Content uploaded by yu nakayama
Author content
All content in this area was uploaded by yu nakayama on May 16, 2018
Content may be subject to copyright.
Tu3L.1.pdf OFC 2018 © OSA 2018
Remotely Controlled XG-PON DBA with Linear Prediction
for Flexible Access System Architecture
Naoki Hanaya1, Yu Nakayama2, Manabu Yoshino2, Ken-Ichi Suzuki2, Ryogo Kubo1
1Department of Electronics and Electrical Engineering, Keio University, 3-14-1 Hiyoshi, Kohoku-ku, Yokohama-shi, Kanagawa 223-8522, Japan
2NTT Access Network Service Systems Laboratories, NTT Corporation, 1-1 Hikarinooka, Yokosuka-shi, Kanagawa 239-0847, Japan
hanaya.naoki@kbl.elec.keio.ac.jp
Abstract: We propose a dynamic bandwidth allocation (DBA) algorithm with linear prediction in
a remotely controlled 10-Gigabit-capable passive optical network (XG-PON). Simulation results
show that the proposed DBA provides low-latency upstream communication compared to non-
predictive DBA.
OCIS codes: (060.4250) Networks; (060.4510) Optical communications
1. Introduction
Recently, along with the emergence of big data and cloud computing, the variety of communication demands has
rapidly expanded with the diversification of services for end users. To adapt to the expansion of communication
demands, a higher-speed passive optical network (PON) is studied as a fiber-to-the-home (FTTH) infrastructure that
connects an optical line terminal (OLT) to multiple optical network units (ONUs) [1]. The PON has the advantages
of cost-effective installation and operation [1]. The 10-Gigabit-capable PON (XG-PON) standardized for the next-
generation PON system in the ITU-T uses time-division multiple access (TDMA) technology for upstream
transmissions [2, 3]. The upstream bandwidth is assigned by a dynamic bandwidth allocation (DBA) function in an
OLT, as shown in Fig. 1(a).
Flexible access system architecture (FASA) has been proposed to meet the diversified requirements for next-
generation communication services [4]. In FASA-based PON systems, we can change and add network functions
quickly whenever new services emerge. FASA also enables the separately/remotely controlled DBA, shown in Figs.
1(b) and 1(c), to coordinate multiple PON services. However, the separately or remotely controlled DBA causes a
longer DBA control time (DCT) on the order of milliseconds, whereas the DCT of the locally controlled DBA (Fig.
1(a)) is negligible. There have been DBA algorithms to achieve low-latency upstream transmissions with traffic
prediction, particularly concerning the separately/remotely controlled DBA with at most a few milliseconds DCT [5],
or sub-millisecond DCT [6].
(a) Locally controlled DBA (b) Separately controlled DBA (c) Remotely controlled DBA
Fig. 1. PON systems with DBA controlled locally, separately, and remotely.
In this research, we consider the remotely controlled DBA with up to 100-ms DCT in the XG-PON to achieve
cloud-based DBA enabling flexible and scalable coordination of multiple central offices, and propose a low-latency
DBA algorithm with linear prediction using a least square fitting technique (LSFT). The LSFT can be implemented
easily, whereas the method proposed in [5] needs additional protocols. In the simulation, non-predictive DBA is
applied to evaluate the effects of the longer DCT. We assess the upstream throughput and transmission delay in the
XG-PON when linear prediction is utilized for the remotely controlled DBA. Our simulation results show that the
proposed DBA provides low-latency upstream communication compared to non-predictive DBA.
2. DBA with linear prediction using LSFT
The basic mechanism of conventional DBA is shown in Fig. 2(a). The OLT has the DBA function. Request and Grant
messages are exchanged between the OLT and ONUs. The DBA function in the OLT calculates the upstream
bandwidth allocated to each ONU according to a Request message, and sends a Grant message with the allocated
bandwidth information to each ONU. In Fig. 2, the DBA control starts when the OLT receives a Request message and
ends when the OLT sends a Grant message after the DBA calculation. After receiving the Grant message, the ONUs
can send user data based on the Grant message. The DBA cycle is the cycle in which the OLT sends a Grant message
after DBA calculation, and equals the integral multiple of a grant cycle. For example, Fig. 2(a) shows the case in
which the DBA cycle is twice the duration of a grant cycle. The ONUs receive several Grant messages and send data
OLT
(DBA included) ONU
OLT ONU
DBA
Controller
Central Office
OLT ONU
DBA
Controller
Remote Server
© 2018 The Author(s)
Tu3L.1.pdf OFC 2018 © OSA 2018
in each grant cycle within a single DBA cycle. The granted bandwidth in each grant cycle remains constant in a DBA
cycle, which means that the user data are divided into multiple grant cycles within a DBA cycle.
