Sleep mode mechanism with improved upstream performance for passive optical networks

Abstract

With the proliferation of advanced and interactive services together with the high demand for premium quality of experience, upstream traffic in passive optical networks is growing fast in recent times. Motivated by this, a new power saving mechanism called the sleep-transmit mode (STM) is proposed, which enables the optical network units (ONUs) at customer premises to transmit upstream data during sleep periods without turning on the receiver, thus conserving energy and improving upstream transmission delay. These advantages are achieved by pre-allocating bandwidth for upstream transmission to ONU before it enters the sleep state. Our analysis results show that using STM (1) the power consumption of an ONU is reduced by up to 29% compared to interrupting the sleep mode, and (2) the average additional upstream packet delay due to sleep mode is reduced by up to 10-fold compared to transmitting the upstream packets only after the sleep duration expires.

Full text

Sleep Mode Mechanism with Improved Upstream
Performance for Passive Optical Networks
Jie Li1, N. Prasanth Anthapadmanabhan2, Chien Aun Chan1, Ka-Lun Lee1, Nga Dinh3, and Peter Vetter2
1
Centre for Energy-efficient Telecommunications (CEET), The University of Melbourne, Melbourne, Australia
2.
Bell Labs, Alcatel-Lucent, New Jersey, USA,
3.
Bell Labs Seoul, Alcatel-Lucent, Seoul, South Korea
E-mail: jie.li@unimelb.edu.au
Abstract – With the proliferation of advanced and interactive
services together with the high demand for premium quality of
experience, upstream traffic in passive optical networks is
growing fast in recent times. Motivated by this, a new power
saving mechanism called the sleep-transmit mode (STM) is
proposed, which enables the optical network units (ONUs) at
customer premises to transmit upstream data during sleep
periods without turning on the receiver, thus conserving energy
and improving upstream transmission delay. These advantages
are achieved by pre-allocating bandwidth for upstream
transmission to ONU before it enters the sleep state. Our analysis
results show that using STM (1) the power consumption of an
ONU is reduced by up to 29% compared to interrupting the sleep
mode, and (2) the average additional upstream packet delay due
to sleep mode is reduced by up to 10-fold compared to
transmitting the upstream packets only after the sleep duration
expires.
Keywords - XGPON; cyclic-sleep mode; sleep-transmit mode;
power saving mechanisms; energy efficiency; upstream packet delay
I.
I
NTRODUCTION
In conventional optical access networks, the optical network
units (ONUs) at the customer premises are constantly powered
on and consume a significant amount of power even during
idle periods. Power saving mechanisms based on sleep modes
are proposed in the International Telecommunication Union
(ITU) standards [1, 2] to improve the energy efficiency of a
passive optical network (PON) based on GPON/XGPON.
Meanwhile, consumer traffic has changed greatly since
triple play services emerged with the rapid growth in the
upstream-hungry services, such as picture and video storage in
the cloud, social media, and video chatting. The forthcoming
era of Internet-of-things (IoT) will further boost the demand of
upstream traffic from surveillance cameras, smart grids,
wireless sensor networks, etc. Hence both power consumption
and upstream quality of service need to be seriously
considered when designing the sleep mode mechanism for
next generation energy-efficient optical access networks.
However, the sleep mode mechanisms proposed in recent
literature [3-7] show an unavoidable trade-off between the
power saving and the delay performance. The performance of
the sleep mode can be improved by shortening the wake-up
overhead [5, 8]. However, these approaches require new
circuits and architectures that are not feasible for the already
deployed PON system. In this work, we propose a new sleep
mechanism called the sleep-transmit mode (STM), which
allows ONU to transmit upstream packets during the sleep
period by adding a Transmit state to the currently employed
mechanism in PON standards. STM aims to optimize both
power saving and transmission delay as applicable to upstream
packets.
The ONU state transition and timing diagram for cyclic-
sleep mode in XGPON (G.987.3) [1, 2] are shown in Fig. 1. In
the Asleep state, both the receiver and transmitter of the ONU
are turned off, and in all the remaining states, they are both
turned on. Thus, sleep mode mechanisms generally require
buffering of downstream packets at the OLT and of upstream
packets at the ONU while the ONU is in the Asleep state. Also,
upstream transmissions are not possible in the Asleep state.
