Evaluation of power consumption of Spanke Optical Packet Switch

The power consumption of an Optical Packet Switch equipped with SOA technology based Spanke switching fabric is evaluated. Sophisticated analytical models are introduced to evaluate the power consumption versus the offered traffic, the main switch parameters, and the used device characteristics. The impact of Amplifier Spontaneous Emission (ASE) noise generated by a transmission system on the power consumption is investigated. As a matter of example for 32 × 32 switches supporting 64 wavelengths and offered traffic equal to 0,8, the average energy consumption per bit is 5,07 · 10−2 nJ/bit and increases if ASE noise introduced by the transmission systems is increased.


I. INTRODUCTION
E NERGY efficiency can be considered as one of the biggest challenges in a large part of industrial and research fields. This arises from the need of reducing the energy related expenses of enterprises, industries as well as residential buildings, while keeping an eye on targets for the reduction of greenhouse gas emissions [1]. For example, as shown in [2], energy consumption of Telecom Italia network in 2006 has reached more than 2TWh (about 1% of the total Italian energy demand), increasing by 7.95% with respect to 2005, and by 12.08% to 2004. Another explanatory example is represented by British Telecom, which absorbed about 0.7% of the total UK's energy consumption in the winter of 2007, making it the biggest single power consumer in the nation [2]. Moreover, as evidenced in [3], about 10% of the UK's entire power consumption in 2007 was related to operating IT equipment.
Research has been initiated in recent years for energy saving of the Internet. This effort has been called "Greening the Internet". Several promising technologies have been being proposed from the component level to the network level [4].
Other solutions are based on power-aware system design that leads to reduce the power consumption of the electronic routers. The optical switch [5]- [6] has long been considered the primary candidate for replacing the electronic routers. For lacking of optical buffer, buffeless Optical Packet Switches are promising nodes in reducing the power consumption [8].
To solve output packet contentions they use the wavelength domain. Contending packets are wavelength converted by using Wavelength Converters (WC). Due to the high power consumption of WCs, especially for bit-rate increasing, Optical Packet Switch (OPS) architectures with shared WCs have V. Eramo [8]. In some cases they allows us to reduce by 80% the power consumption with respect to OPS in which no WCs sharing is performed [8].
In this paper we propose an analytical model to evaluate the average energy consumption per bit of an Optical Packet Switch equipped with a Spanke switching fabric realized in Semiconductor Optical Amplifier (SOA) technology. Sophisticated analytical models are introduced to evaluate the the power consumption of the devices, in particular SOAs, needed to realize the switching fabric. The introduced models allow us to evaluate the impact that Amplifier Spontaneous Emission (ASE) noise, generated by a transport system, has on the SOA's power consumption due to the SOA gain saturation. By means of the these models, we are able to evaluate the average energy consumption per bit of the Spanke switch as a function of the main system and traffic parameters and versus the characteristic of the transmission system (length, number of amplifiers, . . . ). The remainder of the paper is organized as follows. Section II describes the Spanke switch. An analytical model evaluating the average energy consumption per bit in Spanke switches versus the offered traffic, the switch parameters and the characteristics of the transmission system is described in Section III. The main numerical results are illustrated in Section IV where we provide some results on the power consumption of the Spanke switch. Finally Section V provides some final remarks and concludes the paper.

II. SPANKE OPTICAL PACKET SWITCH
The studied general switching architecture is reported in Fig. 1. It is equipped with N input/output fibers (IF/OF) where each IF/OF supports M wavelengths channels. Let λ i (i = 0, . . . , M − 1) be the wavelengths carried on each OF. In order to save power consumption, the OPS is equipped with fully shared Wavelength Converters (WC). Packets not requiring wavelength conversion are directly routed towards the Output Fibers (OF). On the contrary packets requiring wavelength conversion will be directed to a pool of WCs, wavelength converted and next routed to the OF to which they are directed.
