Daniel C. Kilper, Steven K. Korotky and Dusan Suvakovic. Bell Laboratories, Alcatel-Lucent. Murry Hill, NJ, U.S.A.. Estimates suggest that in ten years backbone ...
The 2010 Military Communications Conference - Unclassified Program - Systems Perspectives Track
Energy Efficient Networking: Avoiding a Future Energy Crunch David J. Bishop and Adrian R. Hartman LGS¹ Florham Park, NJ, U.S.A.
Daniel C. Kilper, Steven K. Korotky and Dusan Suvakovic Bell Laboratories, Alcatel-Lucent Murry Hill, NJ, U.S.A.
I.
Traffic (Tb/s)
Estimates suggest that in ten years backbone network traffic will increase by a factor of thirty, which will require a corresponding increase in network capacity. However, at the same time, as the traffic will continue to explosively increase, analyses also indicate that capacity scaling may be nearing an end for optical fibers, Silicon CMOS and cellular systems. Clearly this is undesirable and novel technical approaches are needed to address communications sustainability. In this paper we describe the problem and novel technical approaches to confront the issue.
INTRODUCTION
Historically communication network traffic has exhibited tremendous year-over-year growth, and this trend is expected to continue. Recent trends and future projections for several categories of traffic are shown in Fig. 1. Graphed are the total North America backbone network traffic as well as service specific traffic for peer-to-peer (P2P), internet video, wireless data and wireless voice. The latter, access oriented categories contribute to, but are not the only components of the total backbone traffic. Based on the projections for the individual categories, going forward, wireless voice and today's internet video are anticipated to grow more slowly and P2P will grow at a modest rate. The big driver for backbone traffic will be emerging video and interactive services. Wireless data driven by end user devices like the iPhone, Android, etc. is also projected to show strong growth throughout the next decade. By considering the historical evolution of its growth rate, we project that over the next ten years the total backbone network traffic will increase by approximately a factor of thirty. The worldwide traffic shows similar trends, and the total traffic is roughly twice the numbers in Fig. 1.
3
10
2
10
1
10
0
10
-1
10
-2
T ot
bo ack al B
ne
Internet Video
Da less Wire
P2P
ta
Wireless Voice
2010
2015
2020
Year Figure 1. Aggregate network backbone traffic for North America
In the past, while the traffic has grown exponentially, a number of the underlying technologies such as Silicon CMOS and WDM transmission in optical fibers have also seen an explosive growth in capacity - meaning that orders of magnitude increases in traffic did not translate into orders of magnitude increases in energy consumption. The increase in capacity has almost been matched by a comparable decrease in the required energy per bit to transport the information. However, sustaining this trend requires continuous innovation, which becomes more challenging as physical scales are reduced and fundamental limits are approached.
¹LGS is a wholly-owned subsidiary of Alcatel-Lucent dedicated to serving the U.S. Federal Government market
978-1-4244-8179-8/10/$26.00 ©2010 IEEE
10
2390
ENERGY ANALYSIS OF NETWORK
The results of a detailed analysis of the energy use for a typical network today are shown in Figure 2. One sees a number of basic trends. The first is that the network gets more efficient as you get farther into the core. This is because the multiplexing that takes place where many end users get their traffic added together and the energy cost per bit of moving these aggregated bundles is reduced. The second general trend is that the total energy used is related to bandwidth, e.g. larger bandwidth services use more energy in general than lower bandwidth services. Finally one should note that in general networks today are not designed and built solely around energy considerations. Things like installed cost, operating expense, reliability, performance, etc. are presently much more likely to be the drivers for how a given network is designed, built and operated. We expect additional considerations will be explicitly included in design analyses in the next few years because of the end of unbounded scaling of a number of the basic underlying technologies such as optical fiber capacity, Moore’s Law scaling in Si CMOS and the maturing of RF systems.
slowed significantly and commercial systems are following that trend, as indicated by the solid curve through the WDM commercial points. From 2010 on, total traffic is projected to grow faster than fiber capacity.
1000000 100000
G b /s
II.
10000
TDM Research
1000
WDM Research
100
TDM Commercial
10
WDM Commercial
1
Total Traffic
1980
1990
2000
2010
2020
Year Core Energy ~ 0.1 W/user
Metro/Edge Energy ~ 1 W/user
Access Energy ~ 10 W/user
Figure 3. Capacity trends in optical fibers and the total traffic trends with time
Cu
(ADSL, ADSL2, VDSL, …)
For CMOS, as shown in Fig. 4, a similar picture emerges, as Moore's Law scaling is slowing and the energy per bit reductions achieved with each decrease in line width are also slowing. At smaller feature sizes, the reductions in power per bit are no longer found [1]. In wireless systems, a similar trend is seen with systems approaching Shannon limits on capacity.
Fiber to the Node
Central Office
Remote
Dynamic Dissipation (norm.)
Fiber to the home
Figure 2. Shown is the energy per user for the core of the network, the central office metro/edge portion, and the access piece.
The energy per user is dominated by the access portion of the network with less energy per user being required as you get farther into the network. III.
