energy saving scheme based on multi-modes hybrid dynamic ...dynamic bandwidth optimization for...

Energy Saving Scheme Based on Multi-modes Hybrid

Dynamic Bandwidth Optimization for Software Defined

Distribution Optical Networks

Yang Cao1, Hui Yang

2, and Xiaoxu Zhu

2

1Electrical Power Research Institute, China Southern Power Grid, Guangzhou, 510080, China

2Beijing University of Posts and Telecommunications, Beijing 100876, China

Email: [email protected]; [email protected]

Abstract—As the increasing number of access devices, and the

complex and changeful of access site and the media, energy

consumption of power distribution communication access

networks is expected to further grow in the future, energy

saving has become a key issue with the development of Passive

Optical Networks (PONs). In this paper, we propose a Software

Defined Distribution Optical Network (SD-DON) architecture,

design an energy saving aware control strategy which

support unified optimization and efficient scheduling

from multi-domain perspective. Additionally, a Multi-

Modes Hybrid Dynamic Bandwidth Allocation (MH-DBA)

scheme are proposed and evaluated under SD-DON, in which,

4-saving modes and multi-period mixed adaptive dormancy

method is designed to maximize energy saving of Optical

Network Units (ONUs). Simulation results show that MH-

DBA can significantly save energy under different traffic

loads, and clearly outperform the conventional scheme.

Index Terms—SD-DON, energy saving, multi-domain, MH-

DBA, ONU, dormancy

I. INTRODUCTION

A. Background and Motivation

As the increase of power telecommunication network

and bandwidth, the power distribution network has

gained unprecedented development. Passive Optical

Networks (PONs) have been widely considered as the

most favorable power distribution network architecture to

cater for the growing demand of bandwidth-hungry

applications. Even though PONs are considered more

energy-efficient compared to other wired access

technologies, such as xDSL, due to their massive

deployment nowadays and even more in the future, it is

desirable to further reduce their energy consumption [1].

Software Defined Networking (SDN) is an emerging

technology used to provide more flexible controls on

Manuscript received May 20, 2016; revised December 13, 2016.

This work has been supported by China Postdoctoral Science Foundation (No.2016M602557), NSFC project (61501049), the

Fundamental Research Funds for the Central Universities (2015RC15),

Fund of State Key Laboratory of Information Photonics and Optical Communications (BUPT), P. R. China (IPOC2015ZT01), and funded by

SKL of Advanced Optical Communication Systems Networks, China. Corresponding author email: [email protected]

doi:10.12720/jcm.11.12.1028-1035

networks by abstracting lower-level switch functionality

[2], [3]. By separating the control and data planes and

logically centralizing the traffic management intelligence,

the controller directly manages SDN aware devices via a

well-defined interface and control protocol. The central

control paradigm of SDN enables the controller to utilize

network resources efficiently and facilitate QoS

differentiation for services through network-wide

optimization [4].

B. Prior Art and Limitation

Many successful efforts have used SDN to improve

access network performance [5]-[8]. Reference [9]

presented an architectural design and system

implementation for a GPON based OpenFlow - enabled

SDN virtual switch. By combining the DBA and

bandwidth control of the add-on OpenFlow switch, their

virtual switch can control bandwidth usage accurately and

dynamically. Reference [10] proposed possibilities for

integrating OpenFlow in GPON, in which an OpenFlow

controller directly performs the DBA function. In

addition, ONUs would have to be enhanced to provide

OpenFlow-like matching and processing functions. A

detailed implementation of the system is not addressed in

that work.

Furthermore, a variety of technologies have been

proposed for improving the energy efficiency of PONs

[11]-[15]. Enabling Optical Network Units (ONUs) to

enter sleep mode for energy saving is considered one of

the most effective methods [16]-[22]. In particular,

Reference [11] proposes a parametric extension allowing

the unified power management mode, in which an ONU

periodically turns off both its receiver and transmitter, as

in the cyclic sleep mode, and performs infrequent

bidirectional handshakes, as in the doze mode. The timely

reaction to an external stimulus is ensured by periodic

unidirectional handshakes when only the ONU receiver is

turned on. The unified mode thus combines the

advantages of the two standardized power saving modes,

and a system that supports the unified mode can emulate

either the cyclic sleep or doze behavior as a special case.

