Вы находитесь на странице: 1из 6

IEEE ICC 2015 - Optical Networks and Systems Symposium

Offline Energy-Efficient Dynamic Wavelength and


Bandwidth Allocation Algorithm for TWDM-PONs
M. Pubuduni Imali Dias and Elaine Wong

Dung Pham Van and Luca Valcarenghi

NICTA, Department of Electrical and Electronic Engineering


Parkville, Australia
Email: i.dias@student.unimelb.edu.au

Scuola Superiore SantAnna,


Pisa, Italy

AbstractWe previously proposed and numerically analyzed


a theoretical framework of an energy-efficient offline dynamic
wavelength and bandwidth allocation (DWBA) algorithm designed for a delay-constrained time and wavelength division
multiplexed passive optical network (TWDM-PON). This DWBA
algorithm exploits the tunability and the sleep/doze capabilities
of a 10 Gbps vertical-cavity surface-emitting optical network
unit (10G-VCSEL-ONU) to improve the energy-savings at the
OLT and the ONUs, respectively. In this work, using simulation
results on the number of active wavelengths and the percentage
of energy-savings, we verify the theoretical framework proposed
in our previous study. Most importantly, we show that the
average delay of upstream packets are not adversely affected
by the proposed energy-saving mechanism and is kept below the
specified maximum.

Passive
splitter

!.('
! ('

! )'

,-#'
!

!"#$%&!"#'

!.)'

Passive
splitter

+*"'

+*"'

!.*'
! *'

+*"'

'

'

+*"'

,012345',-#6'27'
+*"6'782169:7'
;1'!.(<<<=!.*'
21>'85?5:@5';1'
! (<<<==! *'
'
'
'

+*"'

Fig. 1: General architecture of a TWDM-PON.

I. I NTRODUCTION
The Full Service Access Network (FSAN) group has specified the network requirements of future broadband networks
under next-generation PON stage 2 (NG-PON2). Based on
these specifications, the future broadband networks are required to support high data rates, high split ratios, and longreach communication. As importantly, these networks should
be cost-effective, energy-efficient, and co-exist with G-PONs
[1]. After evaluating different network configurations, the
FSAN has selected time and wavelength division multiplexed
PON (TWDM-PON) as the favorable network architecture that
satisfies these requirements [2].
Figure 1 presents the general architecture of a TWDMPON. A TWDM-PON consists of multiple wavelengths, tunable transceivers at the optical network units (ONUs), and
tunable or fixed-tuned transceivers at the optical line terminal
(OLT). Unlike in seeded/reflective WDM-PONs, a TWDMPON does not have a centralized light source and as a
result, the transceivers at the ONUs should be able to tune
into multiple wavelengths supported by the network. For
example, in the TWDM-PON shown in Fig. 1, the tunable
transceivers at ONUs transmit on wavelengths l10 ......lN0 and
receive on wavelengths l1 ......lN . Due to the tunability of
ONU transceivers, the OLT can control the ONU distribution
among wavelengths. When certain wavelengths are idle for a
long duration, the OLT can migrate the ONUs supported by
these idle wavelengths to other active wavelengths and switch
off the OLT transcivers associated with these idle wavelengths.
As an active OLT transceiver consumes significant power,

978-1-4673-6432-4/15/$31.00 2015 IEEE

+*"'

+/,'

switching off idle transcivers improves the energy-savings at


the OLT. Following this concept, various solutions have been
proposed for energy-efficient TWDM-PONs.
In [3], the authors have proposed an user-migration scheme
to minimize the number of active wavelengths in the network.
Under the proposed solution, the OLT monitors the network
for wavelengths with lighter traffic. The OLT then migrates
the ONUs supported by the idle wavelengths to other active
wavelengths and switches off the idle ones, thereby saving
significant energy at the OLT. In a similar attempt, the authors
in [4] have proposed and experimentally evaluated a TWDMPON in which, lightly loaded wavelengths are powered off at
the OLT for improved energy-efficiency. In [5], the authors
have proposed wavelengths optimization in conjunction with
sleep mode to improve the energy-efficiency of a TWDMPON further. Under the proposed solution, the OLT monitors
traffic load of the network for a period of T and determines
the number of active wavelengths accordingly. In addition,
the OLT transits idle ONUs into sleep mode to improve the
energy-savings at the ONUs. The authors have also analyzed
the energy-efficiency of a novel TWDM-PON with wavelength
selective switches (WSS) [6]. The proposed scheme optimizes
the grouping of network users among wavelengths based on
data rate and the distance to the ONUs and keeps a minimal
number of wavelengths active.
In existing work [3]-[6], the primary objective is to minimize the energy consumption of a TWDM-PON. While

