that the avail-
able bandwidth is not always enough for all end users. To this
Optical
Network end, a Nash equilibrium is defined and the game parameters
Unit
are settled for devising a fair distribution algorithm called
Proportional Allocation Scheme (PAS).
Fig. 1: Typical architecture of a PON system in a tree The remainder of the paper is organized as follows. Section
topology. II presents the related background. The motivation behind
this paper is provided in III. A detailed description of the
proposed PAS scheme is provided in Section IV. Section V-B
for all, based on the players’ requests then the solution is is dedicated to the evaluation process through numerical
straightforward since each player receives what requested and results. Finally, conclusions are given in Section VI.
b) if the available bandwidth is not enough for all, then a
II. BACKGROUND
bandwidth sharing policy should be devised for meeting the
players’ requests in a fair way. XG-PON systems constitute a promising solution in the
The latest research efforts in the XG-PON bandwidth context of NG-PONs since they are able to support 10 Gbit/s
distribution domain present a growing interest. Some efforts in at least one direction [8]. The XG-PON Transmission Con-
are focused on improving the polling process by introducing vergence (XGTC) layer incorporates functional protocols and
flexible intervals between the message exchange. The work in procedures, including resource allocation and QoS provision-
[2] is an indicative example of dealing with flexible intervals ing between the upper layers and the physical (PHY) layer
for alleviating the high propagation delays in the control [9]. It encloses three sub-layers: a) the service adaptation
message exchanging between the OLT and the ONUs in a sub-layer, where Service Data Unit (SDU) encapsulation and
GPON system. In a different prospective, the efforts in [3] multiplexing takes place, forming XG-PON Encapsulation
intend to provide a fair bandwidth allocation schedule rather Method (XGEM) frames, b) the framing sub-layer, where
than optimizing it in terms of throughput. Those methods use the constructed XGEM frame is received and the downstream
heuristic techniques by maintaining a fairness index during XGTC frame is formed and c) the PHY sub-layer, where bit
the allocation process. Other works try to address the problem error correction algorithms, content scrambling, and frame
of bandwidth insufficiency, i.e., when the available bandwidth synchronization are performed. The Allocation ID (Alloc-ID)
is not enough for all additional ONU requests. For instance, is used to identify the final user behind the XGTC payloads.
in [4], a surplus bandwidth sharing method was proposed The OLT uses a downstream frame that is periodically
where a fair re-distribution process is followed to allocate the transmitted to all ONUs for a) data delivery, b) signaling
available bandwidth to the ONUs that are still unsatisfied. and c) control information delivery. The OLT broadcasts
Applications of XG-PON include the implementation of the frame in a periodic fashion every 125 µsec. Given
hybrid optical-wireless networks, where the XG-PON real- that the downstream transmission rate is 9.95328 Gbit/s,
izes the network backhaul and multiple Long Term Evolu- 155520 bytes are available in the downstream direction.
tion (LTE) radio access networks in the fronthaul part [5]. One of the most important fields in the downstream frame
The proposed application receives traffic-aware knowledge is the BandWidth map (BWmap) array. It is used by the
from the Software Defined Network (SDN) controllers and OLT to broadcast to the ONUs the constructed transmission
applies an adjustment on the uplink-downlink configuration schedules which all ONUs will follow in the upcoming
in the LTE radio communication. The introduced scheme is upstream frames. The ONUs read the BWmap array and they
evaluated in a realistic environment using real traffic traces are informed about the time and the duration of the upstream
such as Voice over IP (VoIP), real-time video and streaming transmission opportunities [10].
video. Another application is presented in [6], where an On the other direction, the supported transmission rate is
optimized DBA is designed for supporting front-hauling over 2.48832 Gbit/s while the upstream frame is 125 µs long.
XG-PONs. The performance evaluation of the optimized As a result, 38880 bytes become available for all ONUs
DBA takes place in virtualized Cloud Radio Access Network in each upstream frame. This fact implies that all ONUs
(C-RAN) architecture. In addition, in [7] a new resource have to share this available bandwidth to claim upstream