Suppose some upstream traffic arrives during T
sleep
, there are
two ways for transmitting upstream traffic when cyclic-sleep
mode is employed:
i) The ONU waits for T
sleep
to expire after which it can send
the upstream packet(s) during the ActiveHeld state; or
ii) The ONU immediately activates a local wake-up indication
(LWI) and transitions to the ActiveHeld state where it can send
upstream packet(s).
Fig. 1: (a) State transition and (b) timing diagram of XGPON cyclic-sleep
mode.
Both of these options suffer from drawbacks. With option 1,
the upstream traffic faces an unnecessary delay due to the wait
for the expiration of T
sleep
. In order to guarantee the QoS for
upstream traffic, the value for T
sleep
needs to be set quite
conservatively which then reduces the power savings that is
actually realized. With option 2, the unnecessary upstream
delay in waiting for T
sleep
to expire is eliminated. However,
since the ONU interrupts the sleep mode, it spends more time
IEEE ICC 2014 - Selected Areas in Communications Symposium
978-1-4799-2003-7/14/$31.00 ©2014 IEEE 3871

in the high power ActiveHeld, ActiveFree and SleepAware(0)
states while waiting for the re-enabling of the low power mode.
Thus, this option results in reduced power savings.
In contrast, STM is specifically designed to address the
above issues. The power saving and upstream delay
performance can be improved by enabling upstream packet
transmission during the low power Asleep state as shown in
Fig. 2. This is achieved by pre-allocating bandwidth for
upstream transmission before an ONU enters the low power
state.
The rest of the paper is structured as follows: Section II
describes our proposed STM mechanism and its operation.
Section III provides mathematical models for performance
analysis and Section IV evaluates the power saving and delay
performance of the proposed mechanism.
$FWLYH)UHH
6OHHS$ZDUH
6OHHS$ZDUH
6OHHS$ZDUHQ
$FWLYH+HOG
$FWLYH+HOG

7LPH
$FWLYH
+HOG
$FWLYH
)UHH
6OHHS
$ZDUH $6OHHS
7DZDUH
H[SLUHV
7VWP H[SLUHV
/:,65$ZDNH
65670
6$2))
RU):,
6$2))RU):,RU/:,65$ZDNH
7KROG H[SLUHV
6$21
7UDQVPLW
7UDQVPLW
7UDQVPLWP

7WUDQVPLW
7UDQVPLWQ
7UDQVPLWQP

7UDQVPLW
7VOHHS
H[SLUHV
7WUDQVPLW H[SLUHV
QWK 670 F\FOH
7VWPQ
$VOHHS

$VOHHS

$VOHHS
P
$VOHHS
Q
$VOHHS
QP
7DZDUH 7DZDUH
7VOHHS
D
E
7VWP
218
2/7
6$21
65670
86'DWD
86'DWD
86'DWD
5HSRUW
'6'DWD
86'DWD
86'DWD
7KROG
7DZDUH 7WUDQVPLW
7VWP
7DZDUH 7WUDQVPLW
F
7VOHHS
6$21
65670
):,
7DZDUH
7VWP
3%:
3%:
7KROG
6OHHS7UDQVPLW0RGH
3RZHU
3DFWLYH
3LGOH
3VOHHS
3WUDQVPLW
7KROG
7WULJJHU
Fig. 2: (a) State transition, (b) timing diagram, and (c) example of the
operation of sleep-transmit mode (STM).
II. P
ROPOSED
S
LEEP
-T
RANSMIT
M
ODE
M
ECHANISM
Sleep-transmit mode (STM) adds a new state called the
Transmit state (see Fig. 2(a)) to the sleep mode state transition
diagram defined by G.987.3. In this Transmit state, the ONU
has the transmitter turned on, but the receiver remains off. The
timing diagram and the operational scheme of STM are shown
in Fig. 2(b) and 2(c), respectively.
