Delay-Aware Dynamic Wavelength Bandwidth Allocation in
Time- and Wavelength-Division-Multiplexed Passive Optical Network
Junsu Kim
1
, Yong-Kyu Choi
2
, Han Hyub Lee
3
and Chang-Soo Park
1
1
School of Information and Communications, Gwangju Institue of Science and Technology (GIST),
123 Cheomdangwagi-ro, Buk-gu, Gwangju 500-712, Republic of Korea
2
Photonic Research Facility Center, Gwangju Institue of Science and Technology (GIST),
123 Cheomdangwagi-ro, Buk-gu, Gwangju 500-712, Republic of Korea
3
Optica Internet Division, Electronics and Telecommunications Research Institute (ETRI),
218 Gajeong-ro, Yuseong-gu, Daejeon 305-700, Republic of Korea
Keywords: Dynamic Wavelength Bandwidth Allocation, Tuning Time, TWDM-PON.
Abstract: Most research works related to dynamic wavelength bandwidth allocation in a time- and wavelength-
division–multiplexed passive optical network are based on bin packing algorithm and present many
advantages in terms of bandwidth utilization and energy efficiency. However, delay performance can be
significantly degraded by frequent wavelength changes. In this paper, we propose a delay-aware dynamic
wavelength bandwidth allocation algorithm. First, we theoretically analyze queuing delay corresponding to
the tuning time, decision cycle, and the number of optical network units. And then, we show that the
queuing delay is decreased by setting the limit value of the number of optical network units in a wavelength
channel. In addition, we derive the threshold ratio of the tuning time to the decision cycle as 0.004.
1 INTRODUCTION
As user bandwidth requirements are escalating to
support multimedia streaming services such as high-
definition television (HD-TV), 3D TV, and mobile
services, the bandwidth capacity of optical access
networks needs to be expanded by upgrading legacy
PONs to the next-generation passive optical network
2 (NG-PON2). Among various types of system
architectures for NG-PON2, time- and wavelength-
division-multiplexed passive optical networks
(TWDM-PON) are known as the best technology in
terms of power budget, key component maturity, and
especially system coexistence with legacy PONs.
Full Service Access Network (FSAN) has studied
TWDM-PONs since 2010, and this technology has
been specified as an ITU-T G.989 recommendation
(Yuanqui and Frank, 2013).
1.1 TWDM-PON
TWDM-PONs increase upstream and downstream
bandwidth capacity by stacking four or eight
(optional) XG-PONs with wavelength division
multiplexing (WDM) technology. An optical line
terminal (OLT) consists of four or eight fixed optical
transceivers and can manage network resources
more flexibly than legacy PONs owing to the
tunability of the optical network unit (ONU), which
is composed of a tunable transmitter and receiver.
Figure 1 shows the block diagram of a general
TWDM-PON architecture (ITU-T Study Group 15,
2013, Yuanqiu, Xiaoping, Frank, Xuejin, Guikai,
Yinbo and Yiran, 2013).
Figure 1: The general architecture of a TWDM-PON (CO:
Central Office, OA: Optical amplification for long reach
or high-split-application system).
ONU1
ONU2
ONU3
ONUn
Splitter
OA
(optional)
.
.
.
MAC
PHY4
PHY3
PHY2
PHY1
WDM MUX/DEMUX
CO
DFB LDs Burst-mode receivers
Tunable LD Tunable receiver
64
Kim J., Choi Y., Lee H. and Park C..
Delay-Aware Dynamic Wavelength Bandwidth Allocation in Time- and Wavelength-Division-Multiplexed Passive Optical Network.
DOI: 10.5220/0005053600640068
In Proceedings of the 5th International Conference on Optical Communication Systems (OPTICS-2014), pages 64-68
ISBN: 978-989-758-044-4
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
1.2 Dynamic Wavelength Bandwidth
Allocation (DWBA)
The general requirements of TWDM-PONs are
specified in the ITU-T G.989.1 recommendation.
However, there have not been specifications related
to system management and bandwidth allocation
mechanisms in the transmission convergence (TC)
layer yet. Ideal methods of dynamic wavelength and
time slot allocation at the same time, well known as
DWBA, have been studied by many researchers. As
a result of extensive research, many algorithms have
been introduced to research societies related to
optical access networks. Most algorithms originate
from the conventional bin packing (BP) algorithm,
whose objective is to achieve an efficient resource
distribution with the minimum number of channels.