allocation scheme for cutting-edge Fiber-Wireless networks transmission opportunities in each upstream window. The
was introduced that can be applied in the fronthaul portion Physical Synchronization Block (PSB) of the upstream frame
of 5G-enabled architectures. The proposed scheme performs (PSBu) contains the preamble and the delimiter fields and
comes first within the upstream frame. Then, the XGTC TABLE I: Game Parameters
burst follows, which includes a control field in the front
C Available upstream capacity
(XGTC header) and a trailer (XGTC trailer). The Dynamic N Set of players (Alloc-IDs)
Bandwidth Report (DBRu) implies the use of the status a = {a1 . . . ai } Vector of assigned bandwidth to players in N
reporting method. Status reporting allows ONUs (Alloc-IDs) values of GrantSize fields in allocation y
Ui (ai ) Utility enjoyed by player i, cause of allocation ai
to report their buffer status. This is accomplished by the
s = {s1 . . . si } Vector of bids of player in N
Buffer Occupancy (BufOcc) field. values of BufOcc fields of Alloc-IDs in cycle y
According to the ITU-T specifications, a portion of band-
width is guaranteed for all Alloc-IDs. It includes a fixed rate,
an assured rate and a maximum rate that each Alloc-ID has
The game equilibrium depends on the player utility func-
the right to claim. The OLT secures the guaranteed bandwidth
tion. Let ai represents the bandwidth assigned to i−th Alloc-
for all Alloc-ID and then allocates the surplus bandwidth to
ID by the DBA mechanism. Hence, a reasonable and realistic
all Alloc-IDs that request additional bandwidth.
utility function is needed to describe the enjoyed satisfaction
III. M OTIVATION of player i by the bandwidth allocation ai . We adopt the
A. Tragedy of the Commons elastic traffic utility function as described in [12]. Let Ui be
this utility function. Note that the explicit definition of Ui is
When every player or partner tries to maximize one’s ben- not required in the context of this game, but it is important
efit from a shared pool of resources, a situation is occurred to assume that the function is log-concave, so the marginal
known as tragedy of commons [11]. In essence, the tragedy utility concept is being applied. The definition of Ui is given
stems from the resource devastating, which comes upon when as follows:
the demand for the resource overwhelms the supply. This
phenomenon is caused by a group of players or partners who Definition IV.1. Ui is concave, strictly increasing and
act selfishly by consuming or requesting additional portion of continuous ∀i ∈ N , where ai ≥ 0. It is continuously
resources and therefore they harm others who can no longer differentiable for ai > 0, while Ui′ (ai )|ai =0 6= ∞. Let
utilize their part of the shared resources. In other words, some U denote the set of all utility functions that satisfy the
players of the game neglect the well-being or the fairness of aforementioned conditions.
the system towards their own interests. Besides, this situation
comes as a result of the availability of the resources to all
As a next step, a pricing scheme is defined which relates
players without rules or policies.
what a player offers to what a player should receive. Consider
B. Game Theory on Bandwidth Distribution that
P the resource manager always try to clear the market,
The tragedy of the commons situation has a lot of similari- i αi = C. Further assuming that the manager does not
price-discriminate, each player is charged the same price µ >
ties with the bandwidth distribution problem in PONs. A pool
0, leading to ai = si /µ. In this case, we expect the price µ
of resources (available bandwidth in the upstream direction)
to satisfy the following equation:
is available to multiple players (ONUs). Each player (Alloc-
ID) is free to request as much bandwidth as desires. Thus, X si
each Alloc-ID might act selfishly by trying to monopolize µ= (1)
i
C
the bandwidth distribution process. Fueled by the similarity
between the tragedy of the commons and the bandwidth This price leads to a market-clearing process, where the
distribution problem in PONs, in this paper we investigate the price is set to such a value so as the offer to be equal to the
function of solutions related with the tragedy of the commons request. In the following, the payoff function is defined which
situation in the XG-PON systems. determines the player satisfaction by taking into account the
strategy from the other players. Considering that s−i stands
IV. G AME - DRIVEN XG-PON BANDWIDTH A LLOCATION for the other players strategy, except i, each player selects a
A. Game Theory Parameters strategy si in order to maximize the following equation [13]:
This section describes the proposed game-driven fair band-
width allocation for XG-PON systems. First, the resource  !
s
Pi