5018

IEEE ICC 2015 - Optical Networks and Systems Symposium

optimizing the number of wavelengths and incorporating sleep


mode improve the energy-efficiency of a TWDM-PON, they
also increase the average delay of the network. Most of the
internet-based applications such as IP telephony and video
gaming are delay-sensitive and have restrictions on maximum
allowable delay to provide guaranteed quality of service (QoS)
to the customers. While the solutions proposed in [3]-[6] report
acceptable delay values, these solutions do not consider a
delay-constrained network. To address this issue, we proposed
an energy-saving mechanism that uses both wavelength optimization and sleep operation in a delay-constrained TWDMPON in [7]. The proposed algorithm however, is specific to a
given sleep control function and is only evaluated for a delay
constraint of 10 ms. The solution proposed in [7] therefore
does not provide a general framework that determines the
number of active wavelengths and the sleep time for a delayconstrained TWDM-PON. As importantly, these methods have
not considered the doze mode capabilities of ONUs to achieve
energy-savings when the idle time of an ONU is less than
sleep-to-active transition time [8].
To address these shortcomings, in [9], we proposed and
numerically analyzed an offline dynamic wavelength and
bandwidth allocation algorithm (OFF-DWBA). For a delayconstrained TWDM-PON, the proposed algorithm determines
(a) the number of active wavelengths at the OLT and (b) the
sleep or doze time of the ONUs. The OFF-DWBA exploits
the tunability [10] and the sleep/doze capabilities [11] of a
10 Gbps vertical-cavity surface-emitting laser (10G-VCSEL)
ONU to improve the energy-efficiency of a TWDM-PON. In
addition, the offline nature of the OFF-DBA algorithm ensures fairness among the ONUs in resource allocation. Based
on the numerical analysis provided in [9], the OFF-DWBA
algorithm optimizes the number of active wavelengths and
improves the energy-efficiency of a TWDM-PON. However,
it is important to verify that in a practical TWDM-PON,
the proposed framework achieves the percentage of energysavings reported in [9]. Most importantly, we should ensure
that when the OFF-DWBA is deployed in a practical delayconstrained TWDM-PON, the average delay of upstream packets does not exceed the specified maximum. To address these
practical concerns and to provide a comprehensive analysis
of our theoretical framework proposed in [9], in this work,
we verify our theoretical framework using simulations. The
simulation results on the number of active wavelengths, the
percentage of energy-savings at the OLT, and the percentage
of energy-savings at the ONUs verify the numerical results
reported in [9]. Most importantly, the simulation results on
average delay indicate that the average delay of a TWDMPON, under the proposed OFF-DWBA algorithm, does not
exceed the specified maximum delay.
The rest of the paper is structured as follows. Section
II describes the proposed OFF-DWBA algorithm in detail.
Section III presents the analytical and simulation results of
the proposed algorithm. Finally, a summary of our findings is
presented in Section IV.

II. P ROPOSED OFFLINE DYNAMIC WAVELENGTH AND


(OFF-DWBA) ALGORITHM

BANDWIDTH ALLOCATION

This section presents the theoretical framework of the OFFDWBA algorithm proposed in [9] in detail. The objective of
the proposed OFF-DWBA algorithm is to minimize the energy
consumption of a TWDM PON while maintaining the average
delay of upstream packets below a specified maximum. This
problem can be mathematically formulated as follows:
minimize

Ecycle

subject to Davg Dcons ,

(1)

Nactive wl Ntotal wl

where parameters Ecycle , Davg , Dcons , Nactive wl , and Ntotal wl


represent the energy consumption of the network per cycle,
the average delay of upstream packets, the delay constraint,
the number of active wavelengths, and the total number of
wavelengths in the network, respectively.
Following the general equation on average delay reported
in [12], we can approximate the Davg of a TWDM-PON as
follows:
Tpoll
+ NTpoll + RT T,
(2)
2
where N is the number of cycles the upstream packets have
to wait in the ONU queue after they are reported to the OLT,
before the OLT allocates any bandwidth for their upstream
transmission. Parameter RT T represents the round trip time
of the network. In general, a longer Tpoll allows an ONU to
sleep for longer, and as a result, saves more energy. However,
Tpoll cannot be increased arbitrarily as it increases the average
delay. The maximum energy-savings at the ONUs is achieved
if the network operates at the maximum polling cycle time,
Tpoll max , that satisfies a given Dcons . Using Eq. (1) and (2), we
can mathematically formulate the relationship between Tpoll
and Dcons as follows:
Davg

