Photonic Switching And The Energy Bottleneck
This document was uploaded by user and they confirmed that they have the permission to share it. If you are author or own the copyright of this book, please report to us by using this DMCA report form. Report DMCA
Overview
Download & View Photonic Switching And The Energy Bottleneck as PDF for free.
More details
- Words: 1,662
- Pages: 2
WA3.5 3:15 PM – 3:30 PM
Photonic Switching and the Energy Bottleneck J. Baliga, R. Ayre, K. Hinton and R. S. Tucker ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN) Department of Electrical and Electronic Engineering, University of Melbourne, Victoria 3010, Australia Abstract: The energy consumption of the Internet is growing exponentially. We examine the potential of photonic switching to reduce energy consumption by determining the contribution of cross connects and buffers to the total energy consumption of the Internet.
network core, each of which is now described in more detail. When we give power consumption figures, these are based on the typical power consumption and when such values are not available, the heat dissipation. The calculations do not include the extra power required to maintain redundant routers for reliability. To calculate cooling requirements, we assume that for every watt of power consumed, another watt of power is required for cooling [9].
Keywords: Packet Switching, Energy Introduction As Internet traffic increases, the quantity, capacity, and energy consumption of transmission and switching equipment required to route this traffic must also grow. Energy consumption is becoming a key environmental, social and political issue [1][2]. In addition, the engineering challenges of constructing and maintaining large data centers and switching centers is becoming a matter of increasing concern [3]. This raises the issue of whether Internet growth may ultimately be constrained by energy consumption rather than by bandwidth [3][4]. A question of particular relevance to the photonics community is whether photonic switching and all-optical networking technologies create opportunities to avert the looming “energy bottleneck” in the Internet. At the level of optical circuit switching, MEMS-based optical cross connects [5] show great potential for reducing energy consumption. However, it is not yet clear whether this will be true for high-speed photonic switching technologies, for example burst switching [6] and packet switching [7]. The objective of this paper is to obtain an estimate of how much energy could be saved if electronic switching and buffering components could be replaced by lower-power photonic components. We begin by describing a networkbased model [8] of the power consumption of a standard Internet Service Provider’s (ISP) network, formulated with data from major equipment vendors. Using this model, we estimate the power consumption of the Internet as a function of the access bit rate provided to users. We then calculate a breakdown of the total power consumed by different functional blocks in the Internet, including electronic switching fabrics in routers, electronic buffers in routers, and the WDM fibre transmission systems between network nodes. We show that, for an access rate of 100 Mb/s, the functional blocks in the network that could possibly benefit from conversion to photonic technologies the switch fabrics and buffers - consume approximately 7% and 2%, respectively, of the total energy consumption of the Internet. Our conclusion is that photonic technologies alone will not be able to solve the looming “energy bottleneck” problem.
2.1 Access Network To minimize both operational and installation costs and to maximize bandwidth, future access networks will have a Passive Optical Network (PON) [10] structure. An Optical Line Terminal (OLT), which resides in an edge node, connects to several groups of Optical Network Units (ONUs). Each ONU serves a home and each group of ONUs share a single connection to an OLT using time multiplexing. The average access rate is set by backhaul capacity, ONU peak access rates, and the number of ONUs that share a connection to an OLT; the last is currently limited by most manufacturers at 32 [11]. The ONU consumes approximately 10 W, while each OLT consumes 100 W [11]. 2.2 Metro Network On the network side, the OLTs are typically connected by four Gigabit Ethernet lines to one of the edge routers which encapsulate the IP packets into a SONET/SDH format for transmission to the network core [12]. In our network model, we use the Cisco 12816 edge router [12]. This device is typical of routers available from other manufacturers with respect to performance and power. The edge router connects to the core node via fourteen 10 Gb/s Packet-over-Sonet (POS) links, and to forty OLTs each via 4x1 Gb/s links. The edge routers do not need to be interconnected as they only perform forwarding. Each Cisco 12816 router consumes approximately 4.21 kW [12]. We assume that for single-chassis routers like the 12816, the breakdown of power consumption among system components is as given in the left column of Table 1 [13]. 2.3 Core Network Network routing is performed at the central node by several multishelf core routers [12], such as the Cisco CRS-1 Multishelf System with a switching capacity of 92 Tb/s fullduplex [12]. The CRS-1 routers are interconnected through 40 Gb/s POS links, connect to the edge routers through 10 Gb/s POS links and have 40 Gb/s POS links to neighbouring long distance locations. Each complete CRS1 system consumes 1020 kW [12]. To estimate the breakdown of power consumption in the CRS-1, we assume that the energy consumption of the multi-chassis switching fabric (including inter-chassis interconnects) is
2. Network Structure An ISP’s network can be logically split into three main sections - the access network, the metro network and the 1-4244-1122-X/07/$25.00©2007 IEEE
125
Authorized licensed use limited to: IEEE Xplore. Downloaded on October 10, 2008 at 13:23 from IEEE Xplore. Restrictions apply.