Simulation results show the energy efficiency superiority

of the unified mode over either of the two standardized

1028

Journal of Communications Vol. 11, No. 12, December 2016

©2016 Journal of Communications

modes in general. Reference [18] presents an energy

efficient medium access control protocol exhibiting

intelligent switching among ONU transceiver power

levels. This is implemented by introducing a new

methodology at the ONU active free state of NG-PONs

based on the adaptation of standard local sleep and local

doze queue indicators to a more comprehensive grated

mode selection policy. Computer simulations have shown

the superiority energy efficiency of this study. Reference

[23] proposed a QL (queue length) -based proportional

and proportional-derivative controllers, effectively reduce

the power consumption of ONUs while maintaining the

downstream queuing delay at a constant level. Besides,

many Dynamic Bandwidth Allocation (DBA) were

usually just used inside a single Optical Line Terminal

(OLT), lacking flexible bandwidth configurations among

multiple OLTs. As a result, a huge waste of network

resources exists in idle areas in all working devices.

What’s more, the majority of traditional schemes realize

bandwidth allocation according to terminal bandwidth

requirements and pre-established 3-saving mode (active,

doze and sleep) switching policies [24], [25]. However, in

the current 3-saving mode DBA method (3M-DBA),

when the end point of the DBA period is reached,

regardless of whether there is a large data traffic, ONU

will always be awakened. So a 4-saving modes (active,

doze, light sleep, deep sleep) and multi-periods mixed

adaptive dormancy method is needed to maximum energy

saving efficiency.

C. Proposed Approach and Key Contributions

This paper designs a Software Defined Distribution

Optical Network (SD-DON) framework. Control

mechanism includes multi-domain scheduling, dormancy

awareness and bandwidth allocation. It is beneficial to

simplify the control process, improve network

management and maintenance level, and promote the

greenization of the distribution networks. Besides, a

novel multi-modes hybrid dynamic bandwidth allocation

algorithm (MH-DBA) is first proposed under SD-DON

architecture. One of the challenges is to specify a

scheduling order in which data transmission, light/deep

sleep period, and doze period are scheduled. Moreover,

ONU sleep/doze time should be carefully determined so

that energy-savings can be achieved whilst maintaining

the required Quality of Service. Simulation results show

that MH-DBA under SD-DON can significantly save

ONU energy, provide lower latency under high traffic

loads, and can ensure the communication quality of the

power system.

The next sections arranged as follows. Section II

describes the SD-DON architecture. Section III describes

our definition for four kinds of ONU energy saving mode,

and the MH-DBA scheme in detail. Section IV gives our

simulation results and analysis. In Section V, we will

conclude the paper.

II. SD-DON ARCHITECTURE

EPON have been widely considered as the most

favorable access network technology of its low cost, easy

to use and easy to update to cater for the growing demand

of bandwidth-hungry applications. Combine PON

technology, architecture of SD-DON from perspective of

multi domain convergence and integration control is

illustrated in Fig. 1. Consisting of control layer, 4-level

backbone communication network layer and access

network layer. The distribution optical access network

refers as underlying network devices including OLT,

splitter and ONU. These ‘goofier’ hardware only need to

transmit data and execute strategies, with the controller

implementing the analysis and management about

network strategies instead. The access layer provides the

control layer with programmable interfaces based on

OpenFlow protocol, which is the standard communication

protocol. Domain controller can be integrated into the

OF-OLT. Convergence occurs in gateways on 4-level

electric power backbone network layer between control

layer and access layer, achieve the information intelligent

collection and gathering. The application layer can realize

more complex functions using the north application

programming interface (API) on the control layer. Based

on the proposed SD-DON architecture, the key

technologies of ONU sleep and energy saving are

proposed in this paper.

4- level electric power backbone network

Power

convergence

ring

OF-OLT OF-OLTOF-OLT

ONU

Multi domain

controller Application

D1

D2

D3

Distribution Optical Networks

Controll layer

Access layer

Fig. 1. Architecture of SD-DON

Based on the proposed SD-DON architecture, the

function and the mutual cooperation relationship of each

control module are shown in Fig. 2. The innovation lies

in adding the Adaptive Dormancy Control Module

(ADCM) and the Multi Domain Control Module (MDCM)

to realize the multi domain ONU energy saving strategy.

In control plane, the Traffic Monitoring Module (TMM)

collects the statistics of services' traffic. MDCM balanced

the allocation of the whole network OLTs bandwidth/

traffic. On receiving the flow state, the traffic scheduling

policy of the Dynamic Bandwidth Allocation Module

(DBAM) will allocate bandwidth according to the

1029



required time slot of different services. According to the

dormancy strategy, the slot calculation unit of ADCM

calculate each ONU’s slot distribution and slot energy-

saving state (active/ doze/ light sleep/ deep sleep).

