Second, the ONUs can be allowed to sleep when the traffic is
low. In reality, devices in access networks, such as ONUs, tend to
be underused by approximately 15% [12]. Because the traffic oc-
curs suddenly and dynamically, the ONUs do not keep busy all of
the time. Consequently, the ONUs can go into sleep modes under
low or zero traffic conditions. Refs. [13,14] proposed algorithms
for ONU sleep when the traffic load is lower than the low-level
threshold, at which point another sleeping ONU will be woken
up when the network traffic load is higher than the high-level
threshold. In this case, the traffic designated to the sleeping ONU
is rerouted to other active ONUs.
Third, energy-efficient protocols can be designed in EPON. This
method designates the sleep start time and wakeup time for ONUs.
Refs. [15,16] proposed two ONU sleep schemes, named Upstream
Centric Scheduling (UCS) and Downstream Centric Scheduling
(DCS), which are based on the use of a Multipoint Control Protocol
(MPCP) in EPON. In UCS, the downstream traffic load only can be
transmitted during the upstream transmission slot. If the down-
stream traffic is significantly more than the upstream traffic, a
large delay is caused for the downstream traffic due to queuing.
In DCS, if any traffic exists either upstream or downstream, the
ONU should not go to sleep. Therefore, DCS reduces the down-
stream traffic delay while consuming more energy than UCS.
In fact, 60–70% of the ONU power consumption is due to the
transceiver and back-end circuit [11]. Hence, allowing the trans-
mitter and receiver to sleep is another possible approach for
ONU energy-saving management. Ref. [17] proposed four power
levels for ONUs, achieved through the use of different transmitter
and receiver states. Two sleep mode scenarios for ONUs are pro-
posed, including multiple Dynamic Bandwidth Allocation (DBA)
cycles and one DBA cycle. In the multiple DBA cycles scenario,
the transmitter and receiver are allowed to sleep for more than
one cycle only if the sleep duration is no more than the maximum
tolerated time. In the one cycle scenario, the transmitter goes to
sleep after completing the transmission, and it is woken up via
the ONU MAC protocol. However, sleeping the receiver is signifi-
cantly more complicated because the receiver does not know the
condition of the downstream traffic. Hence, the transmission inter-
val of an ONU in the downstream is designed to be determined by
the sum of the downstream traffic of all other ONUs. Then, the re-
ceiver can sleep for a period based on a function of the interval
time.
Energy-saving problems also have been studied in various
PONs. Refs. [18–20] noted that a 10G-EPON consumes more energy
than a 1G-EPON due to the significantly higher line rate. Based on
this idea, they developed a single-threshold policy [18,19] and a
dual-threshold policy to prevent frequent switching of the link
rates [20]. In addition to adding the adaptive link rate, the cyclic
sleep mode is also used as one of the energy-saving modes that
are specified in the ITU-T Recommendation G. sup45, while the
other is doze mode [21]. Further, ITU-T G.984 for Gigabit PON
(GPON) specifies the ONU sleep modes for four types, including
the ONU power shedding mode, the ONU dozing mode, the ONU
fast/cyclic sleep mode and the ONU deep sleep mode [22]. In the
first mode, beyond the optical link, some functions and services
can be turned off. The power consumption in this mode is the high-
est among the four modes. In the ONU dozing mode, the transmit-
ter can be turned off without the presence of upstream data. This
mode consumes the second highest power after the ONU power
shedding mode. The ONU fast/cyclic sleep is defined by allowing
the ONUs to stay in a sequence of sleep cycles. Its power consump-
tion is less than that of the ONU dozing mode. The ONU deep sleep
mode consumes the least power because both the transmitter and
the receiver can be turned off. Meanwhile, the network perfor-
mances are also most seriously degraded in this mode. Recently,
a method for combining the four modes has appeared. An ONU
deep sleep mode combined with a dozing mode was proposed to
achieve the maximum possible energy saving in Ref. [23].
Although Refs. [13–16,18–23] studied various energy-saving
methods, they did not focus on energy consumption in the inde-
pendent components of the ONUs. In Ref. [17], although the energy
consumption of the transmitter and receiver was considered, the
performances including the packet delay were not addressed.
Therefore, an energy-saving scheme that considers both the com-
ponents of the ONU and the packet delay remains a challenge,
and this challenge has motivated our work.
In this paper, we consider the energy-saving components of the
ONUs, which include the transmitter, the receiver and the ONU in
its entirety. We first propose an algorithm and a rule for down-
stream packet scheduling under inter- and intra-ONU conditions
to ensure that the real-time packet that arrives earliest also can
be sent first and that all real-time packets can be sent before the
non-real-time packets. Then, we develop a hybrid sleep mode for
ONUs that considers the combination of ONU deep sleep and inde-
pendent sleep for transmitters and receivers. To realize this combi-
nation, a modified GATE control message is designed with 10 time
points for the sleep processes of the transmitter, the receiver and
the entire ONU. Based on the packet scheduling algorithm and rule
in addition to the hybrid sleep mode, we propose an ONU sleep
scheme named the Energy-Saving scheme based on downstream
Packet Scheduling (ESPS), which minimizes the energy consump-
tion of ONUs and reduces the packet delay as much as possible.
To the best of our knowledge, the work in this paper is the first
study to combine the independent sleep mode for transmitters and
receivers with the ONU deep sleep mode, and this work also is the
first study to design the ONU sleep scheme by scheduling both in-
ter- and intra-ONUs. The rest of the paper is organized as follows:
Section 2 gives the problem statement; Section 3 presents the
scheduling of downstream packets; Section 4 proposes the hybrid
sleep mode; Section 5 describes the proposed ESPS scheme in de-
tail; Section 6 details the simulation and its analysis; and Section 7
concludes this paper.
2. Problem statement
2.1. Assumptions and notations
In the ONU deep sleep mode, both the transmitter and receiver
are turned off, while the other components (e.g., the digital cir-
cuitry) remain active at all times. Therefore, the power of the
ONU in a deep sleep state is produced by the always-active compo-
nents. When the transmitter/receiver remains active, the always-
active components are also working. We assume there are K ONUs
in a given EPON, and enable tx and rx to denote the transmitter and
the receiver of ONU, respectively. We let acc denote the always-ac-
tive components. Because the energy-saving components that we
consider in this paper include tx, rx and acc, we give the relative
parameters for those components in the following list.
i: The index of the ONU, 1 6 i 6 K.
N: The maximum polling cycle that OLT polling ONUs in the
simulation time.
n: The index of the polling cycle, 1 6 n 6 N.
E
wakeup
: The energy consumption of the wakeup process for an
ONU.
P
tx
, P
rx
: The power consumed by the transmitter and the recei-
ver while active, respectively.
P
sleep
: The power consumed when an ONU is in a state of deep
sleep.
T
n
tx;i
: The active duration of the transmitter of an ONU
i
in the nth
cycle, 1 6 i 6 K.
170 L. Zhang et al. / Optical Fiber Technology 19 (2013) 169–178