Table 1 : Breakdown of power consumed by a router.
Supply loss and blowers Forwarding engine Switching Fabric Control plane I/O Buffers Total
Table 2 : Breakdown of power consumption by switch fabrics, buffers and WDM links as percentage of total internet power.
Percentage of Total Power Single chassis [13] Multi-chassis 35 33 33.5 32 10 14.5 11 10.5 7 6.5 3.5 3.5 100 100
Switching Fabric Buffers WDM
Average Access Rate (Mb/s) 10 100 1000 1.8 % 6.7 % 9.1 % 0.5 % 1.8 % 2.4 % 1.2 % 4.3 % 5.8 %
increased by 50%. The resulting breakdown of power consumption is given in the right-hand column of Table 1. We assume that 50% of the traffic into a core router is internal, 25% is add/drop and finally 25% is bypass. This means that the total traffic being processed by the router is equal to 133% of the total metro traffic. Core routers are built such that they are able to cope with future growth and so are usually built such that they can handle double the current peak demand [14]. 2.4 WDM Links The WDM terminal systems connecting the edge nodes to the core node consume 1.5 kW for every 64 wavelengths [15]. At most one multiple wavelength amplifier, which consumes 6 W per fiber [15], is required between each pair of terminal systems provided the size of the city is less than 200 km in diameter. The WDM terminal systems connecting the core nodes consumes 811 W for every 176 channels, while each intermediate line amplifier (ILA) consumes 622 W for every 176 channels [16]. For a distance of 1,500 km between core nodes, two terminal systems and 14 ILA systems are required.
Figure 1 : Power consumption per million users of various components of the network.
lower power solutions than electronics, photonic technologies alone will not solve the looming “energy bottleneck” problem. 4. References [1] [2]
2.4 Breakdown of Power Consumption
[3]
Using the energy model described above, we have calculated the power consumption of each component of the network per million homes. This power consumption is plotted against the average access rate in Figure 1. We have extrapolated the calculations (dotted curves) for access rates above 100 Mb/s assuming no future improvements in router efficiency. However, in line with evolution of integrated chip processors, there is some scope for reduction in power consumption for higher capacity routers. Table 2 gives a summary of the power consumed by switch fabrics, buffers and WDM links as a percentage of total internet power. Power consumption is dominated by the forwarding engines and by the access terminals, particularly at low access rates.