The OLT operation includes pre-allocating an appropriate
amount of upstream bandwidth to ONUs before the ONU
transitions to the Asleep state. The pre-allocated bandwidth
(PBW) information consists of one or multiple timeslots
(T
transmit
) when the ONU can transmit an upstream burst of
data during the STM period (T
stm
) and the Asleep state interval
(T
sleep
) between two timeslots. On the ONU side, before
transitioning to the Asleep state, it acquires the PBW
information and it’s available for use by the ONU for any
upstream transmission during the low power period (T
stm
). The
detailed operation is as follows. In the ActiveHeld state, if no
downstream packet is observed for a certain period of time
(T
hold
), the OLT sends the Sleep_Allow (SA(ON)) message
together with the PBW information (T
stm
and T
transmit
) allowing
the ONU to enter into the ActiveFree state. The ONU
acknowledges this by sending a STM Sleep_Request
(SR(STM)) message back to the OLT and immediately
transitions to the SleepAware state. The ONU stays in the
SleepAware state for a period of time (T
aware
). It enters into the
Asleep state if there is no downstream or upstream traffic.
Otherwise, the SleepAware period is interrupted by a forced
wakeup indication (FWI) or local wakeup indication (LWI)
and the ONU returns back to the ActiveHeld state. The main
novelty of this proposed mechanism is to allow the ONU to
initiate one or multiple upstream transmissions by
transitioning to the Transmit state (i.e., T
transmit
) during the T
stm
.
By transitioning between the Transmit and Asleep state, this
low power period T
stm
is not interrupted. After a number of
(T
sleep
+ T
transmit
) cycles, T
stm
expires and the STM ONU returns
to the SleepAware state just as in G.987.3.
Fig. 3: Flowchart of sleep-transmit mode (STM)
If there are multiple ONUs operating in different states with
different T
stm
, the PBW information of each ONU is calculated
by considering the status of other ONUs. Hence, the notion of
a polling cycle (with period of T
p
) is necessary. The OLT
periodically checks and updates the status of ONUs during
each T
p
. Based upon the status of each ONU, the OLT updates
each ONU with different PBW information. This is
summarized in the flowchart shown in Fig. 3. The PBW is
conveyed to the ONU in a common downstream frame
transmitted by the OLT containing a pre-allocated bandwidth
map field, which is referred as PBWmap following ITU-T
Gigabit PONs. The allocation of T
transmit
to the ONUs during
the T
stm
is accomplished using a bandwidth allocation
algorithm similar to traditional DBA [8], but the definition of
IEEE ICC 2014 - Selected Areas in Communications Symposium
3872

such algorithm is beyond the scope of this paper.
Another challenge is the synchronization of upstream
transmissions during the Asleep state. At the instant the ONU
transitions to the Asleep state during T
stm
, it is in fact
synchronized with the OLT. The ONU continues to maintain
its clock and uses the same time reference to initiate an
upstream transmission when transitioning from the Asleep
state to the Transmit state. However, a problem occurs in
which the ONU clock may have drifted with respect to the
OLT clock during T
stm
. A typical crystal has a frequency drift
deviation of smaller than 100ppm [9] which translates to 1μs
of clock drift in 10ms duration of T
stm
. Since this drift is
usually very small, the OLT can take it into account by adding
a guard interval before or after T
transmit
in the PBW to avoid
upstream traffic collision among ONUs.
Fig. 4: Timing diagrams for the upstream transmission in (a) non-interrupted
sleep mode, which waits for T
sleep
expires before transmission, (b) interrupted
sleep mode, which wakes up immediately by LWI, and (c) sleep-transmit
mode (STM).
III. P
ERFORMANCE
A
NALYSIS
In order to evaluate the impact of power consumption and
upstream packet delay in STM, we compare three different
sleep mode mechanisms, as depicted in Fig. 4, for upstream
transmission. We compare the proposed STM mechanism
against the two options mentioned in Section I. Fig. 4
conceptually illustrates the advantages of the proposed STM in
both upstream packet delay and power saving performance.
