1. Introduction

Data center traffic is exponentially increasing due to the explosion of cloud and web applications (e.g., financial transactions, social networks, video streaming and research engine). Effort is needed in providing low power consumption, cost reduction and programmability [1–4]. Vertical cavity surface emitting laser (VCSEL) technology is emerging as an interesting solution particularly for short reach scenarios where massive low-cost production of energy efficient modules is a must. The main benefits of making use of VCSELs with respect to standard edge-emitting lasers, are related to their vertical structure. Indeed, that aspect enables early testing throughout the fabrication process, so as to check for material quality and processing issues [5]. This procedure allows fast testing and limits both waste of time and materials, therefore opening to a potential low cost. Furthermore, their remarkable electro-optical energy conversion efficiency, the possibility to be driven at low driving currents and to operate at high temperatures for relaxing the operational cost related to thermal stabilization [6] and their potentials for hybrid integration into Silicon (Si)-based transceivers [6–11], encourage their employment in integrated modules for optical interconnects and short-reach scenarios up to the access segment [12]. Indeed direct modulation of VCSELs does not enable long fiber reach because of the limited signal performance in terms of emission power, extinction ratio and introduced frequency chirp. However such lasers can be exploited for short distances with acceptable performance by exploiting low-cost direct detection (DD) at the receiver. In particular, multi-mode short-wavelength VCSELs achieve quite impressive data rates up to 71 Gbps [13] even though with a maximum fiber reach limited to a few tens of meters. In intensity modulation (IM)-DD systems the data rate is limited by the modulation bandwidth that, for long-wavelength (LW)-VCSELs, typically reaches up to 10 GHz for commercial devices [14] and up to 18 GHz for research prototypes [15]. Presently, the considered optical wavelength range for such type of VCSELs includes both the second (1.3µm) and the third window (1.55µm, the so called C-band). In the first case the main benefit consists of the low chromatic dispersion impact with potential for access and metro networks distances up to 20km of standard single-mode fiber (SSMF) [12], whereas the C-band exhibits both lower attenuation and lower distortions due to nonlinear effects, which is also a crucial aspect when dealing with WDM signals. In this latter case, due to the higher chromatic dispersion impact, a few kilometers of SSMF are achievable, thus being suitable for intra-data center networks [6] and, more in general, short reach scenarios.

As stated before, the programmability is a topic investigated within data center networks. Software Defined Networking (SDN) provides centralized functionalities/databases for handling network and service resources [16–19], also enabling fast service setup [20]. Recently, NETCONF has been standardized by the Internet Engineering Task Force (IETF) to fulfill requirements to IETF in developing a standard for both control (e.g., setup) and management (e.g., monitoring) of network devices [21,22]. In particular, such protocol standardization is an answer to specific requirements to the IETF [23]: i) developing standards for network configuration and management; ii) use eXtensible Markup Language (XML) for data encoding. NETCONF protocol provides mechanisms to install, manipulate, and delete management states and information of network devices. NETCONF has also the prospects to be particularly indicated for monitoring purposes, through the use of specific messages (i.e., notifications).

NETCONF messages are based on YANG model [22], which is a IETF-standard language developed to describe network devices into NETCONF. YANG is attractive for operators because permits to describe devices in a vendor-independent way [24], thus facilitating the implementation of multi-vendor networks. Although NETCONF protocol is a standard, the content of NETCONF messages – thus, YANG models – is still a subject of studies and research [24–28]. A huge effort is directed in the development of YANG models for vendor-neutral network elements and pluggable devices (e.g., white boxes composed of modules from different vendors). Such effort is demonstrated by the several industrial consortiums and projects (OpenConFig [26], OpenROADM [25], IETF [28]) operating in this field, ranging from data centers to transport networks, and including the most important companies in the world.

In this manuscript, which represents an extension of [29], we present a software-defined reconfigurable transponder based on C-band VCSEL for intra-data center transmission. Similarly, through the use of 1.3µm-VCSEL, the same concept can be easily extended to transmission links for access networks. The transponder is programmable by a centralized controller and can self-adapt its configuration based on the feedback from a monitoring system. The transponder supports bit rate at 10 and 25 Gbps with on-off keying (OOK) intensity modulation, it is modeled with YANG and controlled/managed by NETCONF protocol. The employed VCSEL exhibits an emission wavelength in the C-band but the same modulation performance is guaranteed by similar prototypes in the 1.3µm range [7,12]. Experiments are reported for uncooled and cooled data centers showing the ability of the system to reconfigure itself. In particular, a first example is reported showing the capability of the system to automatically react by suitably reconfiguring the VCSEL working condition so as to compensate for temperature variations (from 35 °C to 45 °C in the exposed example) in case of uncooled system and 25 Gbps transmission. For temperature variations slower than 1.3 °C/s the control plane is able to avoid service outage, whist for faster variations an outage time in the range 8–11.5 s occurs. Similarly, a second example shows a suitable self-reconfiguration of the transponder data rate from 25 Gbps to 10 Gbps, in case of soft failure (optical power decrease) for cooled system, so as to re-establish successful (error free) transmission after service outage messaging. NETCONF messages are shown for data center transmission configuration and for notifications of monitoring information, e.g. bit error rate (BER) or service outage. NETCONF can also re-program the bit rate based on physical constraints.