There are several types of algorithms, including
first fit (FF), first fit decreasing (FFD), next fit (NF),
and best fit (BF). Among these algorithms, FFD is
known as the most optimal algorithm in terms of the
minimum number of wavelength channels utilized.
According to the principle of FFD, OLT collects
reported information, including ONU bandwidth
requirements, and rearranges it in descending order.
Then, the OLT checks the channel capacity from the
first to last wavelength channels and finds an
available channel to support the ONU bandwidth
requirement. After finding a supportable wavelength
channel, the OLT makes the ONU change a
previously allocated wavelength to a new
wavelength and allocates as many time slots as the
ONU requires. Figure 2 shows the operating
principle of FFD with a flow chart.
Figure 2: Flow chart of FFD.
Although DWBA has many advantages,
including flexibility of bandwidth allocation, load
balancing, and the possibility of saving energy, there
are also limitations caused by a long wavelength
tuning time. Several milliseconds are required for
the ONU to change its wavelength (Jens and
Edmond, 2006). Because the wavelength tuning time
is long compared to the service interval (typically
125 µs in an XG-PON), at least approximately 80 or
more frames should be accumulated in the buffer.
Thus, the degradation of the delay performance can
be caused by a large number of accumulated frames.
Although the wavelength tuning time of an optical
tunable component cannot be negligible for this
reason, there have been few studies on DWBA with
consideration of this factor. For the realization of a
delay-aware DWBA algorithm, we should analyze
the relation between several parameters related to
wavelength tuning and queuing delay.
In this paper, we find important parameters that
affect delay performance degradation and reveal the
theoretical relation between the parameters and
delay. We also propose a DWBA algorithm to
reduce delay. Then, we evaluate the system
performance in terms of the queuing delay through
simulation and compare the results with
conventional DWBA algorithms.
2 APPROXIMATION ANALYSIS
Before analyzing the relation between wavelength
tuning and delay, we assume that OLT should check
ONU buffer occupancy and make a decision whether
ONU changes its wavelength or not every period T
(wavelength tuning decision cycle). All decision
points are synchronized simultaneously. ONU
changes a wavelength channel during tuning time τ
as soon as it receives a wavelength tuning message
from the OLT. After ONU wavelength tuning is
finished, ONU notifies OLT of its tuning completion
by sending wavelength tuning completion message.
And then, OLT allocates the time bandwidth to
ONU
We assume that ONU has a stochastic tendency p
of wavelength tuning. If there is no limitation of
wavelength change, ONU changes its wavelength
frequently. As a result, a stochastic tendency of
ONU is increased and approximately same as 1.
However, if OLT allocates wavelength channel with
some restrictions that prevent ONU from changing
wavelength channel time to time, p is decreased. In
the case that ONU’s wavelength channel is fixed
like WDM PON, p is equal to 0.
ONU bandwidth
calculation
Completion
Sorting the calculated
bandwidth in descending
order
Can an ONU be
supportable?
Is there any
remaining ONU
to be allocated?
Inspection of
i
th
wavelength channel
capacity
Set i=1
i++
Preparing the next
ONU wavelength
assignment
i
th
wavelength
assignment
Yes
No
Yes
No
Delay-AwareDynamicWavelengthBandwidthAllocationinTime-andWavelength-Division-MultiplexedPassiveOptical
Network
65
Figure 3: Utilization of time bandwidth during decision
cycle when ONU changes its wavelength.
Figure 4: Utilization of time when ONU maintains its
wavelength.
To approximately analyze the relation, we divide
the situation into two cases, as shown in the above
figures. The first is the case that the ONU changes
its wavelength. The second is the case is that the
ONU does not participate in wavelength tuning.
When the ONU participates in wavelength tuning, it
cannot transmit upstream signals during tuning time
τ. After completing wavelength tuning, adequate
time slots with the consideration of fairness among
N ONUs belonging to the same wavelength channel
group are assigned to the ONU. On the other hand,
ONUs that do not participate in wavelength tuning
can transmit an upstream signal during not only τ but
also the remaining time T - τ. A total of m nontuning
ONUs belonging to the same wavelength group
share the time bandwidth during τ, and N ONUs,
including tuning ONUs, share the bandwidth during
T - τ evenly.