manager is defined as the bandwidth allocation coordinator. U
i C − si , if si > 0
Qr (si ; s−i ) = j sj (2)
This role is played by the OLT. The set of Alloc-IDs that 
U (0), if si = 0

request non-guaranteed bandwidth (N ) are the players of the
game. For simplicity reasons we assume that each Alloc-ID
Given that the price is different for each player (Eq. 1)
is connected to a single ONU. Strategy of each player is
and the payoff function is adopted based on that price, the
the bandwidth that requests, i.e., the value of the BufOcc
game affects the players in N explicitly. In accordance with
field contained in the upstream XGTC burst. The resource
the payoff function, the Nash equilibrium is now feasible:
manager receives all bandwidth requests and proceeds to the
bandwidth allocation schedule construction. Table I summa- Definition IV.2. A Nash equilibrium of the game defined by
rizes the parameters adopted in the game. (Q1 , . . . , Qi ) is a vector s ≥ 0 such that ∀i:
To conclude the proof, observe that GAME has a strictly
Qi (si ; s−i ) ≥ Qi (s̄i ; s−i ), ∀s̄i ≥ 0 (3) concave and continuous objective function over a compact
feasible region, and thus has a unique optimal solution, which
We have now defined all required game parameters. The implies uniqueness of the Nash equilibrium as well [15].
following definition charazterizes the underlying game as an
optimization problem of bandwidth distribution allocation in B. The PAS Algorithm
XG-PON systems. Note that the utility function has been The presented PAS algorithm is triggered upon distributing
edited accordingly, so that the players are price anticipators the guaranteed bandwidth to all Alloc-IDs (Rf stands for the
[14]: fixed bandwidth, Rα stands for the assured bandwidth) and
Theorem IV.1. Suppose |N | > 1. Then there exists a the following conditions are met: a) there is still available
unique Nash Equilibrium s ≥ P 0 of the game defined by bandwidth, b) at least two Alloc-IDs request additional band-
(Q1 , . . . , Q|N | ) and it satisfies i si > 0. In this case, the width and c) the additional bandwidth requests exceed the
vector a defined by: available bandwidth. The Algorithm introduces the proposed
game theory-based bandwidth allocation process.
si The algorithm receives all the (pending) Alloc-IDs’ re-
ai = P C, ∀i ∈ N (4)
j sj quests (R), after addressing the guaranteed bandwidth re-
is the unique optimal solution to the following optimization quests. Also, it receives the available bandwidth C and the
problem (GAME): set of all Alloc-IDs (I). The algorithm exports the set of
final allocations (A), which describes the final bandwidth
X
maximize Ûi (ai ) distribution for all Alloc-IDs (players). First of all, the set
ai of players (N ), strategies (S) and allocations are initialized.
i
X
(5) Then these sets are filled by determining the players of the
subject to ai ≤ C
game and their strategies (lines 4-9). In lines 10 − 13 the
i
algorithm performs the game theory approach based on the
ai ≥ 0, ∀i ∈ N
Nash Equilibrium.
where Ûr is defined as follows:
Algorithm: Proportional Allocation Scheme
 ai  Input: R, C, I
ai  a 1
 Z
i
Ûi (ai ) = 1 − Ui (ai ) + Ui (z)dz (6) Output: A
C C ai 0 1: N ← ∅
Proof. Given the fact that each component of vector s is a 2: S ← ∅
positive value, the pay-off of each user is strictly concave 3: A ← ∅
and continuous in si so that s is a Nash equilibrium if and 4: for i ∈ U do
only if the following first-order conditions hold: 5: if Ri > 0 then
6: N ←N ∪i
! ! P
sj 7: S ← S ∪ ri
s si
Ui′ Pi C 1− P =
j
, if si > 0 8: end if
j sj j sj C 9: end for
(7) 10: for i ∈ N$do %
si · C
P
j sj
Ui′ (0) ≤ , if si = 0 11: ai ← P
C j sj
(8) 12: A ← A ∪ ai
P 13: end for
Replacing ρ = j sj /C and ai = si /ρ then the preceding
conditions can be rewritten as:
V. P ERFORMANCE E VALUATION
Ui′ (ai ) = ρ, if ai > 0 (9) In this section, we provide the performance evaluation of
Ui′ (0) ≤ ρ, if ai = 0 (10) the proposed game-based algorithm.
X
ai = C (11) A. Evaluation Enviroment
i
An XG-PON simulation environment was implemented
Since the utility functions Ûi (ai ) are strictly concave using Matlab. All ITU-T specifications were followed, e.g.,
and continuous over 0 ≤ ai ≤ C, the preceding first- the upstream and the downstream rate of the XG-PON were
order conditions are sufficient optimality conditions for the set to 2.48832 Gbit/s and 9.95328 Gbit/s, respectively. The
game.
P We conclude that s is a Nash equilibrium if and only buffer of each ONU was large enough to avoid packet drops.
if j sj > 0, and the resulting allocation a solvesP the Each ONU consists of a single Alloc-ID. The fixed and the
problem GAME with Lagrange multiplier ρ = j s j /C. assured bandwidth rates were set to 250 and 500 bytes per
Alloc-ID. A guard time of 64 bits was inserted between 1
Load fairness index as ONUs increase