[4]
[5] [6]
[7] [8] [9]
3. Conclusion
[10]
We have estimated the power consumption of the functional blocks in the network. We have shown that the power consumption of the switch fabrics as a percentage of the total power consumption rises slowly with increasing access rate. We also showed that, at an access rate of 100 Mb/s, the switch fabrics and buffers consume approximately 7% and 2%, respectively, of the total energy consumption of the Internet. Our conclusion is that even if photonic switching and buffering technologies provide
[11] [12] [13] [14]
[15] [16]
Telegeography Research & Primetrica Inc. 2005. [Online]. http://www.telegeography.com M. Gupta and S. Singh, “Greening of the Internet,” ACM SIGCOMM, Karlsruhe, Germany, Aug. 2003. A. Vukovic, “Data Centers: Network Power Density Challenges,” ASHREA Journal, vol. 47, p. 55, Apr. 2005 K. Christensen, “The next frontier for communications networks: power management,” Proc. SPIE – Perf. Control of Next-Gen. Comm. Nets., vol. 5244, pp. 1-4, Sep. 2003. T. W. Yeow et al., “MEMS optical switching,” IEEE Comm. Mag., vol. 39, p. 158-163, Nov. 2001. Y. Chen et al, “Optical burst switching: a new area in optical networking research,” IEEE Networks, vol 18, issue 3, pp 16-23, May-June 2004. M. C. Chia et al, “Optical packet switches: a comparison of designs,” ICON, pp 365-369, Sep. 2000. J. Baliga, K. Hinton and R. Tucker, “Energy Consumption of the Internet,” COIN-ACOFT, June, 2007. K. Dunlap, “Cooling Audit for Identifying Potential Cooling Problems in Data Centers,” 2006. http://www.apcmedia.com /salestools/VAVR-5UGVCN_R2_EN.pdf P. Chanclou et al., “Overview of Optical Broadband Access Evolution,” IEEE Comm. Mag., p. 29, Aug. 2006. NEC GE-PON Data Sheets. [Online]. http://www.nec.co.jp Cisco Data Sheets. [Online]. http://www.cisco.com G. Epps, Cisco, Private communication. E. Desurvire, “Capacity Demand and Technology Challenges for Lightwave Systems in the Next Two Decades,” J. Lightwave Technol., vol. 24, p.4697, 2006. Lucent Technologies Data Sheets. [Online]. http://www.lucent.com/eon Fujitsu Data Sheets. [Online]. http://www.fujitsu.com
126
Authorized licensed use limited to: IEEE Xplore. Downloaded on October 10, 2008 at 13:23 from IEEE Xplore. Restrictions apply.
Photonic Switching and the Energy Bottleneck J. Baliga, R. Ayre, K. Hinton and R. S. Tucker ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN) Department of Electrical and Electronic Engineering, University of Melbourne, Victoria 3010, Australia Abstract: The energy consumption of the Internet is growing exponentially. We examine the potential of photonic switching to reduce energy consumption by determining the contribution of cross connects and buffers to the total energy consumption of the Internet.
network core, each of which is now described in more detail. When we give power consumption figures, these are based on the typical power consumption and when such values are not available, the heat dissipation. The calculations do not include the extra power required to maintain redundant routers for reliability. To calculate cooling requirements, we assume that for every watt of power consumed, another watt of power is required for cooling [9].
Keywords: Packet Switching, Energy Introduction As Internet traffic increases, the quantity, capacity, and energy consumption of transmission and switching equipment required to route this traffic must also grow. Energy consumption is becoming a key environmental, social and political issue [1][2]. In addition, the engineering challenges of constructing and maintaining large data centers and switching centers is becoming a matter of increasing concern [3]. This raises the issue of whether Internet growth may ultimately be constrained by energy consumption rather than by bandwidth [3][4]. A question of particular relevance to the photonics community is whether photonic switching and all-optical networking technologies create opportunities to avert the looming “energy bottleneck” in the Internet. At the level of optical circuit switching, MEMS-based optical cross connects [5] show great potential for reducing energy consumption. However, it is not yet clear whether this will be true for high-speed photonic switching technologies, for example burst switching [6] and packet switching [7]. The objective of this paper is to obtain an estimate of how much energy could be saved if electronic switching and buffering components could be replaced by lower-power photonic components. We begin by describing a networkbased model [8] of the power consumption of a standard Internet Service Provider’s (ISP) network, formulated with data from major equipment vendors. Using this model, we estimate the power consumption of the Internet as a function of the access bit rate provided to users. We then calculate a breakdown of the total power consumed by different functional blocks in the Internet, including electronic switching fabrics in routers, electronic buffers in routers, and the WDM fibre transmission systems between network nodes. We show that, for an access rate of 100 Mb/s, the functional blocks in the network that could possibly benefit from conversion to photonic technologies the switch fabrics and buffers - consume approximately 7% and 2%, respectively, of the total energy consumption of the Internet. Our conclusion is that photonic technologies alone will not be able to solve the looming “energy bottleneck” problem.
