Вы находитесь на странице: 1из 8

A Study of Green Optical Networks

Meera Vasudevan,1 School of Electronic and Electrical Engineering, Faculty


of Engineering

ABSTRACT
Green networking is fast becoming popular amongst researchers, as well as service
and network providers. This paper considers Optical Burst Switching (OBS) networks
and proposes an energy saving algorithm to reduce the energy consumption of ICT
equipment. Dynamic sleep cycles are implemented in the network such that some of
the wavelengths associated with dynamically selected links are switched off
according to prevailing traffic load conditions, all the while maintaining a trade-off
between energy consumption and Quality of Service (QoS). Blocking probability
results and the amount of saved energy obtained from the simulation results are
analysed and presented.

Keywords: green networking; OBS; algorithm; dynamic; QoS; blocking probability

1
email address for correspondence: sykretkeeper@hotmail.com
A Study of Green Optical Networks
INTRODUCTION
The energy consumption of backbone networks continues
continue to increase with the rising
numberr of bandwidth intensive applications such as video conferencing and highhigh-
definition IPTV. There is an additional concern of heat dissipation from the limited
locations in the backbone network which deals with energy consumption. Recently,
significant research
arch has been carried out to reduce the energy consumption of IP
over WDM backbone networks [1] [2].

This paper considers an Optical Burst Switching (OBS)( architecture for IP


IP-over-WDM
networks. The IP over WDM backbone networks network [3, 4] consists of two laye
layers: the IP
layer and the optical
ptical layer as shown in Figure 1. The IP layer consists of a core IP
router connected to an optical switch node with the help of short-reach
short reach interfaces.
The IP layer is responsible for the aggregation of data traffic from low end routers.
The optical layer helps the IP routers to communicate with each other. Physical fibre
links containing multiple fibres are used to interconnect optical switch nodes.
Wavelengths travelling through each fibre undergo multiplexing/de-multiplexing
multiplexing/de multiplexing wwith
the help of wavelength multiplexers/de-multiplexers.
multiplexers/de multiplexers. Transponders are associated
with each wavelength and help transmit data. In order to maintain the signal quality in
long distance transmissions, Erbium Doped Fibre Amplifiers (EDFA) are deployed on
fibre links. All these components
onents consume energy. Table 1 shows the energy
consumed by these components.

Figure 1:: IP over WDM optical network

It is clear from Table 1 that a router port is the most energy consuming component in
a node. Therefore switching
hing off router ports will save energy more than any other
component in the network. In this work we propose a dynamic sleep cycle where
wavelengths and therefore routers are switched on and off dynamically, accordin
according to
the network status. Figure 2 shows the flow chart of the algorithm. In the algorithm,
the burst dropping probability and traffic load on different links are monitored.
According to the required QoS level, a burst dropping probability threshold is set. The
higher the dropping probability threshold,
threshold, the higher amount of saved energy. If the
overall burst dropping probability is less than a certain threshold, the M links with the
lowest load are selected to switch off half of their wavelengths. On the other hand, if
the network dropping probability
probability is higher than the dropping probability threshold, all
the wavelengths on one of the sleeping links are switched on to improve the blocking
probability.

Energy consumption of a router port 1000 W


Energy consumption of an optical switch 85 W
Energy consumption of a multiplexer or a de-
16 W
multiplexer using renewable energy
Energy consumption of a transponder 73 W
Electricity energy consumption of an EDFA 8W

Table 1: Energy consumption of different components

Without the energy saving algorithm, all wavelengths in the network are switched on
all the time, i.e. all router ports are switched on. The energy consumption of router
ports in the network is given as:

E ports  2  PR  L  W
(1)

where PR is the energy consumption of a router port, L is the total number of links in
the network, W is the number of wavelengths in each link.

With the energy saving algorithm, the energy consumption of router ports is given by:

L W
E ports  2  PR     ij
i 1 j 1
(2)

where δij=1 if wavelength j on link i is switched on, δij=0 otherwise.


Figure 2: Algorithm flow chart

Performance Evaluation
Simulation is carried out to evaluate the performance of the proposed dynamic
sleeping wavelength algorithm. The National Science Foundation (NSF) network
topology, depicted in Figure 3, is considered as an example of a real world network.
Figure 3: NSF network

The NSF network topology consists of 14 nodes and 21 bidirectional links. The
network is investigated under different number of data channels (16, 32 and 64). Two
control channels are used. The wavelength rate is assumed to be 10 Gb/s. The
transmission rate of nodes is set to 10 Gb/s. All nodes are assumed to be equipped
with full wavelength conversion capability.

The best-fit scheduling algorithm and the just enough time (JET) signalling protocol
are implemented to set OBS reservations for SAN traffic. For OBS the following
assumptions are made: Each burst is allowed 10 hops before its slack time expires.
An offset time of 20 µs is assumed for all bursts. The BCP processing time at each
node is 2.5 µs and data burst switching time is 10 µs. The burst arrival is assumed to
follow a Poisson process and the length of the burst is exponentially distributed with
average burst size of 1 MB. The performance of the network is evaluated under
varying levels of traffic loads. Figure 4 shows the energy consumption with and
without the energy saving algorithm.