An Optical Packet Switching architecture equipped with Spanke switching fabric realized in Semiconductor Optical Amplifier (SOA) technology is studied. An example of Spanke switch is illustrated in Fig. 2 in the case N =2, M =2 and r=1. A full pool of r Wavelength Converters is shared among the arriving packets. The selection of either an OF or a WC is realized by turning on one Semiconductor Optical Amplifier (SOA) of a 1×(N + r) Space Switching Module (SSM) of Control the 1st stage. Each SSM is composed by one splitter and N + r SOAs. The activated SOA allows the splitter loss to be overcome. One N M ×1 SSM of the 2nd stage, composed by one coupler and one SOA, has the function to forward to the WC stage a packet to be wavelength converted. The SOA gain allows the coupler loss to be overcome. The converted packets are sent to the OFs by turning on one SOA of a 1×N SSM of the 3rt stage. Finally each OF is equipped with an (N M + r)×1 SSM in the 4th stage whose function is to address towards the OF the packets coming either from the Input Wavelength Channels or from the pool of shared WCs.

III. ANALYTICAL EVALUATION OF THE POWER CONSUMPTION IN SPANKE SWITCH
To evaluate the Spanke switch energy consumption we use the model illustrate in [9] to evaluate the SOA's power consumption C SOA . According to this model, P SOA can be expressed as follows: where V b is the SOA forward bias voltage, I b is the polarization current, Γ SOA is the confinement factor, α SOA is the material loss, L SOA is the length and i t is the transparency current given by: w SOA being the SOA active region effective width, d SOA the active region depth, q = 1, 6×10 −9 C the electronic charge, N 0 the conduction band carrier density required for transparency, τ the carrier spontaneous decay lifetime. The amount of gain saturation G s is a function of the SOA input power P in SOA and the following nonlinear equation gives the unsaturated gain G us required to produce saturated gain G s [10]: In evaluating the various power consumption in the Spanke switch shown in Fig. 2 we notice that at time t: • there are as many turned on SOAs in 1st stage as the number N a (t) of packets forwarded; • the number of turned on SOAs in both 2nd stage and 3rd stage equals the number N c (t) of converted packets; • there are as many active turned on SOAs in 4th SSM stage as the number N d (t) of OFs in which at least one packet is directed; • we assume that all of the r Wavelength Converters are turned on. According to these remarks we can write the following expression for the average power consumption P Spanke av,T for the Spanke switch shown in Fig. 2: wherein: 4) is the power consumption of a turned on SOA in the ith stage (i=1, . . . , 4); • C W C is the power consumption of a Wavelength Converter; • C SOA of f is the power consumption of a turned off SOA; it is equal to V b i of f where i of f is the polarization current of an inactive SOA and needed to guarantee a high SOA switching rate [11] . . . , 4) we notice that an accepted packet may follow either of the paths reported in Fig. 3.a and Fig. 3.b according to the case in which it is or not wavelength converted. In the Figures we report both the values of loss introduced by the splitters and couplers and the SOA saturated gain. From Fig. 3.b and by using the expression of the power consumption of a cascade of SOA and passive elements reported in [13]- [14], the expressions Eqs (6)- (8) reported at page 3 can be obtained for C SOA i (i=1, 2, 3), wherein: • P in s is the input signal power; • P in ASE is the Amplifier Spontaneous Emission (ASE) noise power; • P se = n sp p ef f hν c B 0 ; each SOA is assumed to emit ASE with constant spectral density within the optical bandwidth B 0 ; ν c is the center frequency, h is the Planck constant, n sp is the excess spontaneous emission factor [15], p ef f is a factor which ranges from 1 for a device which amplifies only one polarization to 2 for a polarization-insensitive device. The evaluation of the power consumption C SOA 4 is evaluated in [14] and it is reported in Eq. (9) at page 4. Finally notice as by inserting Eqs (5)-(9) in Eq. (4) show how the ASE noise generated at the switch input may influence the power consumption of the Spanke