2.1 Access Network To minimize both operational and installation costs and to maximize bandwidth, future access networks will have a Passive Optical Network (PON) [10] structure. An Optical Line Terminal (OLT), which resides in an edge node, connects to several groups of Optical Network Units (ONUs). Each ONU serves a home and each group of ONUs share a single connection to an OLT using time multiplexing. The average access rate is set by backhaul capacity, ONU peak access rates, and the number of ONUs that share a connection to an OLT; the last is currently limited by most manufacturers at 32 [11]. The ONU consumes approximately 10 W, while each OLT consumes 100 W [11]. 2.2 Metro Network On the network side, the OLTs are typically connected by four Gigabit Ethernet lines to one of the edge routers which encapsulate the IP packets into a SONET/SDH format for transmission to the network core [12]. In our network model, we use the Cisco 12816 edge router [12]. This device is typical of routers available from other manufacturers with respect to performance and power. The edge router connects to the core node via fourteen 10 Gb/s Packet-over-Sonet (POS) links, and to forty OLTs each via 4x1 Gb/s links. The edge routers do not need to be interconnected as they only perform forwarding. Each Cisco 12816 router consumes approximately 4.21 kW [12]. We assume that for single-chassis routers like the 12816, the breakdown of power consumption among system components is as given in the left column of Table 1 [13]. 2.3 Core Network Network routing is performed at the central node by several multishelf core routers [12], such as the Cisco CRS-1 Multishelf System with a switching capacity of 92 Tb/s fullduplex [12]. The CRS-1 routers are interconnected through 40 Gb/s POS links, connect to the edge routers through 10 Gb/s POS links and have 40 Gb/s POS links to neighbouring long distance locations. Each complete CRS1 system consumes 1020 kW [12]. To estimate the breakdown of power consumption in the CRS-1, we assume that the energy consumption of the multi-chassis switching fabric (including inter-chassis interconnects) is
2. Network Structure An ISP’s network can be logically split into three main sections - the access network, the metro network and the 1-4244-1122-X/07/$25.00©2007 IEEE
125
Authorized licensed use limited to: IEEE Xplore. Downloaded on October 10, 2008 at 13:23 from IEEE Xplore. Restrictions apply.
Table 1 : Breakdown of power consumed by a router.
Supply loss and blowers Forwarding engine Switching Fabric Control plane I/O Buffers Total
Table 2 : Breakdown of power consumption by switch fabrics, buffers and WDM links as percentage of total internet power.
Percentage of Total Power Single chassis [13] Multi-chassis 35 33 33.5 32 10 14.5 11 10.5 7 6.5 3.5 3.5 100 100
Switching Fabric Buffers WDM
Average Access Rate (Mb/s) 10 100 1000 1.8 % 6.7 % 9.1 % 0.5 % 1.8 % 2.4 % 1.2 % 4.3 % 5.8 %
increased by 50%. The resulting breakdown of power consumption is given in the right-hand column of Table 1. We assume that 50% of the traffic into a core router is internal, 25% is add/drop and finally 25% is bypass. This means that the total traffic being processed by the router is equal to 133% of the total metro traffic. Core routers are built such that they are able to cope with future growth and so are usually built such that they can handle double the current peak demand [14]. 2.4 WDM Links The WDM terminal systems connecting the edge nodes to the core node consume 1.5 kW for every 64 wavelengths [15]. At most one multiple wavelength amplifier, which consumes 6 W per fiber [15], is required between each pair of terminal systems provided the size of the city is less than 200 km in diameter. The WDM terminal systems connecting the core nodes consumes 811 W for every 176 channels, while each intermediate line amplifier (ILA) consumes 622 W for every 176 channels [16]. For a distance of 1,500 km between core nodes, two terminal systems and 14 ILA systems are required.