A. Non-interrupted sleep mode
As shown in Fig. 1, the OLT buffers the downstream traffic
for each ONU and only transmits the buffered-packets during
a pre-determined activity slot [3]. The upstream packets are
buffered and are sent during a specific timeslot after the ONU
is out of the sleep duration.
As shown in Fig. 4(a), in a conventional fixed cyclic-sleep
mode, the ONU will not be interrupted during the sleep
duration (T
sleep_a
). During the SleepAware state, the ONU
wakes up to check the traffic condition and control messages
from the OLT. The ONU enters the ActiveHeld state by either
receiving the FWI from the OLT for any downstream packet
or triggering the LWI locally for any upstream packet. The
ONU goes back to the Asleep state if there is no downstream
or upstream packet. Otherwise, it returns to the ActiveHeld
state and starts to receive/transmit the packets. Since the time
used to send/receive the wake-up indication is almost
negligible, the time spent in the T
aware
is mainly constrained by
the wake up overhead (T
wake
), i.e., synchronizing the
downstream data. T
wake
ranges from 0.6 ms to 14 ms [10]. In
our analytical model, we assume T
aware
=T
wake
and the first
SleepAware period T
aware
(0) together with T
hold
is the
triggering time (T
trigger
= T
hold
+T
aware
(0)).
We assume that the packet arrivals in both downstream and
upstream are independent Poisson processes with rates Ȝ
ds
and
Ȝ
us
respectively, resulting in an overall packet arrival rate (Ȝ)
of Ȝ
ds
+Ȝ
us
. If only upstream traffic is considered, the
probability of no packet arriving is given as exp(-Ȝ
us
T
trigger
)
during T
trigger
and exp(-Ȝ
us
(T
sleep_a
+T
aware
)) during a sleep cycle.
In other words, the probability (pro(n)) of having at least one
packet arrived at the n-th sleep cycle while no packet arrived
during the previous (n-1) sleep cycle(s) is given as [11]
()
_
1
0
() 1
1
1
us trigger
us aware
sleep a
T
TT
n
nism
n
pro n pro n
e
e
λ
λ
−
−+
−
=
°
=®ªº
⋅≥
°«»
¬¼
¯
−
−
, (1 )
where pro
n-1
is the probability of no packet arrived in previous
cycles, which is given as
()
()
_
1
us aware
sleep a
us trigger
nT T
T
ee
λ
λ
−− +
−
⋅
.
The average power consumption of an ONU under non-
interrupted sleep mode mechanism (P
nism
) is listed in Table I,
where P
active
is the power consumption of transmit/receive
packets during the ActiveHeld state, P
idle
denotes the idle
power consumption during T
trigger
in the ActiveHeld state and
the first SleepAware(0) state, P
sleep
is the power consumption
of the Asleep state, and P
cycle
is the total power consumption
during one sleep cycle, which is given as T
sleep
·P
sleep
+
T
aware
·P
idle
. T(0) represents the expected arrival time of the
first packet that arrives during the idle period T
trigger
. Based on
[12], T(0) is given as
1
(0) 1
us trigger
us trigger
T
trigger
T
us
e
TT
e
λ
λ
λ
−
−
=− ⋅
−
. (2)
T
T
denotes the packet transmission time, i.e., the time for an
ONU to finish transmitting backlogged packets. For n=0, as
long as one packet arrived, the ONU is interrupted from T
trigger
IEEE ICC 2014 - Selected Areas in Communications Symposium
3873

and transitions to (or remains in) the ActiveHeld state. During
the packet transmitting time (T
T
(0)), the new packets arriving
during this period should also be considered. If the ONU is
able to enter into the Asleep state (n1), T
T
becomes a function
of the upstream buffered backlog of the ONUs and its link
access rate. Only those packets arrived during the last sleep
cycle (Ȝ
us
(T
sleep_a
+T
aware
)) should be considered. If the packet
forwarding rate of the ONU, denoted as ȝ, is assumed to be
the same as its link access rate divided by the packet size, T
T
for non-interrupted sleep mode (T
T_nism
) is given as
()
()
()
_
_
_
1
10
() 11
1
us aware
sleep a
us rrt
us us
Tnism
us sleep a aware
us rrt
TT
us us
Tn
Tn TT Tn
e
λ
λμλ
λλμλ
−+
+=
°−
°
®ªº
+
°«»
+⋅ ≥
°−
«»
−
¬¼
¯
=
. (3)
The average additional upstream packet delay is the average
delay occurs when an upstream packet arrives at the ONU
during the sleep period before a transmission opportunity
becomes available. The average additional upstream packet
delay shown in Fig. 4(a) for non-interrupted sleep mode (D
nism
)
is depicted in Table I. It is noted that the delay due to the
queuing and the propagation delay during the active state are
not accounted in the delay calculation. Hence D
nism
is
considered as 0 when n=0. Although Fig. 4(a) appears to have
a low average power consumption since it does not interrupt
the Asleep state (T
sleep_a
), the upstream traffic experience a
large additional packet delay in waiting for the Asleep state
(T
wait_a
) and the SleepAware state (T
aware
) to expire, and then
transitions to the ActiveHeld state to initiate the upstream
bandwidth allocation. This allocation includes at least one
round-trip-time (T
rtt
). In the model, it is considered as 200 μs
for network reach of 20 km [2].
