Sie sind auf Seite 1von 5

1569100737

Integrated Bandwidth Allocation and Wavelength


Assignment in WDM-PON Networks
Xin Ye, Chadi M. Assi, and Mohamed A. Ali.

AbstractWDM-based passive optical network upgrades


from traditional single wavelength TDM-PON to satisfy the
growing traffic demands in the access network. Various
Dynamic Wavelength and Bandwidth Allocation algorithms
(DWBA) for WDM-PON have been studied. These DWBA
algorithms always separate the bandwidth allocation from
wavelength assignment. In our previous work , we introduced
a SLA (service level agreement) based dynamic resource
allocation scheme where ONUs are formed into different
groups, and the limited inter-wavelength multiplexing is used.
In this article, we present a new algorithm, which relaxes the
restriction to further increase the statistical multiplexing
opportunities among different ONUs. We also improve early
allocation scheme to support the network under heavy load,
further improving network utilization. In addition, we enhance
excess bandwidth allocation scheme to further improve
network performance. We use extensive simulations to
compare the proposed algorithm with the algorithms from
previous works. The results have shown that the
newly-proposed algorithm outperforms the existing
algorithms.
Index TermsDynamic Wavelength and Bandwidth
Allocation (DWBA), WDM-PON. Quality of Services.

I. INTRODUCTION
Wavelength division multiplexing (WDM) based Ethernet
passive optical networks (EPONs) have been considered as
one of promising technologies for next generation access
network to provide sufficient bandwidth to the increasing
end-users application such as IP telephony (VoIP),
Video-on-demand, High Definition TV (HDTV) and
high-quality audio transmission due to their low operational
costs and huge bandwidth [1]. How to provide effective and
fair resource allocation to Optical Network Units (ONUs)
has become one of key factors to ensure the success of
WDM-PON deployment. The most challenging task for
resource allocation comes from upstream transmission,
where ONUs share multiple wavelengths to transmit packets
to Optical Line Terminal (OLT). Similar to its predecessors,
for upstream transmission allocation, WDM-PON uses the
basic MPCP protocol to collect resource requests from
ONUs via REPORT messages, and disseminate the resource
Xin Ye and M.A.Ali are with the Electrical Engineering Department,
Graduate School of The City University of New York, New York, NY
10016 USA (e-mail: xinye@ ee.ccny.cuny.edu; ali@ccny.cuny.edu).
Chadi M Assi is with Concordia Institute, Information Systems
Engineering Department, Concordia University, Montreal, QC H3G 1M8,
Canada (e-mail: assi@ciise.concordia.ca).

assignment results to ONUs through GATE message .


How to support effective resource allocation in
WDM-PON has been extensively studied lately[1-5]. In [1],
we proposed Enhanced Dynamic Bandwidth Allocation
(EDBA) scheme, where, the ONUs are divided into K
groups based on their service level agreement (SLA), and K
is the number of wavelengths in PON, further one
wavelength is selected as the default wavelength for each
group, and the lightly loaded ONUs always use the default
wavelength for upstream transmission to avoid frequent
wavelength switching and simplify management. However,
the potential drawback of EDBA is that it doesnt handle the
bursty traffic very well. The authors of [2] proposed a WDM
extension to IPACT [3] to use strict priority based
scheduling for bandwidth allocation and first fit based
wavelength selection. In [5], the authors extended EPON
based DBA algorithms in [4] and came up with three variants
of the WDM DBA schemes; among three variants, two phase
based DWBA has the best performance. This scheme offers
heavily loaded ONUs two sub-timeslots within the one
transmission cycle, hence adding extra complexity into
network resource allocation process, furthermore the more
guard time is introduced for each heavy loaded ONUs. To
address this problem, [6] proposed so called Just-In-Time
(JIT) online scheduling framework which schedules ONUs
as soon as their REPORTs are received at OLT. However,
due to the fact that JIT excludes the cycle based DBA
concept, lightly loaded ONUs get more chances to transmit
than heavy loaded ONUs, leading to unfair bandwidth
allocation, in addition, imminent bandwidth allocation
prevents the heavily loaded ONUs from using excess
bandwidth saving from lightly loaded ONUs, the
performance for heavily loaded ONUs is expected to be
degraded.
In this paper, we will introduce a new Integrate Dynamic
Wavelength and Bandwidth (IDWBA) to further increase the
statistical multiplexing opportunities among different ONUs
and our contributions come from the following aspects: 1)
extending early bandwidth allocation to make sure no
bandwidth waste caused by Round Trip Time (RTT) from
GATE message within any cycle, and each ONU only gets
one timeslot per cycle, thus the overhead introduced by [5]
can be drastically reduced. Meanwhile, keeping every ONU
get one timeslot per cycle avoids unfair resource access in
[6]; 2) enhancing excess bandwidth allocation in [5] to
improve excess bandwidth utilization.
This paper is organized as follows: in section II, describes
1