Operation Administration and Maintenance (OAM)

module feedback to DBAM to adjust each ONU

bandwidth slot allocation in order to meet the

requirements of energy saving effect. Particularly,

operators will deploy this energy-saving policy module

into controller in the form of plug-in, and it will make

convenient for control strategy expansion and

customization.

Data/ forwarding plane describes the exchange process

of OLT and ONU. In OF-OLT, the downstream (DS)

traffic is generated by the receiver in the Tx/Rx module

and broadcast to all connected ONU. For the upstream

(US), the Tx/Rx module can be used to receive traffic and

allocate bandwidth slots with different granularity for the

corresponding ONU. OpenFlow protocol agent is

Embedded in the OF-OLT maintaining the information

flow table, and simulate OLT related control information

(e.g. ONU sleep state), the software content is mapped as

the hardware (transmitter and receiver internal switchgear)

control and adaptation. Therefore, the US and DS link

can be virtualized as a logical flow. Similarly, OpenFlow

protocol agent is also embedded in the ONU, ONU

interact with OF-OLT through signaling, directly control

transmitter and receiver internal switchgear.

ADCM DBAMMDCM

Plug-In OAM TMM

Flow_mod sleep switch

receiver

Flow_mod

spll

iter

OF multi domain controller Controll plane

OpenFlow

transmitterTx/Rx

module

sleep switch

receiver

transmitterTx/Rx

module

OF-OLT ONUi

Data

Input

Data planeONUn

...

signaling

exchange

Function structure of power distribution unit

Fig. 2. Function structure of power distribution

III. MH-DBA SCHEME

A. Multi Modes and Periods Hybrid Concept

To maximize energy efficiency, the ONU sleep time

should be extended as much as possible. We design 4-

saving modes (active, doze, light sleep, deep sleep) and

multi-periods mixed adaptive dormancy method, and

assume two DBA cycles (2T) as a polling period.

Definitions are elaborated as follows:

1) Sleep threshold

The limit below which ONU may enter light sleep

mode. It includes cache threshold (Thrcache) and time

threshold (Thrtime). If data in report and queue cache is

smaller than Thrcache during 1th

Thrtime, ONU may enter

light sleep mode. If after 2nd

Thrtime, data cache is still less

than Thrcache, ONU may enter deep sleep mode.

2) Sleep command

OLT assign ONU to enter sleep mode with this kind of

GATE.

3) Awake state

An interim mode, when the clock (ONUclk) of sleep

control achieves sleep interval, ONU enters this mode

spontaneously receiving the trigger from the sleep control.

4) Active interval

The periods ONU in active mode. In active interval, all

models of ONU are active. It receives normal gate and

DS data from OLT and transmits data according to the

normal gate's start time and length.

5) Doze interval

The time ONU in doze mode. In doze interval, ONU

transmitter close off, and stops transmitting data. ONU

also maintains a timer to calculate the doze period which

is specified by OLT. Keep the part function active of the

receiver. The data which arrives during the doze period

will be temporally stored in ONU's buffer. When ONUclk

sleep control arrives a DBA cycle, T, ONU will always

be awaked.

6) Light sleep interval

The time ONU in light sleep mode. Stop all function of

user interface, optical receiver and transmitter. ONU

cannot receive or send any traffic. Similar to doze state.

When ONUclk arrives T, ONU will always be awaked,

and start to transmit data stored in buffer.

7) Deep sleep interval

The same as light sleep interval, ONU cannot receive

or send any traffic, but ONU may sleep for a polling

period which is 2T. When ONUclk sleep control arrives

polling periods, 2T, ONU will always be awaked, and

start to transmit data stored in buffer according to the

sleep gate's start time and length.

B. MH-DBA Scheme Operation

The MH-DBA protocol operation for intra-domain is

illustrated in Fig. 3. For US, OLT calculate the time and

US bandwidth when the US data begin to transmit in the

next cycle. Then, OLT bond the time and the bandwidth

together to constitute a GATE, and put it into the GATE

queue. For DS, The MH-DBA performs dynamic

scheduling where the OLT waits for all the report

messages from ONUs before computing grants for the

next cycle. Before assign the bandwidth for an ONU,

OLT should check the GATE queue. If there is GATE

waiting to transmit, OLT assign the time slot for all

GATE first, then calculate the start time and bandwidth

for the ONU.