B. Interrrupted sleep mode
In Fig. 4(b), any upstream arrival immediately activates the
LWI, which transitions the ONU to the ActiveHeld state with
only a small overhead. The overhead includes the wakeup
overhead for the transceiver (T
wake
) to recovery from the
Asleep state and a small amount of time for the ONU waiting
to send the Sleep_Request (SR(Awake)) back to OLT. After
transitioning to the ActiveHeld state, the ONU spends T
rtt
waiting for the upstream bandwidth allocation. This allows the
upstream delay to be reduced as compared to the case of non-
interrupted sleep mode as shown in Fig. 4(a).
However, since the low power mode is interrupted, this
option leads to a higher opportunity of entering higher power
consumption states (ActiveHeld, ActiveFree, and SleepAware
Table I: The power and delay analysis of the power saving options shown in Fig. 4
ONU power consumption
Non-interrupted sleep mode (fixed cyclic-sleep mode)
()
()
_ _
1
___
(0) (0) ( )
(0) ( )
(0) (0) ()
rtt idle T nism active trigger idle cycle rtt idle T nism active
nism nism nism
n
rtt T nism trigger sleep a aware rtt T nism
T TPT P TPnP TPT nP
Ppro pron
TTT TnT T TTn
∞
=
ªº
ªº
++ +++
«»
+
«»
++ ++++
«»
«»
¬¼
¬¼
=
¦
Interrupted sleep mode
() ()
()
_ _
1
__
(0) (0) ( 1) ( ) ( )
(0) ( )
(0) (0) (1) ()
rtt idle T ism active trigger i dle cycle sleep wake rtt idle T ism active
ism ism ism
n
rtt T ism trigger sleep a aware wak
T TPT P TP nP TnP T TPT nP
Ppro pron
TTT TnTTTnT
∞
=
′′ªº
++ ⋅ +−+ +++
′
+
«»
++ ′
+− + + +
«»
¬¼
=
¦
()
_
_
1__
()
()
( ) ()
erttTism
trigger idle cycle rtt idle T ism active
ism
ntrigger sleep a aware rtt T ism
TT n
TPnP TPT nP
pro n TnT T TTn
∞
=
ªº
«»
′
++
«»
¬¼
ªº
′′
++ + ⋅
′′ «»
+′′
++++
«»
¬¼
¦
Sleep-transmit mode
()
()
()
__
_
1
_
(1) () ()
(0) (0)
(0) ( )
(0) (0)
trigger idle c stm T stm sleep T stm transmit aware idle
rtt idle T st m active
stm stm stm
n
rtt T stm trigger stm aware
TP nPTT nP T nP TP
TTPT P
Ppro pron
TTT T nTT
∞
=
ªº
+− + − + +
ªº
++ «»
+
«»
++ + +
«»
«»
¬¼
¬¼
=
¦
Average additional upstream packet delay
Non-interrupted sleep mode (fixed cyclic-sleep mode)
_
1
() 2
sleep a aware
nism nism rtt
n
TT
Dpron T
∞
=
+
§·
=⋅ +
¨¸
©¹
¦
Interrupted sleep mode
()
11
() () 2
wake
ism ism wake rtt ism rtt
nn
T
DpronTTpron T
∞∞
==
§·
′′′
=⋅++⋅+