1569100737
WDM-PON architecture and its corresponding media access
control protocol. In section III, the new IDWBA algorithm
will be presented. Section IV presents our simulation study;
finally section V concludes this paper.
II WDM PON ARCHITECTURE & MAC PROTOCOL
The development of protocols and algorithms for
WDM-PON is currently at their initial stage due to the fact
that there is no suitable standard network protocol [9-11]. A
comprehensive review of WDM-PONs has been presented in
[11].
1. WDM-PON Architecture
We assume WDM-PON variation I in [11] as our studied
architecture. In this architecture, each ONU has either
tunable transceiver or an array of multiple fixed transceivers
(each has its own wavelength) to transmit traffic to OLT in
the upstream direction. In the case of the fixed transceivers,
at any time, only one wavelength is used, the automatic
switch circuit is used to control the on/off status of
transceivers. In the OLT side, an array of fixed transceivers
are deployed, here the number of transceivers is equal to the
number of wavelengths supported in network. The
bandwidth allocation and wavelength assignment algorithm
is implemented at OLT, and the OLT can simultaneously
receive data from the various ONUs on different
wavelengths and transmit data and control messages to the
ONUs. An example of WDM-PON architecture is shown in
Figure 1.
Rx
ONU

MAN/WAN

OLT

Tx
Rx

III INTEGRATED DYNAMIC WAVELENGTH AND BANDWIDTH


ALLOCATION (IDWBA)
Unlike EDBA proposed in [1], where the early allocated
ONUs have to use the default wavelength, our new algorithm
relaxes this restriction and makes fully dynamic wavelength
and bandwidth integrated. In this way, we expect that
IDWBA further increases statistical multiplexing
opportunities. The IDWBA consists of two steps: 1)
Initialization: based on SLA and transmission overhead,
IDWBA calculates the minimal guaranteed bandwidth that
OLT needs to provide for each ONU; 2) Real time
wavelength selection and bandwidth allocation: IDWBA
computes bandwidth assignment and wavelength selection
for each ONU based on the bandwidth requests carried by
REPORT message(s) and the minimal guaranteed
bandwidth.
A. Minimal guaranteed bandwidth calculation
i

The minimum guaranteed bandwidth B Min for ONUi


defined in [4], is dependant on the weight assigned to each
ONU based on the SLA between the service provider (SP)
and users. We consider a PON with N ONUs, and the OLT
supports K wavelengths, here, we assume K<N, and the
transmission speed per wavelength is R (Mbps). The
tunable transmitter at ONUs can be tuned into any
wavelength supported by OLT. The tuning time of the
tunable laser is in the order of 1us . We denote TCycle as the
grant cycle. Within a grant cycle, every ONU is granted a
timeslot to transmit packets to OLT. Further, we also denote
Tg as the guard time that separates the transmission window
between ONUi and ONUi+1, Tu as the tuning time for a

Tx
Rx

specific ONU is determined, the OLT sends GATE message


to this ONU with the transmission schedule of the ONU (i.e.,
transmission start time, transmission length, and
corresponding wavelength channel identifier).

1: N splitter
Rx
ONU

transmitter tuning into different wavelength, and wi as the


weight assigned to each ONU based on its SLA such that

Figure 1 WDM-PON architecture

= 1. Therefore, the minimum guaranteed bandwidth

i =1

2. Multipoint Control Protocol (MPCP) extension


MPCP developed by the IEEE802.3ah Task Force [13]
has been used to arbitrate time-slots among OLT and ONUs
in EPON. To support the upgrade from EPON to
WDM-PON, the MPCP has to be extended to enable
multiple wavelength support. In WDM extension to MPCP,
the ONUs send REPORT messages to OLT to request the
bandwidth and the supported wavelength information, OLT
then executes the dynamic bandwidth allocation and
wavelength assignment algorithm to calculate the upstream
transmission time slot and corresponding wavelength for all
ONUs. Once the transmission slot and wavelength for a
B. Real time wavelength & bandwidth allocation
The IDWBA gives transmission opportunity for each
2

per cycle the OLT can allocate for an ONUi is computed as


follows:
i
BMin
=

In

(Tcycle N Tg K Tu ) K R wi
8

case

of

no

SLA

classification

per

(1)

ONU,

wi = w =
i
=
BMin

1
, i, and wi = 1 ,
N
i =1
(Tcycle N Tg K Tu ) K R

8N

(2)

ONU within one cycle, generally the maximal length of a


cycle is 2ms according to [5]. Within a cycle, the ONUs
sends REPORT messages to request transmission window

1569100737
for next cycle. Upon receiving REPORT(s), the OLT runs
real time wavelength and bandwidth allocation to determine
the time slot and the selected wavelength for ONU(s). We
i

define Bassign

i
Re q

as the assigned bandwidth to ONUi, and

as the requested bandwidth by ONUi.