Tpoll
+ N Tpoll + RT T Dcons
2
Dcons RT T
Tpoll
,
N + 12

(3)

According to Eq. (3), the maximum value of Tpoll , Tpoll max ,


that satisfies a given Dcons is given by the minimum value of N.
The minimum value of N depends on the type of the algorithm
deployed. In this work, we have used the same offline DBA
process used in [8] and [13] as it ensures fairness among
ONUs in bandwidth allocation. Based on this offline DBA, the
ONUs inform the OLT of their bandwidth requirement using
REPORT message. The OLT waits until it receives REPORT
messages from all ONUs in the PON and then calculates the
average bandwidth allocated to an ONU. Under the OFFDWBA algorithm, upstream traffic has to wait a complete
polling cycle after the REPORT message is sent to the OLT
before any bandwidth is allocated to them. As a result, the
minimum value of N in this case is 1. The Tpoll max of the
OFF-DWBA, therefore, can be approximated as follows:

5019

IEEE ICC 2015 - Optical Networks and Systems Symposium

2 (Dcons RT T )
,
(4)
3
Under the proposed OFF-DWBA algorithm, the network operates at this Tpoll max .
Figure 2 illustrates the flow chart of the OFF-DWBA
algorithm executed at the OLT of a network operating at
Tpoll max . The OLT waits until it receives REPORT messages
from all ONUs in the TWDM-PON. Once it receives all the
REPORT messages, the OLT calculates the average bandwidth
requested, BWavg , by an ONU. Under the assumption that all
ONUs in the network operate at the same network load, it is
reasonable to calculate BWavg . Once BWavg is calculated, the
OLT initiates the wavelength optimization process. The OLT
first calculates the maximum allowable bandwidth, BWmax , of
a TWDM-PON operating with a single wavelength. BWmax
corresponds to the maximum allowable bandwidth allocated
to an ONU operating under Tpoll max . Parameter Tprocess in Fig.
2 represents the processing time of the ONUs. If the BWavg
BWmax , a new wavelength is introduced to the network and
the OLT calculates the new BWmax of a TWDM-PON with
two active wavelengths. This BWmax is again compared with
the BWavg , and this process continues until BWavg is lower than
Tpoll max

BWmax . At this point, the Nactive wl is taken as the optimized


number of active wavelengths required to keep the upstream
packet delay below a specified maximum.
Once the wavelength optimization is complete, the OLT
initiates the sleep/doze allocation process. As shown in Fig.
2, the OLT first calculates the idle time, Tidle , of an ONU.
The parameter Tslot represents the period that corresponds to
BWavg . Once Tidle is calculated, the OLT decides on whether
an ONU is transitioned into doze, sleep, or active state as
follows. Let sleep-to-active transition time and doze-to-active
transition time of an ONU be Tsta and Tdta , respectively. If Tidle
Tsta , the ONUs are transitioned into sleep mode and if Tsta
Tidle
Tdta , the ONUs are transitioned into doze mode. If
neither of these two conditions are satisfied, the ONUs remain
active.
It is important to note that we have considered only upstream traffic and delay constraints in our algorithm. Had
we considered the downstream traffic as well, we have to
take the minimum of the delay constraints in upstream and
downstream. Further, the upstream and downstream traffic
should be synchronized as proposed in [8].
III. S IMULATIONS AND RESULTS
TABLE I: Network and protocol parameters
Parameter
Network reach
Number of wavelengths
Number of ONUs
Delay constraints
Propagation delay
Inter-frame gap in upstream
Average Ethernet packet size
Wavelength tuning and GATE processing

Waits for REPORT messages from


each ONU

Calculate average bandwidth requested by an ONU


BWavg = Total BW requested/Nonu

Calculates BWmax of a TWDM PON with Nactive wl wavelengths


BWmax = (Tpoll max - Tprocess)/(Nonu/Nactive wl)