Meanwhile, ONU could not sent Report package

utilizing SDN, in order to compact time slots to reduce

bandwidth resource occupancy of OLT. OLT allocates

data slot duration as the minimum between US and DS

buffer backlogs. The OLT first sends a GATE to an ONU.

When the ONU receives the GATE, both US and DS data

transmissions take place. When the data slot duration

1030



expires, the ONU enters energy saving mode until the

scheduled time for the next GATE message. The OF-

OLT determines the scheduled time for a GATE by

predicting the total transmission time of all other ONU.

What’s more, to maximize energy efficiency, a preferable

way is to overlap the DS and US transmissions as much

as possible. With this compact sizing policy, Duration of

a transmission slot (Txlen) is computed as:

Txlen = max {Bds, Bus} (1)

where Bds, Bus denotes the bandwidth of DS and US, and

ensures that both OLT and ONU have sufficient

bandwidth for both transmission directions.

When incorporating four energy saving modes into the

operations of a PON system, a scheduling scheme must

specify where to insert the doze period and/or light, deep

sleep period into an existing order of data transmission,

and GATE message. Therefore, the scheduling order

becomes a challenge.

This paper implements MH-DBA schemes as shown in

Fig. 4. If monitoring large bandwidth requests, all models

of ONU remain active. Otherwise, according to Thrtime

that set by controller, judge ONU's energy saving state.

We assume two DBA cycle as a polling period. If the

data in report and cache is smaller than the Thrcache within

the time, ONU may enter doze mode. If exceeding Thrtime,

there are still little data requests or smaller uplink traffic,

ONU may enter light sleep mode, and the sleep interval is

T. Otherwise, inactive state exceeding a DBA period T,

ONU may enter deep sleep mode of the sleep interval 2T.

When the timer reaches the sleep interval, ONU

spontaneously goes to wake mode. ONUs in state doze or

light sleep will judge again whether remain the original

mode. When the timer reaches polling period (2T), ONUs

start to transmit data stored in buffer according to the

sleep gate's start time and length.

light

light

OLT

ONU2

D1 D2

ONU1

ONU3

U2

D4

G

G

G

G

GU1

D2

D1

U1

G

U1 U2 U4

& :Awake & sleep

light

light light

light

deep

deep

:Active :Doze :Light sleep

G

US

DS D1

:Deep sleep

U1

D4U4

D1U2

U2

D2

light

light

G

G D1

U1

G

G U3

D3

light

light

light

US

DS

US

DS

Current polling cycle Next polling cycleDBA DBA NextDBA

Tx

Rx

D1

ONU4G U4

D4

Glight

lightUS

DS

U4

D4

light

light

G

D1

U1

D1

U1

light

light

U1

D2

U1

D1

T T2T

D3

U3

D1

U1

light

light

light

Fig. 3. Scheduling order of MH-DBA scheme operation

start

Controller judges the sleep modes

0：doze mode

1：light sleep mode

2：deep sleep mode

Controller set a time Thrtime t0 for the

ONU, (t0<T, a DBA polling cycle is T)

0：t<t0，little data

requests or US data

traffic, little

Thrcache, ONU

transmitter closed.

1：t0<t<T，little data

requests or US data

traffic, little Thrcache,

ONU transmitter and

receiver closed.

2：t>T，little data

requests or US data

traffic, Thrcache,

ONU transmitter

and receiver closed.

Keep

the

original

mode

During T, monitor

Large bandwidth

request?

ONU awake and

switch to active

No

Yes Arrive at

sleep interval

2T?

Keep deep

sleep

Yes

No

Arrive at sleep

interval T?

Yes

Monitor large

bandwidth requests？No

Yes

ONU awake,

end

Arrive at sleep

interval 2T?

YesNo

N

o

Fig. 4. Illustration of MH-DBA scheme operation

C. MH-DBA under SD-DON

DBA algorithm is one of key technologies of EPON.

The MH-DBA algorithm we propose, means adopting

multi-periodic polling to assign bandwidth in both

upstream and downstream direction. When power service

trigger, ONU mapping as service request. The flow chart

of MH-DBA under SD-DON Implement and definitions

of some parameters are described in Table I. We take DS

DBA for instants.

In general, MH-DBA is divided into two types, inter-

domain and intra-domain, which are described as step2

and step 3. For inter-domain. Each domain OLTs

gathering ONU service requests and ONU network status

report to multi domain controller. According to the OLT

queues cache in per domain and prediction results,

controller evaluate the traffic demand in next T cycle.