The real time wavelength selection and bandwidth


allocation has two components: early allocation and regular
WDM based DBA. Different from [5], there is always early
allocated ONUs within a cycle. There are two early
allocation scenarios: 1) there are lightly loaded ONUs within
a grant cycle, for lightly loaded ONUs, i.e., B

i
Re q

i
Min

the OLT conducts allocation for them right after receiving


i

their REPORTs, i.e, Bassign = B Re q and the wavelength


with the least load will be selected. 2) there is no lightly
loaded ONU in a cycle, then the schemes in [5] assign
resources for all ONUs only after receiving all REPORTs,
the resources between the time that OLT sends first GATE
message and the time the first ONU sends out data get
wasted. This wasted time is equivalent to single trip delay
(STD) from OLT to ONU and processing delay at ONU. In
order to reduce this kind of bandwidth waste without missing
the opportunity of accessing excess bandwidth, we propose
Just-Enough-Time (JET) allocation scheme to schedule the
least loaded ONUs at time:
t start = t resource _ avail t STD t proces sin g ,
Here, t resource _ avail

is time when the wavelength is

available for next cycle transmission, and we allocate


i
i
Bassign
= BMin
for the selected ONUi. Since the least

loaded ONUs are used for JET, this makes sure the
unfairness is very limited.
The regular WDM based DBA component deals with
those
unallocated
ONUs
having
heavy
load
i

(i.e., BRe q > B Min ) after all REPORTs are received.


Resource allocation for heavy load ONUs will leverage the
extra saving bandwidth from lightly load ONUs. Assume
there are M lightly loaded ONUs, the extra saving bandwidth
from
these
lightly
loaded
ONUs
Total

is Bextra =

M 1

(B

i
Min

i
i
i
BRe
q ) , where B Min > B Re q . The

i =0

Total

heavily loaded ONUs will get extra bandwidth from Bextra


in

addition
i
assign

=B

i
extra

to

+B

its
i
min

guaranteed

bandwidth,

. To compute B

i
extra

i.e.,

, we assume

Total

always scheduled first, and Bextra is always updated after


one heavily loaded ONU gets assigned.
Then for a given heavily loaded ONUi
i
i
i
i
BRe
if BRe
q B Min
q < B Min + Bextra
(3)
i
i
if BRe
Bextra
q > B Min + Bextra

i
: Bextra
=
And

Total
Total
i
= Bextra
Bextra
,
Bextra

Total

Bextra = Bextra

(4)

Here L is the number of remaining unallocated ONUs.


The pseudo code of the IDWBA algorithm is shown in
Figure 2. It consists of three functions: the main IDWBA
function, Grant_on_fly function for early allocation, and
Run_wdm_dba function for the resource allocation for
heavily loaded ONUs.
IV PERFORMANCE EVALUATION
To validate our algorithm, we develop a WDM-PON
event driven simulator in C++. This simulator consists of
two modules: a module simulating ONUs and a module
simulating OLT. To simplify the simulator, we will ignore
the downstream traffic transmission.
The following are some parameters used in our simulation:
The number of ONUs N=32; the number of Wavelengths
channels K=2; maximum cycle time = 2 ms; PON Channel
speed R = 1 Gbps; distance between OLT and ONU is 20Km;
the guard time=1 s ; the tuning time =1 s ;
End-users/ONU link speed is 100Mbps; the distance
between ONU and the passive splitter is 5 Km; buffering
queue size is 1Mbytes. An extensive study shows that most
network traffic (i.e., http, ftp, variable bit rate (VBR) video
applications, etc.) can be characterized by self-similarity and
long-range dependence (LRD) [7]. To model the bursty
nature of Internet traffic, we generated self-similar traffic
based on Pareto distribution with a Hurst H=0.8; the source
code was provided by [5], where packet sizes are uniformly
distributed by 64 and 1518 bytes.
In the simulation, we conduct performance analysis of our
algorithm against previous EDBA in [1] and DWBA3
control excess algorithm in [5]. The performance parameters
we compare are: average packet delay, maximum packet
delay, and network throughput.
Figure 3 presents the network average delay for IDWBA,
EDBA and DWBA respectively. Here, the traffic load of

each heavily loaded ONU gets the same share of excess


bandwidth ( Bextra =

Total
Bextra
N M

), and we sort the

unallocated ONUs according to their requested bandwidth,


and the ONUs having the lower requested bandwidth are
3

1569100737

IDWBA()
{
Early__allocation_flag=0;
Num_HeavyLoad=0;
While (receive_Report( ONUi ))
{
if (

i
i
BRe
q <= B Min

Total
extra +=

BMin - B

Grant_on_fly(

i
Re q

i
BRe
q , i,);