No

Parameter
Doze-to-active transition time (VCSEL) [11]
Sleep-to-active transition time (VCSEL) [11]
Power consumption VCSEL (active) [11]
Power consumption VCSEL (doze) [11]
Power consumption VCSEL (sleep) [11]
OLT TRX (active) [7]
OLT base power (EDFA Preamp +
Booster + L2 switching capacity) [7]

Take the current Nactive wl as


the number of active
wavelengths

No

Nactive wl < Ntotal

TABLE II: Power consumption and switching values of 10GVCSEL-ONUs and OLT

Yes

BWavg < BWmax

Yes
Calculate Tidle
Tidle = Tpoll max - Tslot

Nactive wl = Nactive wl + 1
Tidle >
Tsta

Yes

Enter sleep mode for


Tidle - Tsta

No

Tsta>Tidle >Tdta
No

Yes

Enter doze mode for


Tidle - Tdta

Value
40 km
4
64
7.5, 10, and 15 ms
200 s
1s
791 bytes
50s

Remains active

Creates GATE message with Sleep/Doze time,


Sleep/Doze start time, and Sleep/Doze command

Fig. 2: Flow chart of the offline DWBA algorithm executed at


the OLT.

Value
330 ns
2 ms
3.984 W
3.85 W
0.75 W
11 W
64 W

The simulations for the proposed OFF-DWBA algorithm


were performed using C++. We have simulated a 40 km
TWDM-PON with 64 ONUs and four wavelengths. Table
1 lists the network and protocol parameters of the TWDMPON used in our simulations. Table 2 lists the power consumption and switching values of the VCSEL-ONU and the
OLT considered in this work. First, we verify Eq. (4) using
simulations as the validity of Tpoll max depends on this equation.
In this initial simulation, we have considered processing and
tuning times to be negligible. Figure 3 plots the average delay
of the network under the OFF-DWBA algorithm for a Tpoll
of 10 ms. As shown in Fig. 3, the average delay of the

5020

IEEE ICC 2015 - Optical Networks and Systems Symposium

16
14
Time (ms)

12
10
8
6
4

Polling cycle time

Offline DBA

0
0

0.2

0.4
0.6
Normalized network load

0.8

Fig. 3: General architecture of a TWDM-PON.

OFF-DWBA algorithm is approximately 1.5 times the polling


cycle time. The minor difference between the theoretical and
simulated values is due to the fact that in our equations, we
have not considered the time, which is in nano second range,
it takes a packet to reach the beginning of the ONU queue.
This result therefore, verifies the calculation of Tpoll max in our
OFF-DWBA algorithm. The rest of this section discusses and
compares the numerical and simulation results of the OFFDWBA algorithm for Dcons of 7.5 ms, 10 ms, and 15 ms.
Figure 4 plots the average delay of the OFF-DWBA algorithm for Dcons of 7.5 ms, 10 ms, and 15 ms. It is important
to note that the simulated values of average delay for each
Dcons is slightly higher than the Dcons . As explained before,
we have not considered the time it takes for a packet to reach
the beginning of the ONU queue. As a result, the simulated
average delay is not exactly 1.5 times the polling cycle times.
Had we taken the simulated value of 1.6, it would have resulted
in lower Tpoll max and lower average delay. However, we have
considered the factor of 1.5 in our simulations, resulting in
this reported increase in average delay.

Figures 5, 6, and 7 compare the simulation results of the


OFF-DWBA against the numerical results reported in [9].
Based on these plots, we can verify the performance of
the OFF-DWBA in terms of the optimum number of active
wavelengths, the percentage of energy-savings at the ONUs,
and the percentage of energy-savings at the OLT.
Figures 5 (a) and (b) plot the number of active wavelengths
as a function of normalized network load based on simulations
and numerical analysis, respectively. In both simulations and
numerical analysis, for any given Dcons , when the network load
increases, the number of active wavelengths increases. When
the network load increases, BWavg increases. As explained in
section II, when BWavg exceeds BWmax , a new wavelength
is introduced to the network. As such, the number of active
wavelengths increases with the increase in network load. It
is also important to note that when the Dcons increases, the
same traffic could be transmitted with lower number of active
wavelengths. For example, when the network load is 0.4,
a network with a Dcons of 15 ms requires only 1 active
wavelength while a network with a Dcons of 10 ms requires 2
active wavelengths to support the same traffic. When Dcons