(a) Conventional DBA (b) Remotely controlled DBA with a longer DCT
Fig. 2. DBA mechanisms in the XG-PON.
We assume a FASA-based PON system with a remotely controlled DBA, shown Fig. 1(c). The DBA controller is
located remotely from the OLT to achieve FASA-based flexible operation. The DBA controller receives the Request
message information from the OLT, conducts the DBA calculation according to the request, and sends the calculation
results to the OLT. The remotely controlled DBA has a longer DCT, due to the communication between the OLT and
DBA controller, as shown in Fig. 2(b). The remotely controlled DBA cannot utilize the latest Request message, as the
DCT is longer than a DBA cycle, while the locally controlled DBA can utilize the latest Request message. In order to
predict the latest request from each ONU based on the history of request messages and to achieve more appropriate
bandwidth allocation, we propose the DBA with linear prediction using LSFT.
In the proposed DBA, the last several Request messages, which have been received by the DBA controller, are
utilized for the DBA calculation. We can change and optimize how many Request messages are utilized depending
on communication circumstances, such as DCTs or traffic patterns. The last several requests are defined as R(xi, yi).
The variables xi and yi denote the request number and the number of requests xi, respectively. The predicted request
Rpred is calculated by (1)–(3),
bxa
R
pred
+=
, (1)
−⋅
−=
∑∑∑∑∑∑
======
2
11
2
111
2
1
n
i
i
n
i
i
n
i
ii
n
i
i
n
i
i
n
i
i
xxnyxxxya
, (2)
−⋅
−⋅= ∑
∑∑∑∑
=====
2
11
2
111
n
i
i
n
i
i
n
i
i
n
i
i
n
i
ii xxn
yxyxnb
, (3)
where a and b denote the intercept and slope of the linear approximation line, respectively. The intercept and slope
are calculated by using (2) and (3). Intercept a indicates the either upward or downward tendency of the last several
requests. In the calculation of Rpred, the variable x is determined based on the DCT. The variable x equals the sum of
the number of requests used for prediction and the number of grant cycles within the DCT. After the DBA control,
the allocated bandwidth information is sent to each ONU as a Grant message, and finally, the ONUs can send the user
data. It is expected that the proposed DBA can reflect the dynamic change of the ONUs’ requests when there exists a
longer DCT caused by remote control.
3. Simulation results and discussion
Simulations were performed based on the topology shown in Fig. 1(c) and the XG-PON, whose upstream bandwidth
was 2.5 Gbps. The number of ONUs was set to 32, and the distance between the ONUs and OLT was set to 20 km.
The non-predictive DBA was implemented as a conventional DBA algorithm. In the simulations, the proposed DBA
with linear prediction using LSFT and the conventional DBA were compared. In all cases, three requests were used
for the prediction for simplicity. A DBA cycle was set to 1 ms, which equals eight grant cycles. Exponentially
distributed user datagram protocol (UDP) flows were input to the ONUs in order to assess the effects of not only the
DBA but also of the dynamic change of traffic distribution. We performed the simulations to evaluate both the average
upstream throughput and average transmission delay from the ONUs to OLT in each network load, normalized based
on the upstream bandwidth (2.5 Gbps) of the XG-PON. The simulation results are shown in Fig. 3. Figs. 3(a), 3(b),
3(c), and 3(d) show the simulation results when the DCTs were set to 0 ms, 1 ms, 10 ms, and 100 ms, respectively.