Then it allocate OLT bandwidth as principle of reduce

free domain OLTs bandwidth, as well as prior to occupy

busy domain. In each domain, the OLT has to store

downstream data to separate cache for each ONU first,

and then poll every ONU and assign the bandwidth

according to the caches. Take instants, OLT collect all

1031



ONUs report in advance according to the different types

of ONU request in the OLT feedback and the length of

the reported message (caches). Weighting and

convergence, calculate the length of time slot and the

upstream bandwidth for each ONU, and then return to the

domain controller. For intra-domain. An upstream or

downstream DBA cycle means the summation of

transmission time slot for each ONU when OLT has

polled all ONUs once. Because the time slot for each

ONU is related to the report in upstream, the cache in

downstream, the cycles are different. The calculation of

upstream and downstream transmit time slot is associated

to each other. Particularly, step 4 utilizes the dormancy

strategy we presented above, the slot calculation unit of

ADCM calculate each ONU’s slot distribution and slot

energy-saving state (active/ doze/ light sleep/ deep sleep).

In step 5, 6 and step 7, OF-OLT function module receive

OF south interface message, maintain the information

flow table. ONU interact with OF-OLT through signaling,

and start ONU sleep-state-switch module to make the

corresponding action, due to the active period, doze

period and/or light, deep sleep period, directly control

transmitter and receiver internal switchgear. Do the

bandwidth allocation according to the scheduling order of

data transmission. Besides, maximum allowable delay

TDmax is derived directly through controller which is

alterable in step8. If TDret>TDmax, then service block. Else,

service succeed.

TABLE I: FLOW CHART OF MH-DBA UNDER SD-DON

TDmax, denotes the maximum allowable delay set by controler; TDret, denotes the retransmission latency;

TBavailable, denotes the US available bandwidth of OLT;

UBrequest, denotes the US bandwidth request of ONU;

Step1: ONU mapping as service request.

Step2: For inter-domain. Each domain OLT gathering ONU

service requests and network status report to multi domain controller. Controller evaluate the traffic in next T cycle due to the

OLT queue cache and prediction results per domain, then allocate

OLT bandwidth as principle of reducing free domain OLT bandwidth, and prior to occupy busy domain. If TBavailabl>UBrequest,

then go to step 3, else, jump to step 6.

Step3: For intra-domain. OLT collect all ONUs report in

advance according to the different types of ONU request in the

OLT feedback and the length of the reported message. Weighing and convergence, calculate the length of time slot and the US

bandwidth for each ONU, and then feedback to multi domain controller.

Step4: According to currently network-wide energy saving state,

DBAM start the sleep slot calculation unit of ADCM to calculate the next time slot saving state.

Step5: OF-OLT function module receive of south interface message, open/close OLT transceiver due to the calculated energy

saving state.

Step6: OLT and ONU interact by signaling, start ONU sleep-state-switch module to make the corresponding action.

Step7: Bandwidth allocation. If succeed, then jump to step9, else return null.

Step8: Start latency retransmission mechanism, controller

update TDmax. If TDret>TDmax then service block, jump to Step 9, else, jump to Step 2.

Step9: Service processing end.

IV. EXPERIMENTAL RESULTS AND ANALYSIS

We evaluate the effectiveness of the proposed MH-

DBA by means of extensive simulations of a PON with

one OLT and eight ONUs per domain using C-based

discrete-event simulator OPNET 14.5. More values of

evaluation parameters are shown in Table II. The physical

distance between the OLT and ONUs is set randomly

following a uniform distribution between 15km and 20km.

Both US and DS channels operate at a data rate of 1Gbit/s.

The energy consumption of an ONU in the active mode,

doze mode, light sleep mode and deep sleep mode are

equal to 3.75W, 1.70W, 1.08W and 0.40W, respectively.

Upon bandwidth allocation, the OLT sets the length of a

given cycle based on the allocated bandwidth and the

start time, which may be adjusted dynamically by

controller within a permissible range.

TABLE II: VALUES OF EVALUATION PARAMETERS

Parameters Values

DS and US data rate 1Gbit/s

Data frame size 1250bytes

Data buffer size 4MB

Number of domains (N) 3

Number of ONUs (N) per domain 8

ONU power consumption(Pa, Pd, Pls and Pds)

3.75W, 1.70W, 1.08W, 0.40W,

physical distance between the OLT and ONUs

Uniform distribution between 15km and 20km

In the following, the performance metric used in the

evaluation is average energy-savings.