upstream allocations of different ONUs. The ONUs have 0.9


been placed uniformly having a distance from 20 to 60 Km. 0.8
PAS was compared to the single standard bandwidth

Load fairness index


0.7 Default
allocation scheme (XG-PON Standard) [1]. The standard PAS

scheme applies the pure allocation method where: a) the 0.6

guaranteed bandwidth requests are addressed and b) the sur- 0.5

plus bandwidth is shared all Alloc-IDs that request additional 0.4

bandwidth equally. No fairness index is monitored. 0.3


Each simulation experiment was conducted for 60 sec. The 0.2
traffic used is synthetic and generated based on real traffic
0.1
traces. Three real services were considered, namely: a) VoIP, 4 8 12 16 20 24 28 32
ONUs
b) video streaming application, and c) live stream session.
Synthetic traffic flows were produced for each ONU in the Fig. 2: Load fairness subject to the number of ONUs.
upstream direction by combining multiple services so that
the aggregate traffic demand of each ONU was 100 Mbit/s, 1
Delay fairness index as ONUs increase

in order to exceed the available bandwidth (38880 bytes each


125 µs) when the number of ONUs gets higher than 24. To 0.9

this end, a background traffic of constant bit rate was also 0.8

Delay fairness index


multiplexed with the real traces.
0.7
Default
B. Numerical Results PAS
0.6
Numerical results are presented in this sub-section in terms
of fairness provisioning. In order to assess fairness we adopt 0.5

the well known Jain’s fairness index in two modes: a) the 0.4
load fairness index and b) the delay fairness index. The load
fairness index denotes how fair is the bandwidth distribution 0.3
4 8 12 16 20 24 28 32
subject to the allocated traffic each ONU received. It is ONUs

defined as follows: Fig. 3: Delay fairness subject to the number of ONUs.