Early_allocation_flag=1;
}
Else
{
if ( t current

t start )

Grant_on_fly(

BRe q , i)

{
i
assign =

i
Re q ;

Sort(

Max_Capacity=0;
for (j=0; j<K; j++)
if (max_cap<channel_capacity[j])
{
Max_Capacity = channel_capacity[j];
lambda=j;
}
}
Update (channel_capacity);
Send the GATE(

i
Bassign

i
BMin

Total
Bextra = Bextra

/(Num_HeavyLoad j);

Find the corresponding ONU i;


if (

i
i
BRe
q < B Min
i

; Bexcess

Bextra

= BRe q - B Min ;

else
i
Bexcess
= Bextra ;

, lambda)

i
Bassign

to ONUi;
}

= B Min +

i
Bexcess
;

max_cap=0, lambda=0;
for (k=0; k<=K; k++)
if (max_cap<channel_capacity[k])
{
max_cap=channel_capacity[k];
lambda=k;
}

, i)

}
Else
Num_HeavyLoad++;
}

Update(channel_capacity);

if (i==N)
{
Run_wdm_dba(

BRe q ); // sort the remaining unassigned bandwidth)

for (j=0; j<Num)HeavyLoad; j++)


{

{ if (Early_allocation_flag==0)
Grand_on_fly(

Total
Bextra
, Num_HeavyLoad)

Run_wdm_dba(

Total
extra ,Num_HeavyLoad);

Send GATE (

i=0;
Break;
}
i++

i
Bassign

, lambda) to ONUi;

i
Total
Total
Bextra
= Bextra - Bexcess ;

}
}

}
}

Figure 2: Psuedo code for the proposed IDWBA.


ONUs varies from 0.1 to 0.9. Under lightly load (load<=0.3),
these three algorithms archive similar performance. Under
medium load (where 0.3<load <0.6), IDWBA and EDBA
outperform DWBA. Since in EDBA and IDWBA, the heavy
loaded ONUs have opportunities to exploit the resources from
all wavelengths. When network load becomes heavy, EDBAs
performance will be similar to the performance of DWBA
because the network is overloaded, and there is no ONU that is
qualified for early allocation. Due to the fact that IDWBA
employs the enhanced early allocation scheme to allow early
resource allocation for even heavy loaded ONUs, hence, the
performance of IDWBA is better than the previous two
algorithms.
0.1 6

Average Delay(Second)

0.1 4
0.1 2
0 .1

EDBA
D WBA

0.0 8

ID WBA

0.0 6
0.0 4
0.0 2
0

0.2

0 .4

0.6

0 .8

Link Load

Figure 3. Average Packet Delay V.S. Link Load

In Figure 4, we compare the maximum delay under IDWBA,


EDBA, and DWBA. When the link load is very light (load
<=0.2), there is a small amount of traffic in the network, these
three algorithms have the similar results in maximum delay.
Since IDWBA always tunes to the maximum capacity
wavelength, wavelength-tuning time adds additional delay to
the maximum delay of IDWBA. When the link load is medium
(0.3< =load < 0.6), DWBA has the worst performance since it
always uses the default wavelength while EDBA has the best
performance since it always tries to use its default
4

wavelength first, if the default wavelengths capacity is not


enough, then it will turn to other wavelength to transmit.
IDWBAs performance is not as good as EDBA because of the
delay introduced by wavelength switching. When the link load
becomes heavy, IDWBA achieves the best results while
DWBA and EDBA have the similar results. Since IDWBA
fully integrate the dynamic bandwidth allocation and
wavelength assignment together, and enhanced early allocation
scheme can improve network utilization, hence, even though
the network is overloaded, IDWBA can still achieve relatively

1569100737
ideal results.
0.1 6

Maximal Delay ( Second )

0.1 4
0.1 2

EDBA

0 .1
0.0 8

DWBA

0.0 6

IDWBA

efficiency and improve the network performance which


comparing with EDBA in [1] and DWBA in [6]. Since the
tunable transceiver is expensive, in the future study, we will
investigate a hybrid WDM-PON architecture where only a
limited number of ONUs employ tunable lasers. We will look
for the tradeoff between cost and network performance.
REFERENCES
[1]

0.0 4
0.0 2

[2]

0
0

0.2

0 .4

0 .6

0 .8

Link Load

[3]
Figure 4. Maximum Packet Delay V.S. Link Load

[4]

Figure 5 shows the throughput of these three algorithms.