switch due to SOA gain saturation. We perform the analysis under the following assumptions: • The SOA's power consumption model illustrated in [9] is adopted and allowing us, according to Eq. (1), to express the SOA power consumption as a function of the main SOA parameters (V b , i b , w SOA , . . . ); A 1 commercial SOAs [12] produced by manufacture A is used to implement the switching fabric. The A 1 parameter values are reported in Table I. • As Wavelength Converter, the Delayed Interference Signal Wavelength Converter (DISC) proposed in [16] is used. DISC utilizes an SOA and an Optical Bandpass Filter placed at the amplifier output. It can be constructed by using commercially available fiber-pigtailed components. It has a simple configuration and allows photonic integration. Its power consumption has been evaluated in [16] when commercial SOA produced by some manufactures are employed. In particular we consider the A 2 SOA characterized by a Multiple Quantum Well (MQW) type structure and produced by manufacture A. We report in  [16], is also reported. It equals 187mW when the WC is operating at bit-rate B=40 Gb/s. • We assume that the ASE noise P in ASE is generated by a Wavelength-Division Multiplexing (WDM) transmission system comprising S identical stages, each of length (i.e., EDFA amplifier spacing) L as illustrated in Fig. 4. The total length of the transmission system is L tot = SL. Each stage in Fig. 4 is modeled by a fiber attenuation the attenuation and the gain of the splitters, couplers and SOAs are reported  transmitter, S identical stages of optical gain. Each stage has length L and the used fiber is characterized by an attenuation equal to α dB/Km. Ltot is block with a power loss of D fiber = e αL , where α is the power attenuation per unit length of the fiber and an amplifier gain block with power gain G EDF A which is equal to the loss per stage (i.e., G EDF A = D fiber ). At each wavelength, the ASE noise P in ASE is given by the following expression [17]: where n EDF A sp is the excess spontaneous emission factor of each EDFA amplifier. We choose the switch parameters N =32, M =64 and the offered traffic p=0,8. The operation bit-rate is B=40 Gb/s and the optical bandwidth is B 0 =100Ghz. The used SOAs are characterized by the parameters n sp =3,5, p sp =2. Power consumption is not taken into account for the turned off SOAs (i of f =0). The energy consumptions are reported in Fig. 5 varying the number S of stages from 0 to 30 of the transmission system generating ASE noise at the switch input. Each stage has length L=60Km with attenuation α = 0, 2 dB/Km and each EDFA is characterized by n EDF A sp =1. The case S=0 corresponds to the case in which no ASE noise is generated because electrical regeneration is performed before the switching operation. We also report in Fig. 6 the SOA power consumption and versus the number r of WCs for S=0 and S=20. From Fig. 5 we can notice how the increase in ASE noise makes less energy efficient the Spanke switch. For instance in the case of Spanke switch and when r=320 WCs are used, E Spanke av,T increases from 3, 85 · 10 −2 nJ/bit to 5, 02 · 10 −1 nJ/bit when S increases from 0 to 30. That is consequence of the increase in ASE noise that saturates the SOAs gain leading to the need to increase the power consumption as indicated by the Eqs (5)- (9).
The increase in power consumption due to the ASE noise is confirmed in Fig. 7 where we report the average energy  We have proposed a sophisticated analytical model in order to evaluate the average energy consumption per bit of the Spanke switch. In the evaluation of the energy consumption we take into account the ASE noise generated by the transmission system that can degrade the performance in power consumption because of the gain saturation of the SOA gates needed to realize the switching fabric. We have verified the the ASE noise generated by a transmission system may strongly degrade the switch performance in terms of power consumption. As a matter of example, if a switch with N =32 and M =64 is taken into account and the offered traffic p equals 0,8, the average energy consumption per bit is 1, 53 · 10 −1 nJ/bit in the case of a transmission system of total length 2100Km and S=30 spans.