( )d d ls ls ds ds

a a d d ls ls ds ds

P t P t P t

P t P t P t P t

(2)

where Pa, Pd, Pls and Pds are the ONU power

consumption in active, doze, light sleep, and deep sleep

states; ta, td, tls and tds are the average time an ONU

sojourns in each state within an operating cycle.

In the case of limited cache, the system will appear the

packet loss. Firstly, we consider packet loss in the

simulation to evaluate the performance of all algorithms

in terms of average energy saving ratio, average latency

and packet loss ratio.

By comparing the proposed MH-DBA and

conventional 3M-DBA scheme in Fig. 5, the presented

results clearly show that MH-DBA has a superior

performance in terms of energy saving of ONUs. For 3M-

DBA, when the end point of the DBA period is reached,

ONU will always be awakened. But for MH-DBA, one

more sleep state implies more waiting packets and

extended sleep time which could greatly improve the

energy efficiency. As we can see from Fig. 5, the average

energy saving ratio is 0.85 of MH-DBA and 0.28 of 3M-

DBA when network load under 50 Mbit’s. Thus, MH-

DBA could reduce the energy consumed by about 32.9%

compare with 3M-DBA under low network load. ONU’s

1032



energy-savings of both two schemes encounter significant

decline at light traffic loads. Since higher delays implies

more waiting packets and longer sleep time. The

extended sleep time also helps improve the energy

efficiency of the ONU. At medium and high traffic loads,

MH-DBA is able to offer a lower delay, resulting in a

higher energy consumption of the ONU transmitter.

Hence, the lower cycle time leads to a lower energy

saving rate of ONUs. When network load exceed 300

Mbit’s, it starts to level off. It also because MH-DBA

considering one more energy saving mode than 3M-DBA,

and the ONU sleep time can be adjusted as much as

possible to maximize energy efficiency.

Fig. 5. Average energy saving ratio

Fig. 6. Average latency

US average latency of two schemes is demonstrated in

Fig. 6. In most cases, latency of 3M-DBA is lower than

MH-DBA scheme because of the smaller cache in

channel. However, for increasing traffic loads, dynamic

sleep polling cycle plays a main role. This advantage

helps the transmission system adapt to real access

environments, where traffic is mostly large or bursty.

What’s more, the efficient utilization of bandwidth also

improves the maximum available channel capacity. MH-

DBA shows its superiority by providing similar delay

under mid- and heavy-load conditions compare with 3M-

DBA. Since an adjusted shorter DBA cycle can reduces

the occupation of cache. The channel capacity is not

occupied fully and all the traffics can be serviced without

more delay.

Fig. 7 shows that the packet loss ratio is increasing as

the growth of traffic. For the reason that higher delays for

increasing traffic loads lead to a large number of data

packets waiting in the buffer. In most cases, packet loss

ratio of 3M-DBA is lower than MH-DBA scheme

because of the smaller cache in channel. Due to the

longer sleep time during deep sleep state, there may be

large data cache store in the queue, and the total cache is

easy to exceed Thrcache, which may lead to the packet loss.

In this situation, controller will adjust Thrtime shorter

automatically, so as to guarantee the QoS of service.

Fig. 7. Packet loss ratio

Secondly, in order to reflect the superiority of

improved energy-saving mechanism than the traditional

mechanism more directly, we make the simulation

without any packet loss for all algorithms by increasing

the queue buffer and to further evaluate the performance

in terms of average energy saving ratio and average

latency.

Fig. 8. Average energy saving ratio

Fig. 9. Average latency

1033



When the throughput tends to be stable, we conclude

the results of energy efficiency that can be seen from Fig.

8. With the increase of traffic load, energy saving ratio of

the two mechanisms is reduced, for the reason that the

probability of the system to trigger a sleep mechanism is

reduced as the load increases. But the presented results

still clearly show that MH-DBA has a superior

performance in terms of energy saving of ONUs in all

scenarios. In Fig. 9, MH-DBA shows its superiority by

providing similar lower delay under heavy-load

conditions compare with 3M-DBA. Since an adjusted

shorter DBA cycle can also reduce the packet waiting

time.

V. CONCLUSIONS

This paper proposes a SD-DON architecture that aims

to incorporate ONU’s four dormancy modes into DBA

algorithms to reduce ONU energy consumption.

Meanwhile, though MH-DBA, ONU energy saving time

are maximized as differentiable four modes for any given

network condition to maximize energy efficiency. Results

show that MH-DBA achieves a good energy saving

efficiency, lower delay and can significantly ensure the

quality of the distribution optical networks.

