Power savings provided by elastic optical networks ...

Power savings provided by elastic optical networks considering yearly traffic fluctuations

8th CEF Networks Workshop

Prague, Czech Republic

15th of September 2014

Ioan Turus

2/17

Outline

• Introduction

– Energy efficiency in ICT

– Global traffic forecasts

– Green networking

– Predictable traffic fluctuations

• Proposed traffic model

• Energy reduction strategies

• Control plane implementation

• Implementation and results

• Conclusions

3/17

Energy efficiency in ICT

• The ICT industry accounts for approx. 2% of global CO2 emissions, a figure equivalent to aviation – Gartner 2007”

• “The share of electricity demand for ICT purposes is almost 11% of the overall final electricity consumption in Germany”

• “The ICT sector produces between 2% and 3% global greenhouse emissions annually”

• 3x traffic increase between 2013 and 2018

4/17

Global traffic forecasts

• 3x traffic increase between 2013 and 2018

[1] Cisco VNI, 2014

5/17

Green networking

[2] Rhee, ICNC, 2012

6/17

Predictable traffic fluctuations and growth

• NORDUnet – The overlay network of Nordic National Research and Education Networks

• NORDUnet traffic with Customers

– Day/night fluctuations

– Weekend drops

– Yearly growth

[3] http://stats.nordu.net

60%

max

avg

min 16%

54%

100%

7/17

• Predictable fluctuations

– Diurnal and weekly fluctuations

• Yearly traffic growth

– Traffic growth within one connection

Proposed traffic model

8/17

Energy reduction strategies (I)

• On/Off (Sleep mode) of OE devices:

– Transponders (TRX)

– Regenerators (REG) – back-to-back transponder configuration

• 100 G PDM-QPSK

ON IDLE OFF

Power(TRX) 350 W 8 W 0 W

Power(REG) 700 W 16 W 0 W

9/17

Energy reduction strategies (II)

• Data-rate adaptation

• Elastic transponder/regenerator

– 25, 50, 75, 100 Gbps datarate configuration

Payload (Gbps)

SR (GBd)

MF Reach (km)

Power (W)

100 28 PDM-QPSK 1200 350

75 28 21

PS-QPSK PDM-QPSK

1800 1200

350 255

50 28 14

PDM-BPSK PDM-QPSK

2500 1200

350 206

25 28 14 7

SP-BPSK PDM-BPSK PDM-QPSK

3000 2500 1200

350 206 189

TABLE I. Elastic transponder power consumption

10/17

Energy reduction strategies (III)

• Modulation Format (MF) adaptation

• Symbol Rate (SR) adaptation

• Mixed (SR+MF) adaptation

TRX REG TRX

100Gb/s

100G PDM-QPSK 100G PDM-QPSK

100Gb/s 50Gb/s 50Gb/s

50Gb/s PDM-BPSK

zzz…

50Gb/s

50G PDM-QPSK 14 GBd 50G PDM-QPSK 14 GBd

50Gb/s

TRX REG TRX

100Gb/s 100Gb/s

100G PDM-QPSK 28 GBd 100G PDM-QPSK 28 GBd

350W 700W 350W

350W 700W 350W 206W 412W 206W

Ch. Power: 1400W Ch. Power: 700W

Ch. Power: 1400W Ch. Power: 824W

11/17

Control plane implementation

• Automatic node configuration based on RSVP-TE signaling and a policy controller

• RSVP-TE used to:

– Set-up, tear-down Lambda LSPs according to the power state of OE devices

• Policy controller

– Decides on reconfiguration and/or recovery

– Provides the necessary information to the GMPLS control plane

12/17

• Reference topology: NORDUnet and GEANT topologies

• Three types of demands equally distributed:

– 50, 75 and 100 Gbps (peak capacity)

• MIT (Mean inter-arrival time) of 1.6h

• Holding time of 38h

– Total load of 24 Erlangs

• 80 wavelengths

Implementation

13/17

Scenario definition

MF SR

Scenario 1 (Fixed)

fixed (100G) fixed (100G)

Scenario 2 (MF)

adapt fixed

Scenario 3 (SR)

fixed adapt

Scenario 4 (Mixed)

adapt adapt

TABLE I. Scenario definition

14/17

Results – Power consumption NORDUnet

• MF lower power (REGs placed in mode OFF)

– Peaks given by diurnal and weekly fluctuations (…from day 150)

• SR even lower power (symbol-rate adaptation)

– Higher peaks given by diurnal and weekly fluctuations

• Mixed - lowest power consumption

15/17

Results – Power consumption GEANT

• MF lower than SR in this case

– Mainly due to long spans and higher need for regeneration

• Mixed - still the lowest power

16/17

Results – Power savings

34,4

42,7

48,9 45,7

42,4

50,9

0,0

10,0

20,0

30,0

40,0

50,0

60,0

MF SR Mixed

Averag

e p

ow

er s

avin

gs

norm

ali

zed

to

baseli

ne

[%

]

Energy reduction strategy

NORDUnet GEANT

17/17

Conclusions

• Traffic increase overprovisioning increased power consumption

• Periodical and predictable traffic variation in core networks

• Energy saving strategies based on:

– Sleep-mode of OE devices

– Data-rate adaptation (MF, SR and mixed)

• 50% energy savings for both networks in Mixed scenario

• MF outperforms SR in large footprint networks (e.g. GEANT)

• SR only is preferred in small networks due to less complex signaling

18/17

Ioan Turus [email protected]

[email protected]

+45 31627817

Find me on:

Thank you!

19/17

Acknowledgements

- Annalisa Morea and Dominique Verchere (Alcatel-Lucent Bell Labs) for guidance and valuable feedback during the external stay at Alcatel-Lucent Bell Labs France.

- Elastic Optical Networks Project (Celtic EO-Net) for valuable data regarding elasticity.

- GreenTouch consortium for valuable input with regards to energy efficiency strategies.