Generally speaking, µ is occasionally changed by
the network situation in a real PON system.
However, we consider only the effective service rate
(µ) during T to simplify the calculation. The
effective service rate can be calculated by the
following equations:
(1)
_
(2)
µ
tuning
is the effective service rate of ONUs
needed to participate in wavelength tuning, and
µ
not_tuning
is the effective service rate of other ONUs.
µ
max
is the maximum service rate of a wavelength
channel, typically, 2.5 Gbps.
On the basis of an M/M/1 queuing model (Henry
and John, 1994.), the number of accumulated
packets in the ONU buffer is derived by the
following equation:
(3)
For our system, B should be modified to a new
equation that reflects the wavelength tuning
condition with stochastic tuning tendency p.
Therefore, the expected accumulated number of
packets β is derived as follows:
∗
∗
1
1
_
(4)
The queuing delay (D) is derived by Little’s
Theorem and converted into a simple form by
inserting equations (1) and (2) into equation (4).
1
(5)
1
,
1
(6)
/
(7)
Because the queuing delay is improved as δ and
N are decrease, we have to set the limit value of the
maximum number of ONUs in the same wavelength
group (N
th
) for efficient DWBA realization.
3 PROPOSED ALGORITHM
In the previous section, we explained why the limit
(N
th
) is required for wavelength assignment and
bandwidth allocation. We propose the modified FFD
algorithm reflecting N
th
under the following
assumptions: (1) all ONU bandwidth requirements
are bounded between R
min
and R
max
,
and (2) the sum
of every bandwidth requirement is lower than the
sum of the capacity of all wavelength channels.
Briefly speaking, the system can always support all
bandwidth requirements.
To satisfy the assumption, the value of R
max
is
⁄
(8)
An ONU that changes its wavelength channel by DWBA
only uses time bandwidth during this time.
service interval (125μs)
…
…
Others can use time bandwidth
during full time of decision cycle.
Time bandwidth is available
…
…
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
66
C is the channel capacity of a single wavelength
channel, I is the total number of ONUs, and J is the
number of wavelength channels in the system.
R
min
can be changed by a system operator
corresponding to the target service rate in the system
because it depends on the decision of the system
operator.
Figure 5: Flow chart of the proposed algorithm.
As shown in Figure 5, the basic principle of the
algorithm is similar to FFD except for checking the
wavelength availability condition. Although the
ONU bandwidth requirement is lower than the
capacity of a wavelength channel, the ONU cannot
be assigned to the wavelength channel if the number
of accommodated ONUs in the channel is equal to
N
th
because we should avoid a biased wavelength
assignment that causes severe degradation of the
delay performance.
4 SIMULATION RESULTS
We evaluate the performance of the proposed
algorithm by conducting simulations. We only
consider the case in which 32 ONUs exist in a
TWDM-PON system that utilizes eight wavelength
channels (2.5 Gbps for upstream line rate, 10 Gbps
for downstream line rate per single wavelength
channel). The wavelength tuning time is 10 ms, and
the decision cycle of wavelength assignment is fixed
to 1 s. The optical link distance between the OLT
and ONU is 20 km (The distance value is a general
requirement of TWDM-PON). The ONU generates
frames randomly, and the load is followed by a
Poisson distribution. We control the mean offered
load of the ONU from 100 Mbps to 600 Mbps when
we evaluate the queuing delay of the system.
Because the ONU queuing delay values differ from
each other as a result of the randomness of the
simulation, we acquire the mean value of the delay
to show a tendency and compare values that result
from FFD and the proposed algorithm.
Figure 6: Performance comparison of mean queuing delay
correspoding to offered loads.
Figure 7: Average queuing delay corresponding to T.
As shown in Figure 6, the delay performance
grows worse as N
th
increases. Because of the
randomness of traffic generation, many ONUs can
be assigned to the same wavelength channel if there
is no limitation on N
th
. Consequently, the OLT
cannot allocate time slots when many time slots are
simultaneously required. On the other hand, the
proposed algorithms exhibit significantly better
results than the conventional algorithm. Secondly,
we also evaluate the average queuing delay
ONU bandwidth
calculation
Completion
Sorting the calculated
bandwidth in descending
order
Can an ONU be
supportable?
Is there any
remaining ONU
to be allocated?