¨¸
©¹
¦¦
Sleep-transmit mode
_
1
() 2
sleep c
stm stm
n
T
Dpron
∞
=
§·
=⋅
¨¸
©¹
¦
IEEE ICC 2014 - Selected Areas in Communications Symposium
3874

states) for a certain upstream traffic load. The analytical
models for the average power consumption (P
ism
) and the
average additional upstream packet delay (D
ism
) of an ONU
under interrupted sleep mode mechanism are listed in Table I,
respectively. The probability of an upstream packet arrived
during T
trigger
, T
sleep_a
, and T
aware
, (pro
ism
) are given as
()
()
_
_
_
1
1
() 0
( ) 1 during
( ) 1 during
1
1
1
us trigger
us sleep a
us sleep a us aware
T
ism
T
ism sleep a
TT
ism aware
n
n
pro n n
pro n pro n T
pro n pro n T
e
e
ee
λ
λ
λλ
−
−
−−
−
−
==
′=⋅ ≥
′′ =⋅ ⋅ ≥
−
−
−
. (4)
The upstream packet transmission time (T
T_ism
) during these
three states are given as
()
()
_
_ _
_
1
() 1 0
1
( ) 1 1 during
1
( ) 1 during
1
us wake
T ism us rtt
us us
T ism us wake rtt sleep a
us us
us wake
T ism us rtt aware
T
us us
Tn T n
Tn TT n T
T
Tn T n T
e
λ
λμλ
λμλ
λλμλ
−
+=
−
′=ª++º⋅≥
¬¼
−
§·
′′ =+⋅ ≥
¨¸
−
−
©¹
=
.(5)
It is noted that T
wake
= T
aware
in the last term of our model.
Since T
wake
cannot be interrupted by an upstream arrival, a
complete P
cycle
has to be accounted if the upstream packets
arrived during T
wake
. The expected arrival time of the first
packet that arrived during these three states (T) are given as
__
_
_
1
() 0
1
1
( ) 1 during
1
1
( ) 1 during
1
trigger
us
sleep a sleep a
us
aware aware
us
trigger
trigger
sleep a
sleep a
aware
aware
T
T
T
T
T
T
us
us
us
us
us
us
e
Tn T n
e
e
Tn T n T
e
e
Tn T n T
e
λ
λ
λ
λ
λ
λ
λ
λ
λ
−
−
−
−
−
−
=− ⋅ =
−
′=− ⋅ ≥
−
′′ =− ⋅ ≥
−
. (6)
C. Sleep-Transmit Mode (STM)
In comparison, STM (as shown in Fig. 4(c)) enables a
reduction of upstream delay. The average additional upstream
packet delay for the STM (D
stm
) is listed in Table I. The
upstream bandwidth pre-allocation eliminates the need for
downstream data recovery during T
stm
. Hence the wakeup
overhead for the transmitter to transition from the Asleep state
to Transmit state is with a value of less than a microsecond,
which mainly depends on the laser ON/OFF time [8, 13]. Thus,
the transmitter of the ONU can potentially be turned ON/OFF
as required during any of the states. T
wait_c
shown in the Fig.
