Determinism, Randomness, and Harvesting: Strategies for Reduced-Energy Telecommunication Networks

Prof. Stuart D. Walker & Dr. Michael C. Parker

Access Networking Laboratory School of Computer Science & Electronic Engineering

University of Essex Colchester, UK

stuwal@essex.ac.uk

ETSI TC EE Workshop, Genoa, Italy, 21st June 2012

– Moore’s Law (processing power)

– Nielsen’s Law (internet bandwidth)

– Metcalf’s Law (network value, n2 or nlogn)

– Edholm’s Law (converging network technologies)

– Parker-Walker’s Law (conservation of network energy)

The Essex University Dept. of Laws

How to measure network energy consumption? J/b, J/b/s, W/b, W/b/s...J/b/s/Hz, W/b/s/Hz (spectral efficiency) What’s the minimum J/b for information transmission/processing? Is there an absolute minimum to network energy consumption?

Power Effective bit-rate Energy/bit [J/b] dBε

BT Network 1 GW 22 Tb/s 45x10-6 162.0

Dell laptop 80 W 1.87 GHz (clock) 42.8x10-9 131.7

Ultra-low power DSL/fibre 165 mW 10 Mb/s 16.5x10-9 127.6

Tb/s Router 10 kW 1 Tb/s 10x10-9 125.4

200 km Raman-pumped link 60 W 10 Gb/s 6x10-9 123.2

Currently most-efficient CPU 2.8 W 1 Gflops 2.8x10-9 119.9

Currently most-efficient 10 Gb/s system 10 W 10 Gb/s 1x10-6 115.4

Human Brain (T=310 K, or 37°C) 20 W 40 Tb/s 0.5x10-12 82.3

20 photons/bit, 10-9 BER, λ=1.55µm 25.6 nW 10 Gb/s 2.56x10-18 29.5

1 photon/bit , λ=1.55µm (quantum limit) 1.28 nW 10 Gb/s 0.128x10-18 16.5

Absolute Energy Efficiency Figure

=

ln2RateBit Powerlog10dBε 10 kT

Boltzmann’s constant k=1.381x10-23 J/K Absolute temperature T=300 K

‘Roadmapping ICT: An Absolute Energy Efficiency Metric ’, M.C. Parker, S.D. Walker, JOCN, 3(8), pA49, 2011

Absolute Energy Efficiency Metric

=

ln2RateBit Powerlog10dBε 10 kT

Boltzmann’s constant k=1.381x10-23 J/K Absolute temperature T=300 K

Access Networks

Stochastic energy-efficiency approaches to power consumption minimisation (complementary and/or in addition to sleep/idle modes) using master-slave configurations, applied to equipment featuring high overhead power consumptions, i.e. that are inelastic to traffic volumes, e.g.: - Routers, Switches, Rack equipment, Optical Amplifiers

Master-Slave Equipment Configurations

“Stochastic Energy-Efficiency Optimization in Photonic Networking by use of Master-Slave Equipment Configurations”, M.C. Parker, S.D. Walker, ONDM’12, Colchester, UK, April 2012

Master-Slave Equipment Upgrade Paths

• Master-Slave devices don’t have to be switched-on or –off at high speeds: • Only “simple” input switch must be fast • Master & Slave devices can have slow switch-on & -off times – not operated in “hard” manner • An overlapping operation time ∆T is possible:

• Hysteresis mode of operation (avoids HF “oscillations”) • Master & Slave devices pre-emptively powered-up in anticipation of threshold crossing

• Inherent capacity redundancy in a Master-Slave configuration: • Greater reliability & resilience in operation • Leading to a greater longevity & a higher mean-time before failure (MTBF)

• Optimisation of alternative upgrade paths for minimisation of OpEx/CapEx/TCO

Green Telecommunications

• Motivations: – Environmental (CO2 footprint) – Economic ($) – Practical (battery life of mobile devices)

• Lots of Approaches: – Low-power devices

• Un-cooled, high-temperature devices - (SiC, GaN) materials

• Minimise energy dissipation • ...

– Power saving strategies • Powering down of un-utilised sections of chips & equipment • Energy harvesting (e.g. convective currents) • Green routing protocols • ...

Energy-efficiency possibilities for converged wireless and optical access networks

• Energy efficiency • Including green radio (adaptive antennas, MIMO beam shaping) • High differential energy efficiency base stations (BSs) • Handover to alternative low-power BSs within a PON infrastructure; • Selective powering-down/idle states of network cells • Arbitrage between different marginal costs ($/CO2) of optical & wireless bandwidths

• Caching & Locality • Content distribution networking • Resource/bandwidth allocation algorithms

• Network Simplification & Topologies • Holistic optimisation of a fully integrated wireless-wireline NGOA architecture •NGOA node consolidation v. Wireless network cognition, intelligence near end-users

• Cost sharing of equipment (techno-economics) • Relative cost of Wireline final-drop (FTTH) v. Wireless final-drop • Fibre infrastructure serving end-users as well as BS ONUs, with mobile backhaul a fully integrated aspect to the complete optical network

Renewable Power Strategies in Access

Optimised design of active nodes (CAN/CO/RN/BS) to minimise operating carbon footprint using: • Renewable energy sources, with optimised power security/reliability:

• Solar/Wind – 10s & 10s of KWs power supply – ideal matching • DC-designed & optimised active nodes:

• DC-centric Central Offices • Photo-voltaic power activated to traffic loads via Golomb Ruler:

• (sub-)optimisation designs

10 Independent Ways to Cut Energy Consumption in Half

• Mathematical Trick: 210=1024 ~ 1000=30dB

• 4 Main Areas to Concentrate Upon, each with independent approaches: – Improved inherent energy-efficiencies as offered by electronics

technologies • (1) More efficient CMOS technologies, (2) High temperature operation of ICs

– More sophisticated management and exploitation of network resources • (3) Source coding & caching, (4) Multi-layer traffic engineering (MLTE), • (5) Powering down, sleep/idle modes and burst-mode operation.