Figure 1 : Power consumption per million users of various components of the network.
lower power solutions than electronics, photonic technologies alone will not solve the looming “energy bottleneck” problem. 4. References [1] [2]
2.4 Breakdown of Power Consumption
[3]
Using the energy model described above, we have calculated the power consumption of each component of the network per million homes. This power consumption is plotted against the average access rate in Figure 1. We have extrapolated the calculations (dotted curves) for access rates above 100 Mb/s assuming no future improvements in router efficiency. However, in line with evolution of integrated chip processors, there is some scope for reduction in power consumption for higher capacity routers. Table 2 gives a summary of the power consumed by switch fabrics, buffers and WDM links as a percentage of total internet power. Power consumption is dominated by the forwarding engines and by the access terminals, particularly at low access rates.
[4]
[5] [6]
[7] [8] [9]
3. Conclusion
[10]
We have estimated the power consumption of the functional blocks in the network. We have shown that the power consumption of the switch fabrics as a percentage of the total power consumption rises slowly with increasing access rate. We also showed that, at an access rate of 100 Mb/s, the switch fabrics and buffers consume approximately 7% and 2%, respectively, of the total energy consumption of the Internet. Our conclusion is that even if photonic switching and buffering technologies provide
[11] [12] [13] [14]
[15] [16]
Telegeography Research & Primetrica Inc. 2005. [Online]. http://www.telegeography.com M. Gupta and S. Singh, “Greening of the Internet,” ACM SIGCOMM, Karlsruhe, Germany, Aug. 2003. A. Vukovic, “Data Centers: Network Power Density Challenges,” ASHREA Journal, vol. 47, p. 55, Apr. 2005 K. Christensen, “The next frontier for communications networks: power management,” Proc. SPIE – Perf. Control of Next-Gen. Comm. Nets., vol. 5244, pp. 1-4, Sep. 2003. T. W. Yeow et al., “MEMS optical switching,” IEEE Comm. Mag., vol. 39, p. 158-163, Nov. 2001. Y. Chen et al, “Optical burst switching: a new area in optical networking research,” IEEE Networks, vol 18, issue 3, pp 16-23, May-June 2004. M. C. Chia et al, “Optical packet switches: a comparison of designs,” ICON, pp 365-369, Sep. 2000. J. Baliga, K. Hinton and R. Tucker, “Energy Consumption of the Internet,” COIN-ACOFT, June, 2007. K. Dunlap, “Cooling Audit for Identifying Potential Cooling Problems in Data Centers,” 2006. http://www.apcmedia.com /salestools/VAVR-5UGVCN_R2_EN.pdf P. Chanclou et al., “Overview of Optical Broadband Access Evolution,” IEEE Comm. Mag., p. 29, Aug. 2006. NEC GE-PON Data Sheets. [Online]. http://www.nec.co.jp Cisco Data Sheets. [Online]. http://www.cisco.com G. Epps, Cisco, Private communication. E. Desurvire, “Capacity Demand and Technology Challenges for Lightwave Systems in the Next Two Decades,” J. Lightwave Technol., vol. 24, p.4697, 2006. Lucent Technologies Data Sheets. [Online]. http://www.lucent.com/eon Fujitsu Data Sheets. [Online]. http://www.fujitsu.com
126
Authorized licensed use limited to: IEEE Xplore. Downloaded on October 10, 2008 at 13:23 from IEEE Xplore. Restrictions apply.
Related Documents
Photonic Switching And The Energy Bottleneck
November 2019 1
The Photonic Bottleneck
November 2019 4
Switching
November 2019 21
Switching
December 2019 24
Energy And The Internet
November 2019 30