4(c) is the time that the ONU waits to enter the Transmit state
while it is still in the Asleep state. On average, for a fixed
T
sleep_c
, T
wait_c
is equal to T
sleep_c
/2 and the maximum length of
each T
sleep_c
during the STM period (T
stm
) can be approximately
given as
_
_
()
1
stm T stm
sleep c
transmit
TT n
TN
−
=+
, (7)
where N
transmit
is the number of T
transmit
state within a STM
duration (T
stm
). In order to compare with the interrupted sleep
mode as shown in Fig. 4(b), a reasonable value of N
transmit
is
chosen to provide a similar upstream delay D
stm
with respect to
D
ism
, where T
wait_c
is approximately equal to T
wake
+T
rtt
. T
T_stm
during each state is given as
()
()
()
_
1
10
() 11
1
us stm aware
us rtt
us us
Tstm
us stm aware
TT
us us
Tn
Tn TT n
e
λ
λμλ
λ
μλ
−+
+=
°−
°
®ª+º
°⋅≥
«»
°−
−
¬¼
¯
=. (8 )
Since the low power period (T
stm
) is not interrupted, the
power saving is much higher compared to the interrupted sleep
mode. The power consumption of an ONU with STM (P
stm
) is
listed in Table I, where P
c
is the total power consumption of a
T
stm
cycle without any upstream transmission, which is given
as T
stm
·P
sleep
+T
aware
·P
idle
. P
transmit
is the power consumption in
the Transmit state. During this state, only the transmitter is
turned ON at the ONU.
IV. N
UMERICAL
R
ESULTS
Fig. 5 illustrates the power consumption (in %) relative to
the full power consumption and the average additional
upstream packet delay of an ONU with the non-interrupted
sleep mode and with the proposed STM. The parameters used
in analytical models are listed in Table II.
3RZHUFRQVXPSWLRQSHUFHQWDJH
$YHUDJHDGGLWLRQDOXSVWUHDP
SDFNHWGHOD\ PV
Fig. 5: Power savings and average additional upstream packet delay
performances of the ONU with non-interrupted sleep mode and with sleep-
transmit mode (STM).
Table II: Key parameters for performance analysis
Parameter Value Parameter Value
T
trigge
r
2ms T
rtt
0.2ms
T
aware
1ms N
stm
10
T
sleep_a
20ms P
sleep
23%×P
active
T
stm
20ms P
idle
74%×P
active
T
wake
1ms P
transmit
60%×P
active
ȝ
us
2.5Gbps
Packet size
1518B
The ONU with the proposed STM demonstrates slightly
lower power consumption compared to the non-interrupted
sleep mode. For the presence of any upstream packet, the
ONU with the non-interrupted sleep mode is required to
transition to the ActiveHeld state before it re-enters into the
low power state. Transitioning into the ActiveHeld state also
incurs additional power consumption consumed during T
rtt
and
T
trigger
. In contrast, once the ONU with STM transitions into
the low power period (T
stm
), it does not require to return to the
ActiveHeld state to transmit upstream packets. Hence, at the
low traffic condition, the ONU can always remain in T
stm
to
conserve power.
IEEE ICC 2014 - Selected Areas in Communications Symposium
3875

In terms of the additional upstream delay performance, the
non-interrupted sleep mode suffers from a large delay at low
traffic due to the fact that the ONU needs to wait for the
expiry of the sleep duration. As shown in Fig. 5, with the same
sleep duration, the ONU with STM can attain up to 10 fold of
upstream packet delay reduction (from 10.6 ms to 1 ms)
compared to non-interrupted sleep mode at low upstream
traffic. In order to guarantee the QoS for upstream delay, the
sleep duration (T
sleep_a
) of the non-interrupted sleep mode
would need to be set conservatively. As shown in Fig. 5,
shortening T
sleep_a
improves the delay performance. However,
it also results in a significant power consumption increase
(15%) at low traffic load.
Fig. 6 shows the power saving and the average additional
upstream packet delay of an ONU with the interrupted sleep
mode and an ONU with the proposed STM. The ONU with the
interrupted sleep mode shows relatively poorer power saving
performance as expected. The proposed STM also
demonstrates its advantage in providing almost the same
upstream packet delay compared to the interrupted sleep mode.
Fig. 6: Power savings and average additional upstream packet delay
performances of the ONU with interrupted sleep mode and with sleep-
transmit mode (STM).
Fig. 7: Power saving performance of an ONU with sleep-transmit mode (STM)
with shortened T
trigger
.