TECHNOLOGY LIMITATIONS TO SCALING
Arguably, the increasing difficulty in the scaling of networks to higher traffic volume is already visible in the details of the growth of capacity of the underlying networking technologies. For example, as shown in Fig. 3, in an optical fiber the aggregate backbone ingress/egress traffic has for many years and until recently been less than that of the demonstrated capacity of a single fiber link. However, with backbone traffic predicted to grow by a factor of a thirty over the next decade, and the innovations in fiber transport beginning to slow, many optical fibers will be required to carry the traffic where in the past a small number would suffice. Clearly this would require a significant increase in energy. Note that the capacity increase in research demonstrations has
10
0
250 nm 180 nm
-1
10
45 nm 65 nm (2009)
130 nm 90 nm
15 nm 22 nm 32 nm (2018) (2015) (2012) -2
10
Moore's Law
-3
10
10
100
CMOS Feature Size (nm) Figure 4. Scaling of CMOS dynamic power dissipation as a function of feature size.
An analysis using a transaction based model [2] was recently carried out to understand the consequences for network energy consumption. As shown in Fig. 5a, based on
2391
current trends, going forward over the next decade the factor of thirty increase in capacity will require at least a one order of magnitude increase in energy consumption per user. Given that data networks today use roughly one percent of the electricity currently generated, this clearly presents a dilemma.
(a)
100
10
Power/User (W)
Power/User (W)
100
(b)
Fixed Access Wireless Access Metro/Edge Core
1
Fixed Access Wireless Access Metro/Edge Core
10
1
0.1
0.1
0.01
0.01
reduce the carbon footprint of future data networks. The goal of GreenTouch is to demonstrate key technologies that would enable a reduction in the total energy use per bit by a factor of one thousand. The consortium is using models for the network in 2020 as a basis for these technologies and anticipates that future networks will evolve in these directions driven by the network and the urgent need for efficient and sustainable growth [6]. Driven by similar motivations, the Mobile Virtual Center of Excellence (VCE) consortium initiated a green radio program that targets a factor of 100x reduction in energy requirements for high bit rate mobile services. Other organizations such as the Green Grid and the Global eSustainability Initiative (GeSI) provide tools and information to better understand and address key Information and Communications Technologies (ICT) energy related challenges [7]. V.
2010
2015
Year
2020
2010
2015
2020
Year
Figure 5. a) Energy consumption per user as a function of time assuming the growth in traffic shown in Figure 1 and business-as-usual technology trends. b) Optimistic benefit of changes in energy use per user that is possible using efficient networking technologies [3].
IV.
CONCLUSION
In conclusion, given the unrelenting demand for bandwidth and an expected end of scaling for fiber optical systems, Silicon CMOS, and RF systems, the current course and speed suggests that if not addressed IT networks would require an increase in power per user over the next decade that only becomes worse in time. More aggressive technology innovation including radical network changes focused on energy efficiency are needed. The GreenTouch consortium is an industry wide research effort to make these innovations a reality.
ENERGY EFFICIENT NETWORKING
Recent work in Alcatel-Lucent Bell Labs [3] has considered a variety of measures for increasing efficiency that are under focus today, such as sleep mode operation, energy proportional routing and computing, dynamical optical transparency, limited introduction of small cells, and new generation RF power amplifiers, and applied those expected improvements to the business-as-usual network energy model. An estimate of the maximum impact of these innovations is shown in Fig. 5b, which indicates that at best the network power can be kept roughly flat over the next decade. More pragmatically, practical scenarios can be expected to result in an increase in power. Looking beyond 2020, having exhausted these improvements, the power would be expected to continue to increase largely unabated with traffic growth. Thus, one must look to far more radical solutions to curb increases in energy consumption and to enable the Internet to continue to grow, which requires targeted, long-term research. Technologies that might be needed include new CMOS architectures, such as adiabatic computing; new network architectures and algorithms, such as the High Leverage Network; new amplifier designs; optical techniques, as described in [4]; ubiquitous picocells for wireless systems; and perhaps even new network protocols looking beyond IP [5].
[1] [2]
[3]
[4]
[5]
[6] [7]
Due to the pending explosion in anticipated energy needs for networks that we have described in this paper, something needs to be done. The scope of the problem has recently motivated industry wide response. Recently, a group of organizations from across industry and academia came together to form a unique research consortium, GreenTouch, which is inviting the entire community to work together to drastically
2392
REFERENCES M. Horowitz, et al., "Scaling, power, and the future of CMOS," Electron Devices Meeting, IEEE, 2005. J. Baliga, R. Ayre, K. Hinton, W. V. Sorin, and R. S. Tucker, "Energy consumption in optical IP networks," J. Lightwave Technol. 27, 2391-2403 (2009). D. C. Kilper, G. Atkinson, S. K. Korotky, S. Goyal, P. Vetter, D. Suvakovic, and O. Blume, “Power Trends in Communication Networks”, has been accepted for publication, IEEE J. Spec. Top. Quantum Electron. (2011). E. Desurvire, "Capacity demand and technology challenges for lightwave systems in the next two decades," J. Lightwave Technol. 24, 4697 (2006). S. Samuel, D. Kilper, S. Goyal, S. Venkatesan, D. Neilson, S. Korotky, D. Suvakovic, and G. Rittenhouse are preparing a manuscript entitled, "Sustainable Communication Networks” for publication. www.greentouch.org www.mobilevce.com; www.thegreengrid.org; www.gesi.org