Number of active wavelengths

18

3.5
3
2.5
2
1.5
Dcons = 15 ms

Dcons = 10 ms

0.5

Dcons = 7.5 ms

0
0

0.2

0.4
0.6
0.8
Normalized network load

1.2

1.2

(a) Simulations
3.5
Number of active wavelengths

18
Average delay (ms)

16
14
12
10
8
6

Dcons = 15 ms

Dcons = 10 ms

Dcons = 7.5 ms
0.2

0.4
0.6
0.8
Normalized network load

2.5
2
1.5
1

Dcons = 15 ms
Dcons = 10 ms
Dcons = 7.5 ms

0.5
0

0
0

1.2

0.2

0.4
0.6
0.8
Normalized network load

(b) Numerical analysis


Fig. 4: Average delay as a function of normalized network
load.

Fig. 5: The number of active wavelengths reported in (a)


simulations and (b) numerical analysis

5021

IEEE ICC 2015 - Optical Networks and Systems Symposium

75

+,-.,/012,"34",/,-25"6178/26"9,-"
.5.:,"10"0;,"<=>"?@A"

Percentage of energy-savings per


cycle per ONU (%)

80
70
65
60
55
50
45

Dcons = 15 ms

40

Dcons = 10 ms

35

Dcons = 7.5 ms

30
0

0.2

0.4
0.6
0.8
Normalized network load

1.2

&#"
&!"
%#"
%!"
$#"
H.3/6"I"$#"C6"

$!"

H.3/6"I"$!"C6"

#"

H.3/6"I"J'#"C6"

!"
!"

!'%"

(a) Simulations

$"

$'%"

$"

$'%"

(a) Simulations

80
75
70
65

+,-.,/012,"34",/,-25"6178/26"
9,-".5.:,"10"0;,"<=>"?@A"

Percentage of energy-savings per


cycle per ONU

!'("
!')"
!'*"
B3-C1:8D,E"/,0F3-G":31E"

60
55
50
45

Dcons = 15 ms
Dcons = 10 ms
Dcons = 7.5 ms

40
35
30
0

0.2

0.4
0.6
0.8
Normalized network load

1.2

(b) Numerical analysis

&!"
%#"
%!"
$#"
H.3/6"I"$#"C6"
H.3/6"I"$!"C6"
H.3/6"I"J'#"C6"

$!"
#"
!"
!"

Fig. 6: Percentage of energy-savings per cycle per ONU


reported in (a) simulations and (b) numerical analysis
increases, the corresponding increase in BWmax allows the
same amount of traffic to be transmitted with a lower number
of active wavelengths.
Figures 6 (a) and (b) plot the percentage of energy-savings
per ONU per cycle as a function of normalized network load
based on simulations and numerical analysis, respectively. The
percentage of energy-savings is calculated as the proportion
of energy-savings achieved using a sleep/doze mode VCSELONU compared to an always-active VCSEL-ONU. In both
simulations and numerical analysis, for a given Dcons , the
percentage of energy-savings decreases with the increase in
network load. When the network load increases, it also increases Tslot . As the increase in Tslot increases Tactive , the energy
consumption increases, and thereby decreases the percentage
of energy-savings. It is important to note that the percentage of
energy-savings of more than 60% reported in this work is due
to the ONUs entering into sleep mode for the Dcons considered
in this work. The large difference in power consumption, 2.134
W, between active and sleep modes of a 10G-VCSEL-ONU
results in maximum energy-savings of 64%, 56%, and 47%
for Dcons of 15 ms, 10 ms, and 7.5 ms, respectively. When
the Dcons increases, it allows the ONUs to sleep for a longer
duration, thus saves more energy.
Figures 7 (a) and (b) plot the percentage of energy-savings

&#"

!'%"

!'("
!')"
!'*"
B3-C1:8D,E"/,0F3-G":31E"