Inspection of
i
th
wavelength channel
capacity
Set i=1
i++
Preparing the next
ONU wavelength
assignment
i
th
wavelength
assignment
Yes
No
Yes
No
N in i
th
channel
< N
th
No
Yes
100 150 200 250 300 350 400 450 500 550 600
10
-4
10
-3
10
-2
10
-1
10
0
Load (Mbps)
Average delay (s)
Proposed algorithm Nth = 4
Proposed algorithm Nth = 6
Proposed algorithm Nth = 8
Conventional FFD
0.5 1 1.5 2 2.5 3
10
-3.9
10
-3.8
10
-3.7
10
-3.6
Decision cycle (s)
Average delay (s)
Proposed algorithm Nth = 4
Proposed algorithm Nth = 6
Proposed algorithm Nth = 8
Conventional FFD
Delay-AwareDynamicWavelengthBandwidthAllocationinTime-andWavelength-Division-MultiplexedPassiveOptical
Network
67
corresponding to the decision cycle (T) to study the
effect of T. The result is plotted in Figure 7 under
the assumption that the mean offered load of the
ONU is equal to 96 Mbps. As shown in Figure 7,
there exists trade-off relation between the decision
cycle time and delay. To improve delay performance,
long decision cycle time should be set.
When the offered load is 400 Mbps, there is a dip
in the graph of conventional FFD because more
wavelength channels are allocated compared to the
case of the 300-Mbps load. That is because the long
and fixed decision cycle in simulation framework
prevents OLT from reacting to network traffic
condition instantly. When traffic condition is
extremely changed compared to the beginning of
wavelength decision, channel under-utilization or
overflow problem can happen in the system which
uses conventional FFD. In this case, when the load is
300 Mbps, four wavelength channels are required.
However, six wavelength channels are required
when the load is 400 Mbps. This difference of the
number of wavelength channels during fixed cycle
time makes a peak at the point of 300-Mbps load.
As explained in section 2, δ decreases as the
decision cycle duration increases. As a result, the
queuing delay is inversely proportional to the
decision cycle, which means that many packets that
accumulate during the wavelength tuning time
cannot be completely transmitted within a shorter
decision cycle. Therefore, it is important to set a
proper decision cycle length to compensate for the
defect caused by wavelength tuning. In this
simulation with the assumption mentioned earlier,
the saturation value of T is 2.5 s (δ = 0.004).
5 CONCLUSIONS
We proposed a DWBA algorithm with a limitation
on the number of ONU assignments to prevent the
queuing delay from increasing. By theoretical
simulations, we reveal the relation between the
number of ONUs in a wavelength channel and the
queuing delay and ensure that N
th
is an important
parameter affecting the delay performance through
the simulation. In addition, we found that the
optimized value of δ should be lower than 0.004 for
efficient DWBA in TWDM-PON.
ACKNOWLEDGEMENTS
This work was partly supported by the
MSIP(Ministry of Science, ICT and Future
Planning), Korea, under the C-ITRC(Convergence
Information Technology Research Center) support
program (NIPA-2014-H0401-14-1009) supervised
by the NIPA (National IT Industry Promotion
Agency) and the ICT R&D Program 2014 (14-000-
05-002).
REFERENCES
Yuanqui, L., Frank, E., 2013. “TWDM-PON: The solution
choice for NG-PON2”, Huawei, no. 68.
ITU-T Study Group 15, 2013. “40-gigabit-capable passive
optical networks (NG-PON2): General Requirements”,
ITU-T G.989.1 Recommendation.
Yuanqiu, L., Xiaoping, Z., Frank, E., Xuejin, Y., Guikai,
P., Yinbo, Q., Yiran, M., 2013. “Time- and
Wavelength-Dvision Multiplexed Passive Optical
Network (TWDM-PON) for Next-Generation PON
Stage 2 (NG-PON2)”, Journal of Lightwave
Technology, vol. 31, no. 4, pp. 587-593.
Jens, B., Edmond, M., 2006. “Tunable Lasers in Optical
Networks”, Journal of Lightwave Technology, vol. 24,
no. 1, pp. 5-11.
Henry, S., John, W., 1994. “Probability, random processes
and estimation theory for engineers”, Prentice Hall, 4
th
edition.
OPTICS2014-InternationalConferenceonOpticalCommunicationSystems
68