ACKNOWLEDGEMENT

This work has been supported by China Postdoctoral

Science Foundation (No.2016M602557), NSFC project

(61501049), Fund of State Key Laboratory of

Information Photonics and Optical Communications

(BUPT), P. R. China (IPOC2015ZT01), and funded by

State Key Laboratory of Advanced Optical

Communication Systems Networks, China.

REFERENCES

[1] F. Jiang, C. M. Tu, Z. W. Liu, Z. Wei, and T. Q. Dong,

“Differentiated energy-saving planning method of

distribution networks and its applications,” Power System

Technology, vol. 39, no. 10, pp. 2712-2718, 2015.

[2] W. Xia, Y. Wen, C. H. Foh, D. Niyato, and H. Xie, “A

survey on software-defined networking,” IEEE Commun.

Surveys Tut., vol. 17, no. 1, pp. 27–51, Jan.–Mar. 2015.

[3] D. Kreutz, F. M. V. Ramos, P. Verissimo, C. E.

Rothenberg, S. Azodolmolky, and S. Uhlig, “Software-

defined networking: A comprehensive survey,” Proc. IEEE,

vol. 103, no. 1, pp. 14–76, Jan. 2015.

[4] N. Bitar, “SDN and potential applicability to access

networks” in Proc. OFC/NFOEC 2014, San Francisco,

USA, March 2014, pp. 1-3.

[5] T. S. R. Shen, S. Yin, A. R. Dhaini, and L. G. Kazovky,

“Reconfigurable long-reach ultra flow access network: A

flexible, cost-effective, and energy-efficient solution,”

Journal of Lightwave Technology, vol. 32, no. 13, pp.

2353-2363, July 2014.

[6] P. Parol and M. Pawlowski, “Towards networks of the

future: SDN paradigm introduction to PON networking for

business applications,” in Proc. Federated Conf. Comput.

Sci. Inf. Syst., 2013, pp. 829–836.

[7] H. Yang, J. Zhang, Y. Zhao, J. Han, Y. Lin, and Y. Lee,

“SUDOI: Software defined networking for ubiquitous data

center optical interconnection,” IEEE Communications

Magazine, vol. 54, no. 2, pp. 86-95, 2016.

[8] H. Yang, J. Zhang, Y. Zhao, J. Wu, Y. Ji, Y. Lin, J. Han,

and Y. Lee, “Experimental demonstration of remote

unified control for OpenFlow-based software-defined

optical access networks,” Photonic Network

Communications, vol. 31, no. 3, pp. 568-577, 2015.

[9] S. S. W. Lee, K. Y. Li, and M. S. Wu, “Design and

implementation of a GPON-based virtual OpenFlow-

enabled SDN switch,” Journal of Lightwave Technology,

vol. 34, pp. 2522-2561, May 2016.

[10] H. Woesner and D. Fritzsche, “SDN and openflow for

converged access/aggregation networks,” in Proc. Optical

Fiber Communication Conf./Nat. Fiber Optic Engineers

Conf., Anaheim, CA, USA, 2013.

[11] D. A. Khotimsky, D. Z. Zhang, L. Q. Yuan, R. O. C.

Hirafuji, and D. R. Campelo, “Unifying sleep and doze

modes for energy-efficient PON systems,” IEEE

Communications Letters, vol. 18, no. 4, pp. 688-691, 2014.

[12] L. Zhang, Y. Liu, L. Guo, and X. Gong, “Energy-Saving

scheme based on downstream packet scheduling in

Ethernet passive optical networks,” Optical Fiber

Technology, vol. 19, no. 2, pp. 169–178, 2013.

[13] H. Yang, J. Zhang, Y. Ji, Y. He, and Y. Lee,

“Experimental demonstration of multi-dimensional

resources integration for service provisioning in cloud

radio over fiber network,” Scientific Reports, vol. 6, 30678,

2016.

[14] Y. Yan, S. W. Wong, L. Valcarenghi, S. H. Yen, D. R.

Campelo, S. Yamashita, L. Kazovsky, and L. Dittmann,

“Energy management mechanism for Ethernet Passive

Optical Networks (EPONs),” in Proc. ICC 2010, Cape

Town, South Africa, May 2010.

[15] N. Imanaka, Y. Nakahira, and M. Kashima, “Effects of

OLT activation control on quality of communication

services in Virtualized PON,” in Proc. 18th

OptoElectronics and Communications Conference Held

Jointly with 2013 International Conference on Photonics

in Switching, OECC/PS, 2013.