PN
( i=1 li )2
J(l1 , l2 , · · · , lN ) = PN (12)
N i=1 li2 to do with the (network) saturation point which deliberately
The parameter li expresses the fraction of the bandwidth occurs when the number of ONUs is 24. During the first
phase, the performance of the two schemes is similar while
received by i ONU divided by the total bandwidth allocated
to all ONUs. For example, in an XG-PON system with three PAS succeeds to slightly excel, especially when the number
of ONUs is 16. This minor improvement is due to the narrow
ONUs consider that the first ONU receives 1400 bytes, the
operation of the PAS bandwidth allocation algorithm, which
second ONU receives 1800 bytes and the third ONU receives
1100 bytes. Hence, l1 = 1400 1800 1100 is partly triggered when the requested bandwidth is larger
3800 , l2 = 3800 , l3 = 3800 , while than the offered capacity. As expected, both schemes present
J(l1 , l2 , l3 ) = 0.9615. The optimal value of the load fairness
index is 1. losses in terms of load fairness, even though PAS manages
On the other hand, the delay fairness index denotes how
fair is the bandwidth distribution subject to the delay mea- Mean packet delay as ONUs increase
180
sured in each ONU. It is defined as follows: Default
160 PAS
PN
( i=1 ai )2 140
J(a1 , a2 , · · · , aN ) = PN (13)
Mean Packet Delay (msec)

N i=1 a2i 120

100
The parameter ai stands for the fraction of the delay expe-
rienced by i ONU divided by the average delay experienced 80

by all ONUs. Similar to the load fairness index, the optimal 60

value of the delay index is 1. 40

Fig. 2 illustrates the load fairness index as the number of 20


ONUs change within [4, 8, 12, 16, 20, 24, 28, 32]. Two phases 0
can be distinguished as the number of ONUs becomes larger. 4 8 12 16
ONUs
20 24 28 32

The first phase is defined from 4 to 20 ONUs while the


second phase applies when the number of ONUs is larger Fig. 4: Average upstream delay subject to the number of
than 20. The subtle point that divides the two phases has ONUs.
Goodput as ONUs increase
2500
VI. C ONCLUSIONS
2000 This paper validates the superiority of the game theory
approach in handling bandwidth allocation in XG-PON sys-
tems when the requested bandwidth is likely more than the
Goodput(Mbps)