One possible way to enhance the power saving performance
of the ONU with STM is to shorten the triggering period
(T
trigger
=T
hold
+T
aware
(0)). Since the ONU with STM is able to
transmit upstream packet during the STM period, T
trigger
could
be set as short as possible without affecting the upstream delay
performance. Fig. 6 and Fig. 7 show the power saving
comparison of the ONU with STM and with interrupted sleep
mode for T
trigger
of 2ms and 0.1ms, respectively. The power
saving is improved from 11% to 29% by shortening T
trigger
.
V. C
ONCLUSIONS
With the rapid growth of upstream traffic from user devices
in passive optical networks, it is important to emphasize the
upstream performance (i.e., power consumption and delay)
when sleep mode mechanisms are employed at the ONUs.We
propose a new sleep mechanism called sleep-transmit mode
(STM) that provides a solution addressing this concern by
adding a Transmit state to the currently employed mechanism
in PON standards. This idea enables the ONU to transmit its
upstream packets without interrupting the low power state.
Compared to non-interrupted fixed-cycle sleep mode, the STM
demonstrates a similar power saving performance while at the
same time provides up to a 10-fold upstream packet delay
reduction. When compared to interrupted sleep mode, which
provides the lowest upstream packet delay, STM achieves up
to 29% power reduction while also providing comparable
delay performance for upstream packets. Moreover, STM is
compatible with and can be integrated into current PON
standards such as XGPON.
A
CKNOWLEDGEMENTS
This work is supported by Alcatel-Lucent Bell Labs,
Victoria State Government of Australia, and the Seoul
Metropolitan Government R&BD Program WR080951.
R
EFERENCES
[1] ITU-T, "GPON power conservation," in Series G Supplement 45:
Transmission systems and media, digital systems and networks, ed,
2009.
[2] ITU-T, "Transmission convergence (TC) specifications," in G.987.3:
10-Gigabit-capable passive optical networks (XG-PON), ed, 2010.
[3] A. Dixit, et al., "ONU power saving modes in next generation optical
access networks: progress, efficiency and challenges," Opt. Express,
vol. 20, pp. B52-B63, 2012.
[4] S. Lei, et al., "Energy-efficient PON with sleep-mode ONU: progress,
challenges, and solutions," IEEE Network, vol. 26, pp. 36-41, 2012.
[5] W. Shing-Wa, et al., "Sleep Mode for Energy Saving PONs:
Advantages and Drawbacks," in GLOBECOM Workshops, 2009 IEEE,
2009, pp. 1-6.
[6] A. R. Dhaini, et al., "Toward green next-generation passive optical
networks," IEEE Communications Magazine, vol. 49, pp. 94-101, 2011.
[7] J. Zhang and N. Ansari, "Toward energy-efficient 1G-EPON and 10G-
EPON with sleep-aware MAC control and scheduling," IEEE
Communications Magazine, vol. 49, pp. s33-s38, 2011.
[8] B. Skubic, et al., "A comparison of dynamic bandwidth allocation for
EPON, GPON, and next-generation TDM PON," IEEE
Communications Magazine, vol. 47, pp. S40-S48, 2009.
[9] Improving the Accuracy of a Crystal Oscillator [Online]. Available:
www.semtech.com
[10] L. Valcarenghi, et al., "Energy efficiency in passive optical networks:
where, when, and how?," IEEE Network, vol. 26, pp. 61-68, 2012.
[11] N. Dinh and A. Walid, "Power saving protocol for 10G- EPON
systems: A proposal and performance evaluations," in Proc. IEEE
Global Communications Conference (GLOBECOM) 2012, pp. 3135--
3140, 3-7 Dec. 2012.
[12] N. P. Anthapadmanabhan et al., "Analysis of a probing-based cyclic
sleep mechanism for passive optical networks," in Proc. IEEE Global
Communications Conference (GLOBECOM) 2013, 9-13 Dec. 2013.
[13] A. G. Weber, et al., "Measurement and simulation of the turn-on delay
time jitter in gain-switched semiconductor lasers," IEEE Journal of
Quantum Electronics, vol. 28, pp. 441 - 446, 1992.
IEEE ICC 2014 - Selected Areas in Communications Symposium
3876