[16] S. W. Wong, L. Valcarenghi, S. H. Yen, D. R. Campel, S.

Yamashita, and L. Kazovsky. “Sleep mode for energy

saving PONs: Advantages and drawbacks,” in Proc. IEEE

Globecom Workshops, November - December, 2009.

[17] S. Panagiotis, G. Athanasios, K. Vasiliki, L. Malamati, P.

Georgios, N. Petro, and O. Mohammad, “On forecasting

the ONU sleep period in XG-PON systems using

exponential smoothing techniques,” in Proc. GLOBECOM,

Austin, TX USA, December 2014, pp. 2580-2585.

[18] H. Yang, J. Zhang, Y. Zhao, Y. Ji, J. Han, Y. Lin, and Y.

Lee, “CSO: Cross stratum optimization for optical as a

service,” IEEE Communications Magazine, vol. 53, no. 8,

pp. 130-139, 2015.

[19] A. F. Y. Mohammed, S. H. S. Newaz, J. K. Choi.

“Modeling and simulation of EPON with sleep mode

enabled using OPNET,” in Proc. International Conference

on ICT Convergence, 2014, pp. 16-21.

[20] L. C. Zhang, Y. J. Liu, L .Guo, and X. X. Gong, “Energy-

Saving scheme based on downstream packet scheduling in

1034



ethernet passive optical networks,” in Proc. IEEE Optical

Fiber Technology, March 2013, pp. 169-178.

[21] H. H. Lee, K. Kim, J. H. Lee, and S. Lee, “Power saved

OLT and ONU with cyclic sleep mode operating in WDM-

PON,” in Proc. International Conference on ICT

Convergence, October 2012, pp. 217-218

[22] H. H. Lee, K. Kim, J. H. Lee, and S. Lee, “Efficient power-

saving 10-Gb/s ONU using uplink usage-dependent sleep

mode control algorithm in WDM-PON,” ETRI Journal, vol.

35, no. 2, pp. 253-258, April 2013.

[23] Y. Maneyama, and R. Kubo, “QoS-aware cyclic sleep

control with proportional-derivative controllers for energy-

efficient PON systems,” Journal of Optical

Communications and Networking, vol. 6, nol. 11, pp.

1048-1058, November 2014.

[24] A. Nikoukar, I. S. Hwang, A. T. Liem, and C. J. Wang,

“QoS-aware energy-efficient mechanism for sleeping

mode ONUs in enhanced EPON,” Photonic Network

Communications, vol. 30, no. 1, pp. 59-70, 2015.

[25] V. P. Dung, L. Valcarenghi, M. P. I. Dias, K. Kondepu, P.

Castoldi, and E. Wong, “Energy-saving framework for

passive optical networks with ONU sleep/doze mode,”

Optics Express, vol. 23, no. 3, pp. A1-A14, 2015.

Yang Cao received the Ph.D. degree with the Electronic

Information Science and Technology in Huazhong University of

Science and Technology. His main research interests include the

network architecture, access network, power telecommunication

network, and energy efficiency of access network.

Hui Yang was born in Tianjin Province, China, in 1987. He is

currently a lecturer of the Institute of Information Photonics and

Optical Communications at BUPT. He received Ph.D. degree

from Beijing University of Posts and Telecommunications

(BUPT), Beijing, China, in 2014. He is currently an assistant

professor of institute of information photonics and optical

communications at BUPT. His main research interests include

network architecture, flexi-grid optical network, software

defined networks, and cross-stratum optimization. He has

authored or coauthored more than 60 peer-reviewed papers

related to the previously mentioned research topics in

prestigious international journals and conferences, and he is the

first author of more than 30 of them (e.g., IEEE Commu. Mag.,

Opt. Express).

Xiao-Xu Zhu was born in Tianjin Province, China, in 1991.

She is currently a postgraduate of the Institute of Information

Photonics and Optical Communications at BUPT. She received

her bachelor degree in Electronic Information Science and

Technology from Beijing Forestry University, Beijing, China in

2014. She is currently pursuing her M.S. degree in Beijing

University of Posts and Telecommunications (BUPT), Beijing,

China. Her main research interests include the network

architecture, elastic optical network, software defined networks,

and energy efficiency of access network.

1035



energy saving scheme based on multi-modes hybrid dynamic ...dynamic bandwidth optimization for...

Documents