1500
offered capacity. The approach is realized by the PAS scheme
1000
which is able to adequately coordinate the transmission
opportunities of ONUs, when contending each other, by
500
providing a fair resource allocation policy. The rationale
Default
behind the applied game is based on the tragedy of the
0
PAS common concept. The PAS scheme was applied to the XG-
4 8 12 16 20 24 28 32 PON paradigm by offering an efficient resource allocation
ONUs
solution, resulting in a fair and efficient transmission schedule
Fig. 5: Aggregated system goodput subject to the number of in the upstream direction. Simulation results demonstrate
ONUs. the superiority of the proposed scheme against the standard
bandwidth allocation in terms of fairness while maintaining,
yet slightly improving, the average upstream latency and the
to limit these losses about 5% on average. The equilibrium network goodput.
that PAS offers does actually take place in the second phase
where the network is saturated. The designed game is totally R EFERENCES
triggered since the case where the requested bandwidth is not
[1] G.987.1 : 10-Gigabit-capable passive optical networks (XG-PON):
enough for all ONUs is much frequent than in the first phase. General requirements, International Telecommunication Union, ITU,
Thus, while the default scheme collapses, the proposed game- 2016. [Online]. Available: https://www.itu.int/rec/T-REC-G.987.1/en
theoretic scheme is able not only to maintain the recorded [2] K. Kanonakis and I. Tomkos, “Offset-based scheduling with flexible
intervals for evolving gpon networks,” Lightwave Technology, Journal
load fairness in the levels of the first phase, but to increase of, vol. 27, no. 15, pp. 3259–3268, Aug 2009.
it to the maximum when the number of ONUs is more than [3] P. Sarigiannidis, G. Papadimitriou, P. Nicopolitidis, E. Varvarigos, and
24. K. Yiannopoulos, “Towards a fair and efficient downlink bandwidth
distribution in xg-pon frameworks,” in Mediterranean Electrotechnical
Fig. 3 shows the latency fairness index. Once again the Conference (MELECON), 2014 17th IEEE, April 2014, pp. 49–53.
performance interpretation is divided into two phases, i.e., [4] M.-S. Han, “Dynamic bandwidth allocation with high utilization for
before and after the saturation point. Here, the observed xg-pon,” in Advanced Communication Technology (ICACT), 2014 16th
International Conference on, Feb 2014, pp. 994–997.
performance of PAS is much more intensive. While the [5] P. Sarigiannidis, A. Sarigiannidis, I. Moscholios, and
default bandwidth allocation scheme collapses when N = 24 P. Zwierzykowski, “Diana: A machine learning mechanism for
(it seems to improve the recorded latency fairness index later adjusting the tdd uplink-downlink configuration in xg-pon-lte
systems,” Mobile Information Systems, vol. 2017, 2017.
due to queue stabilization), the proposed scheme achieves to [6] A. M. Mikaeil, W. Hu, T. Ye, and S. B. Hussain, “Performance
optimize the latency fairness index to almost 1. The good evaluation of xg-pon based mobile front-haul transport in cloud-
performance of the PAS scheme is due to the fact that ran architecture,” IEEE/OSA Journal of Optical Communications and
Networking, vol. 9, no. 11, pp. 984–994, Nov 2017.
the variance of the latency each ONU experiences depends [7] T. Lagkas, P. Sarigiannidis, P. Bellavista, and C. Giannelli, “Multi-
strongly on the propagation delay rather than the transmission stage resource allocation in hybrid 25g-epon and lte-advanced pro fiwi
delay. As a result, the expected differences in terms of latency networks for 5g systems,” IET Networks, 2018.
[8] F. J. Effenberger, “The xg-pon system: Cost effective 10 gb/s access,”
are slightly reflected in Fig. 3 compared to Fig. 2. J. Lightwave Technol., vol. 29, no. 4, pp. 403–409, Feb 2011.
The performance analysis is not enough if only fairness in- [Online]. Available: http://jlt.osa.org/abstract.cfm?URI=jlt-29-4-403
dices are examined without observing the average latency and [9] Z. Li, L. Yi, and W. Hu, “Key technologies and system proposals
of twdm-pon,” Frontiers of Optoelectronics, vol. 6, no. 1, pp. 46–56,
the goodput. In order to acquire safe performance evaluation 2013. [Online]. Available: http://dx.doi.org/10.1007/s12200-012-0305-
results, the measurement of the average latency is provided 7
in Fig. 4. As expected, the average latency increases as the [10] G.987.2 : 10-Gigabit-capable passive optical networks (XG-
PON): Physical media dependent (PMD) layer specification,
number of ONUs becomes larger. However, what is important International Telecommunication Union, ITU, 2016. [Online].
to point here is that the performance of the two schemes is Available: https://www.itu.int/rec/TREC- G.987.2/en
almost the same, while PAS offers a slight improvement. This [11] G. Hardin, “The tragedy of the commons,” Science, vol. 162, no. 3859,
pp. 1243–1248, 1968.
is quite important so as to verify the complete superiority of [12] S. Shenker, “Fundamental design issues for the future internet,” IEEE
the PAS since it succeeds to offer a fair bandwidth allocation Journal on Selected Areas in Communications, vol. 13, no. 7, pp. 1176–
while keeping, yet slightly improving, the average latency in 1188, 9 1995.
[13] R. Johari and J. N. Tsitsiklis, “Efficiency loss in a network resource
the XG-PON system. This fact is also validated by Fig. 5 allocation game,” Mathematics of Operations Research, vol. 29, no. 3,
which shows the network goodput. Fig. 5 verifies two main pp. 407–435, 8 2004.
points: a) it clearly highlights the saturation point (it happens [14] B. Hajek and G. Gopalakrishnan, “Do greedy autonomous systems
make for a sensible internet?” in Conference on Stochastic Networks,
at the point the goodput is stabilized) and b) it validates the Stanford University, vol. 222, 2002.
superiority of PAS since it is evident that the applied game [15] T. R. Noam Nisan, Algorithmic Game Theory. Cambridge University
offers equilibrium in the bandwidth allocation process. Pr., 2007.

Das könnte Ihnen auch gefallen