As we expected, IDWBA always achieves better performance
than EDBA and DWBA. With the increasing network link
load, the throughput also increases. IDWBA picks up to around
92% at a network link load of 0.5 and keeps this throughput
while the network link load increases to 0.9. However, the
throughput improvement of IDWBA over EDBA is not
significant, since EDBA utilizes traffic grooming to evenly
distribute traffic into different wavelengths and it also allows
limited wavelength.
1.2

[5]
[6]
[7]
[8]
[9]

[10]

Throughput

0.8

EDBA

[11]

DWBA

0.6

IDWBA
0.4

[12]

0.2

[13]
0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Xin Ye, A. Sana, C.Assi, M.Ali, Enhanced Dynamic Bandwidth


Allocation in WDM-PON, SPIE Optics East 2007..
K.H.Kwong, D. Haarle, and I. Andonovic, Dynamic Bandwidth
Allocation Algorithm for Differentiated Services over WDM_PONs
IEEE International Conference on Communications Systems (ICCS)
pp116-120,September 2004, Singapore.
G. Kramer, G. Pesavento Ethernet Passive Optical Network (EPON):
Building a Next-Generation Optical Access Network, IEEE
Communication Magazine. Vol 40, pp.66-73, 2002.
Chadi M. Assi, Yinghua Ye, Sudhir Dixit, and Mohamed A. Ali,
Dynamic bandwidth allocation for Quality-of-Service over Ethernet
PONs, IEEE JSAC, Vol 21, pp. 1467-1477, 2003.
A. R. Dhaini, C. M. Assi, M. Maier, and A. Shami, Dynamic
Wavelength and Bandwidth Allocation in Hybrid TDM/WDM EPON
Networks, Journal of Lightwave Technology.
Michael P. McGarry etc., Just-in-Time Online Scheduling for WDM
EPONs, ICC2007, June 2007.
Glen Kramer, Synthetic traffic generation, C++ source code version,
http://wwwcsif.cs.ucdavis.edu/ kramer/research.html.
W. Leland, M. Taqqu, W.Willingler, and D.Wilson. On the Self Similar
Nature of Ethernet Traffic (Extended Version), IEEE/ACM
Transactions on Networking, pp.1-15 Feb 1994.
F.An, K.S.Kim, D.Gutierrez, S. Yam, E. Hu, K.Shrikhandle and
L.G.Kazovsky, SUCCESS: A Next-Generation Hybrid WDM/TDM
Optical Access Network Architecture. Journal of Lightwave
Technology, Vol 22, pp. 2557-2569, 2004.
K.S.Kim, D.Gutierrez, F.An, L.G.Kazavsky, Design and performance
Analysis of Scheduling Algorithms for WDM-PON under
SUCCESS_HPON Architecture, Journal of Lightwave Technology, Vol
23, pp.3716-3731, 2005.
Amitabha Banerjee,,etc., Wavelength-division-multiplexed passive
optical network (WDM-PON) technologies for broadband access: a
review [Invited], J. of Optical Networking, Vol 4, pp.737-758, 2005.
M .McGarry, M. Maier, and M. Reisslein, Ethernet PONs: A Survey of
Dynamic Bandwidth Allocation (DBA) algorithms, IEEE Optical
Communications, August 2004.
IEEE 802.3ah, Ethernet in the First Mile Task Force,
http://www.ieee802.org/3/efm/index.html

Link Load

Figure 5. Throughput V.S. Link Load

As we can see from Figure 3 to Figure 5, when the traffic


load is light, there is no big difference among these three
algorithms, since the OLT has enough resources to
accommodate the incoming packets. When the network traffic
becomes heavy, IDWBA always outperform EDBA and
DWBA in average delay, maximum delay, network throughput
and packet loss due to fully flexible integrated dynamic
wavelength assignment and bandwidth allocation scheme.
V CONCLUSION
In this paper, we present an improved integrated dynamic
wavelength assignment and bandwidth allocation (IDWBA)
algorithm for WDM-PON. IDWBA fully exploits the statistical
multiplexing opportunities among different ONUs. Simulation
results have showed that IDWBA increases the network
5

Das könnte Ihnen auch gefallen