– The inherent energy efficiencies as offered by optics technology solutions such as

• (6) Optical bypass, (7) Coherent detection, (8) Polarisation multiplexing. – More environmentally sustainable approaches to network design such as

• (9) Micro-power generation, (10) Increased reliability and robustness of network equipment.

10

(1) More efficient CMOS, P∝C2/3 • CMOS power dissipation P increases as the two-thirds power of clock rate C,

i.e. P∝C2/3. • 50% (i.e. 3dB) improvement in energy-efficiency for each independent network

design principle is achieved if CMOS clock-rates increase only by a factor of 8.

Michael C. Parker, Stuart D. Walker 11

(2) High Temperature Chips • High-temperature semiconductors such as SiN, SiC or GaN platforms can

reduce or remove the need for forced cooling. • Current Power Usage Effectiveness (PUE) is about 2.0 • Avoiding cooling overhead, and reaching PUE≈1.0 offers a 3dB improvement

(3) Source Coding & Caching As a routine part of any end-to-end transmission link, source coding for data compression is still somewhat underutilised

HDTV over IP, and even 3DTV etc., have significant redundancy making compression relatively straightforward. Compression factors can reach 40 times, so 16dB improvement in dBe network energy efficiency may be possible.

Content distribution/caching

(4) Multilayer Traffic Engineering (MLTE)

• MLTE exploits traffic statistics to re-route traffic away from under-utilised nodes

• If the traffic rate reduces below a pre-defined low-level threshold (LLT) , traffic is groomed onto relatively more popular router nodes – IP routers more efficient (c.f. CMOS) when running at higher capacities

• The routers in relatively under-utilised nodes switched into an idle state. – 3dB (45%) improvement in energy efficiency is plausible

(5) Powering-down, sleep-mode, burst-mode operation • Disadvantage of circuit switching is its inefficient use of network

resources, – E.g. a data pipe is setup but kept in operation with no traffic.

• Statistical signatures of most traffic sources don’t lend themselves to efficient aggregation. Fortunately, central limit tendencies allow the data power envelope to be “surfed”. – Preliminary experiments show that 3.4 dB energy saving can be

expected, – Interdependence with MLTE limits overall combination of (4) and

(5) to 6 dB.

13

(6) Optical bypass – avoid OEO, OE or EO conversions • Current IP routers still require full OEO conversion

– Future predictions project them to consume 90% of overall network power. • If at least half the number of nodes can be optically bypassed by a packet traversing the

network, this represents at least a 50% saving in IP router energy – a further 3 dB reduction in energy consumption.

(7) Coherent Detection

• Coherent detection technology offers an intrinsic 3dB improvement in receiver sensitivity;

• This represents a potential 50% improvement in energy efficiency for same system performance.

(8) Polarisation Multiplexing (Pol-Mux)

• Principle of pol-mux often now appears with coherent detection – pol-mux is an independent technique to improve energy-efficiency.

• Maximum pol-mux 50% energy saving only achieved in the limit of all signals remaining in the optical domain across the network.

– A degree of interdependence with (6) – optical bypass – Energy efficiency advantages of pol-mux increases with the degree of optical bypass

employed. • Interdependence with optical bypass limits combination (6) and (8) to <6 dB.

(9) Local energy production (micro-power generation) • Micro-generated and local renewable power sources deployed at network

nodes reduce overall network carbon footprint, and avoid power-line transmission losses (estimated at 8%).

• Overlap period where conventional, fossil-fuel based provision for node powering required; – IPv6 allows “green” bits to minimise fossil-fuel consumption. – energy efficiencies improvements of more than 50% are possible.

14

(10) Increased reliability & robustness – avoid dualling

• A large (50%) degree of redundancy is frequently required in network architectures to satisfy the demands for a certain level of quality of QoS.

• 50% saving in overall network energy consumption/CO2-footprint is possible – Avoid network dualling – reduced no. of truck-rolls – Enhanced network intelligence: Zero-touch & soft-photonics

Selected Publications:

[1] “An Absolute Network Energy Efficiency Metric”, M.C. Parker, S.D. Walker, 1st International ICST Workshop on Green Grids (Green Grids 2009), Athens, Greece, September 2009 [2] “Roadmapping ICT: An Absolute Energy Efficiency Metric”, Journal of Optical Communications and Networks, vol.3(8), pA49-A58, 2011 [3] “Stochastic Energy-Efficiency Optimization in Photonic Networking by use of Master-Slave Equipment Configurations”, M.C. Parker, S.D. Walker, ONDM’12, Colchester, UK, April 2012 [4] “Energy-Efficiency Optimised Upgrade Paths for Cascaded, Stochastically-Based, Master-Slave IP Router Configurations”, M.C. Parker, S.D. Walker, ECOC’12, Paper P5.08, Amsterdam, The Netherlands, September 2012