250
Energy Consumption (kwh)

200

150

100

50

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Network Load
without energy saving
with energy saving, burst dropping probability threshold=0.03
with energy saving, burst dropping probability threshold=0.05
with energy saving, burst dropping probability threshold=0.1

Figure 4: Energy Consumption under different values of the burst dropping probability
threshold
Different dropping probability thresholds are examined: 0.03, 0.05, and 0.1.The size
of the traffic monitoring window is set to 0.2 seconds and M=6. It is clear from the
figure that the algorithm has succeeded in saving energy for loads up to 0.8 with a
dropping probability threshold of 0.05 compared to the energy consumption without
the energy saving algorithm where all the wavelengths are switched on (in this case
the energy consumption is calculated from equation 1 as 224 kWh). For loads up to
0.8 the network can accommodate its traffic while some wavelengths are off. Under a
load of 0.1 the maximum saving is achieved as the maximum number of wavelengths
(half of the wavelengths on all the M links) is switched off during the whole simulation
course. Energy consumption in this case is calculated from equation 2 as 176 kWh
i.e. about 21% of the energy is saved.

As the load increases, the amount of energy saved decreases as fewer wavelengths
are switched off. However for loads higher than 0.8, the network becomes heavily
loaded therefore the algorithm fails to switch any wavelengths off. The figure also
compares the energy consumed under the different dropping probability thresholds. It
is clear that as we increase the threshold, more energy is saved as the algorithm will
be able to switch more wavelengths off.

Saving energy comes at the price of increasing the burst dropping probability as a
number of wavelengths will be switched off under the energy saving algorithm. Figure
5 compares the burst dropping probability with and without the energy saving
algorithm. Different dropping probability thresholds are also examined. It is clear that
there is no difference in the dropping probability between the two curves at very high
loads (loads higher than 0.8) because the network blocking probability is higher than
the blocking probability threshold which results in not switching any nodes off. At a
very low load (load=0.1) the difference in the dropping probability is very small as
switching nodes off does not affect the performance seriously due to the low
congestion levels at that load. The difference is clear loads between 0.2 and 0.8.

0.3

0.25
Burst Dropping Probability

0.2

0.15

0.1

0.05

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Network Load
without energy saving
with energy saving, burst dropping probability threshold=0.03
with energy saving, burst dropping probability threshold=0.05
with energy saving, burst dropping probability threshold=0.1

Figure 5: Burst dropping probability under different values of the burst dropping
probability threshold

Figures 6 and Figure 7 show the energy consumption and the burst dropping
probability under different monitoring window sizes: 0.1, 0.2 and 0.3 seconds. We
assume that the burst dropping probability threshold is set to 0.05. It is shown in
Figure 6 that the energy consumption decreases as the monitoring window
decreases as monitoring the network more frequently allows the energy saving
algorithm to quickly take advantage of any decrease in the dropping probability. As
expected in Figure 7 the burst dropping probability increases as the monitoring
window decreases.

250

200
Energy Consumption (kwh)

150

100

50

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Network Load
without energy saving
with energy saving, monitoring window size=0.1s
with energy saving, monitoring window size=0.2s
with energy saving, monitoring window size=0.3s

Figure 6: Energy Consumption under different monitoring window sizes

0.3

0.25
Burst Dropping Probability

0.2

0.15

0.1

0.05

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Network Load
without energy saving
with energy saving, monitoring window size=0.1s
with energy saving, monitoring window size=0.2s
with energy saving, monitoring window size=0.3s

Figure 7: Burst dropping probability under different monitoring window sizes

Conclusion
This paper has proposed a dynamic wavelength sleep cycles algorithm for energy
efficiency. It is clear from the simulation results that the proposed dynamic
wavelength sleep cycles algorithm for energy efficiency has succeeded to save
energy for loads up to 0.8 while maintaining acceptable levels of burst dropping
probability. It is also clear that selecting the suitable size of the dropping probability
threshold and the monitoring window size affects the energy consumption of the
algorithm.

ACKNOWLEDGMENTS
I would like to sincerely thank Professor Jaafar Elmirghani for his invaluable
expertise, guidance and encouragement. I would also like to thank Taisir El-Gorashi
for her help towards the completion of this work. Finally, I would like to thank my
family, especially my mother, for their constant encouragement and support.

REFERENCES
[1] Nedevschi, S., L. Popa , G. Iannaccone , S. Ratnasamy , D. Wetherall, ‘Reducing
network energy consumption via sleeping and rate-adaptation’, Proceedings of the
5th USENIX Symposium on Networked Systems Design and Implementation, p.323-
336, April 16-18, 2008, San Francisco, California

[2] Shen, G. and R. S. Tucker, ‘Energy-Minimized Design for IP Over WDM


Networks,’ Optical Communication and Networking, vol.1, pp. 176-186, 2009

[3] Ghani, N., S. Dixit, and T. S. Wang, ‘On IP-over-WDM integration’, IEEE
Communications Magazine., vol. 38, no. 3, pp. 72–84, Mar. 2000.

[4] Shub, Charles M. ‘On the roles of simulation, analytical modeling, and
measurement in solving complex problems’, Winter Simulation Conference Archive.
Proceedings of the 17th Conference on Simulation, 1985.

Вам также может понравиться