When the DCT was set to 0 ms, 1 ms, or 10 ms, the average throughput was equivalent to the flow rate in both the
conventional and proposed DBA. When the DCT was set to 100 ms, the average throughput of the proposed DBA
was larger than that of the conventional DBA. The transmission delay indicated a remarkable difference between the
Request Grant Data Transmission DBA Control
OLT
ONU
Grant Cycle
DBA Cycle
Grant Cycle
OLT
ONU
DBA
Contro ller
DBA Cycle
・・・
Tu3L.1.pdf OFC 2018 © OSA 2018
conventional and proposed DBA compared to throughput. In particular, with the introduction of linear prediction,
when the DCTs were set to 10 ms and 100 ms, 40.3% and 35.4% of average transmission delays decreased compared
to the conventional DBA, respectively. In the cases in which the DCTs were 0 ms and 1 ms, the average transmission
delays were similar at each flow rate. Therefore, the proposed DBA could effectively reduce transmission delays. This
indicates that the proposed DBA can be utilized in a PON system with remotely controlled DBA, even if traffic
distribution is dynamically changed. However, the transmission delay was largely dependent on the DCT. Therefore,
appropriate parameters, including the number of requests used for prediction and the DBA cycle, should be considered
further for reliable communication.
(a) DCT: 0 ms (b) DCT: 1 ms
(c) DCT: 10 ms (d) DCT: 100 ms
Fig. 3. Simulation results. In the figures, “conv.” shows the conventional DBA, and “prop.” shows the proposed DBA.
4. Conclusion
We demonstrated the concept of the remotely controlled DBA in the XG-PON and proposed a DBA algorithm with
linear prediction using LSFT for the remotely controlled DBA, which can be simply implemented. We assessed the
non-predictive DBA and the proposed DBA under the consideration of the remotely controlled DBA in the FASA-
based PON systems. We confirmed that there were slight differences in upstream throughput, but beneficial
differences in the upstream transmission delay, especially when the DCT was set to above 10 ms. However, the
transmission delay was largely dependent on the DCT. Hence, appropriate parameters should be considered further
for reliable communication.
5. Acknowledgment
This research was supported in part by JSPS KAKENHI Grant Number 16K16049.
6. References
[1] H. Shinohara, “Broadband access in Japan: Rapidly growing FTTH market,” IEEE Commun. Mag., vol. 43, no. 9, pp.72–78 (2005).
[2] ITU-T G.987.3 Recommendation, “10-gigabit-capable passive optical networks (XG-PON): Transmission convergence (TC) layer specification,”
(2010).
[3] F. J. Effenberger, “The XG-PON system: cost effective 10 Gb/s access,” J. Lightw. Technol., vol. 29, no. 4, pp. 403–409 (2011).
[4] J. Kani, M. Yoshino, T. Tanaka, K. Asaka, H. Ujikawa, K. Suzuki, and A. Otaka, “Flexible access system architecture to support diverse
requirements and agile service creation,” in Proc. 43rd European Conf. Optical Communication (ECOC), paper W.3.D.1 (2017).
[5] K. Nishimoto, M. Tadokoro, T. Fujiwara, T. Yamada, T. Tanaka, A. Takeda, and T. Inoue, “Predictive dynamic bandwidth allocation based on
the correlation of the Bi-directional traffic for cloud-based virtual PON-OLT,” in Proc. 2017 IEEE Int. Communications Quality and Reliability
Workshop (CQR) (2017).
[6] R. Yasunaga, A. Walid, and P. Ananth, “Modular dynamic bandwidth allocation for a flexible PON: concept and evaluation,” in Proc. 43rd
European Conf. Optical Communication (ECOC), paper P2.SC8.61 (2017).
0.1
1
10
100
0
10
20
30
40
50
60
70
80
0 0.2 0.4 0.6 0.8 1
Delay (ms)
Throughput (Mbps)
Network Load
0 ms Throughput Conv. 0 ms Throughput Prop.
0 ms Delay Conv. 0 ms Delay Prop.
0.1
1
10
100
0
10
20
30
40
50
60
70
80
0 0.2 0.4 0.6 0.8 1
Delay (ms)
Throughput (Mbps)
Network Load
1 ms Throughput Conv. 1 ms Throughput Prop.
1 ms Delay Conv. 1 ms Delay Prop.
0.1
1
10
100
0
10
20
30
40
50
60
70
80
0 0.2 0.4 0.6 0.8 1
Delay (ms)
Throughput (Mbps)
Network Load
10 ms Throughput Conv. 10 ms Throughput Prop.
10 ms Delay Conv. 10 ms Delay Prop.
0.1
1
10
100
0
10
20
30
40
50
60
70
80
0 0.2 0.4 0.6 0.8 1
Delay (ms)
Throughput (Mbps)
Network Load
100 ms Throughput Conv. 100 ms Throughput Prop.
100 ms Delay Conv. 100 ms Delay Prop.