(b) Numerical analysis


Fig. 7: Percentage of energy-savings at the OLT per cycle
reported in (a) simulations and (b) numerical analysis
at the OLT as a function of normalized network load based
on simulations and numerical analysis, respectively. The percentage of energy-savings is calculated as the proportion of
energy-savings achieved by switching off idle wavelengths,
compared to having all four wavelengths active. In both
simulations and numerical analysis, for a given Dcons , when
the network load increases, the number of active wavelengths
increases as discussed in Fig. 3. As a result, the percentage of
energy-savings decreases with the increase in network load.
As explained before, higher Dcons requires a lower number
of active wavelengths to support the same network traffic and
therefore results in more energy-savings at a given network
load. It is important to note that irrespective of the network
load, an OLT requires a base power of 64 W. As a result, even
when only a single wavelength is active, the OLT consumes
75 W of power, resulting in a maximum percentage of energysavings of only 30 % at the OLT.
IV. C ONCLUSION
In this work, we have simulated a previously proposed theoretical framework of the energy-efficient OFF-DWBA algo-

5022

IEEE ICC 2015 - Optical Networks and Systems Symposium

rithm to verify its performance in a practical delay-constrained


TWDM-PON. The algorithm provides a general framework
that can be used for TWDM-PONs with different parameters
such as delay constraints and number of wavelengths. In this
work, we have provided a comprehensive analysis of the
OFF-DWBA algorithm and have compared its simulation and
numerical results. The simulation results verify our previous
claim that the proposed OFF-DWBA algorithm results in
significant energy-savings both at the OLR and the ONU. Most
importantly, our simulation results indicate that under the OFFDWBA algorithm, the average delay of upstream packets does
not exceed the specified maximum. As a result, the proposed
energy-efficient OFF-DWBA algorithm is suitable to provide
delay-sensitive services over a TWDM-PON.
R EFERENCES
[1] A. Dixit, S. Lambert, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester,
Toward energy efficiency in optical access networks, Proc. of IEEE
International Conference on Advanced Networks and Telecommuncations
Systems (ANTS), 2013.
[2] Y. Luo, M. Sui, and F. Effenberger, Wavelength management in time
and wavelength division multiplexed passive optical networks (TWDMPONs), Proc. of IEEE Global Communications Conference (GLOBECOM), 2012.
[3] H. Yang, W. Sun, J. Li, and W. Hu, User migration in time and
wavelength division multiplexed PON, Proc. of 15th IEEE Transparent
Optical Networks (ICTON), We.B3.2 (2013).
[4] N. Cheng, L. Wang, D. Liu, B. Gao, J. Gao, and X. Zhou, Flexible TWDM PON with load balancing and power saving, Proc. of
39th European Conference and Exhibition on Optical communication
(ECOC), We.3.F.6 (2013).
[5] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester, Novel
DBA algorithm for energy efficiency in TWDM-PONs, Proc. of
39th European Conference and Exhibition on Optical communication
(ECOC), 2013.
[6] A. Dixit, B. Lannoo, D. Colle, M. Pickavet, and P. Demeester, Flexible
TDMA/WDMA passive optical network: energy-efficient next-generation
optical access solution, J. Optical switching and networking, 10(4),
491506(2013).
[7] M. P. I. Dias, D. P. Van , L. Valcarenghi, and E. Wong,Energy-efficient
dynamic wavelength and bandwidth allocation algorithm for TWDMPONs with tunable VCSEL ONUs, Proc. of IEEE Optoelectronic and
Communications conference (OECC), 2014.
[8] M. P. I. Dias, B. S. Karunaratne, and E. Wong,Bayesian estimation and
prediction based dynamic bandwidth allocation algorithm for sleep/doze
mode passive optical networks, J. of Lightwave technology, 32(14), 2560
2568(2014).
[9] M. P. I. Dias, D. P. Van , L. Valcarenghi, and E. Wong, Energy-efficient
TWDM-PON with VCSEL ONUs, Asia Communications and Photonics
Conference (ACP), ATh2F.4, (2014).
[10] E. Wong, M. Muller, and M.C. Amann, Characterization of energyefficient and colorless ONUs for future TWDM-PONs, OpticsExpress,
21(18), 20747-20761 (2013).
[11] E. Wong, M. Muller, P.I. Dias, C.A. Chan, and M.C. Amann, Energyefficiency of optical network units with vertical-cavity surface-emitting
lasers, OpticsExpress, 20(14), 14960-14970 (2012).
[12] G. Kramer, Ethernet passive optical networks, McGrow-Hill (2005).
[13] M. P. I. Dias and E. Wong,Sleep/doze controlled dynamic bandwidth allocation algorithms for energy-efficient access networks, OpticsExpress,
21(8), 9931-9946 (2013).

5023