The manuscript is organized as follows: in Section 2 a static and dynamic characterization of the employed VCSEL is reported for various temperatures including chirp measurement in case of direct intensity modulation. Section 3 concerns the characterization of the VCSEL-based transponder at both 10 and 25 Gbps. Section 4 focuses on the use of NETCONF for the control and management of the transponder. Section 5 reports the experimental results in the two considered scenarios: cooled and uncooled system. Finally Section 6 sums up the conclusions related to the presented work.

2. High-speed C-band VCSEL characterization

The proposed solution for the data plane of the intra-data center (detailed in the next section) is based on a high-speed long-wavelength (LW)-VCSEL with emission wavelength in the C-band (~1530 nm in this particular case) designed and fabricated by Vertilas GmbH. In order to understand the potential of such device, especially in terms of fiber reach and transmitted information rate, a full characterization has been first conducted. In this sense, the main key features consist of the modulation impulse response, optical output power, achievable extinction ratio and the introduced chirp associated to its direct intensity modulation.

Optical output power and corresponding bias voltage have been measured as functions of the driving bias direct current (DC) for temperatures ranging from 25 °C to 45 °C as shown in Fig. 1. A threshold current of about 1.5 mA is measured and a roll-over occurs between 18 and 20 mA where the maximum output power is reached. The maximum achievable optical power decreases as the device temperature increases, ranging from 4.7 mW (T = 25 °C) down to 3.65 mW (T = 45°C) and the corresponding diode voltage (Fig. 1(b)) leads to typical electrical power consumption below 37 mW.

 

Fig. 1 Measured optical output power from the VCSEL (a) and measured bias voltage (b) as a function of the applied bias current, for different operation temperatures.

Download Full Size | PDF

The investigated LW-VCSEL exhibits a modulation bandwidth as high as 17.3 GHz at 25 °C (Fig. 2). The modulation bandwidth response (S21 parameter) for various temperatures has been measured through a vector network analyzer (VNA) and a 40 GHz photodiode placed at the VNA receiver port. In particular, a bias-tee was used to couple the bias DC component and the radio frequency (RF) signal coming from the VNA which was set so as to have a peak-to-peak voltage equal to 0.2 V. The 3dB-bandwidth of the modulation response degrades with increasing the temperature, mainly due to the impact of temperature on both the threshold and the roll-over current therefore affecting the resonance frequency. This can be appreciated from Fig. 2, where the −3 dB level is also reported. The curves shown (lines) are fits of the measured raw data (dot lines) through a typical 3-pole transfer function which takes into account both intrinsic and parasitic influences on the modulation response of the device [30,31]. In particular, for temperatures equal to 25–35–45–55 °C a 3dB-bandwidth equal to 17.3–15.3–14–11 GHz has been measured, respectively. The frequency notch at about 13.5 GHz and the peak at about 21.8 GHz, both present in each data plot, are due to imperfections of the employed RF setup that could not be compensated during the VNA calibration process.

 

Fig. 2 Modulation frequency response with bias current of 11 mA and for temperatures ranging from 25°C to 55°C, obtained by fitting the raw data (dotted lines) with a three-pole transfer function modeling the VCSEL response (lines).

Download Full Size | PDF

As said before, another key feature when making use of directly-modulated (DM)-VCSELs consists in the dynamic frequency chirp associated to the applied RF modulation current. Indeed, as VCSELs exhibit a cavity length hundreds of times shorter than typical edge-emitting lasers, the needed current density to induce intensity modulation of the output power is very high therefore causing high refractive index changes [31]. The tight power budget caused by such aspect represents the main concern in employing VCSELs for applications requiring C-band wavelengths and SMF links. In order to quantify the chirp induced when modulating the VCSEL, both in its adiabatic and transient components, the method detailed in [32] has been applied. In particular, a 5 Gbps OOK intensity modulation obtained through the DM-VCSEL was injected into a liquid crystal-on-Silicon (LCoS)-based programmable WaveShaper (WS) by Finisar, employed to emulate the necessary Mach-Zehnder interferometer (MZI) with a free spectral range equal to 150 GHz. Through user-friendly software it was possible to suitably spectrally shift the WS spectral response with respect to the signal central wavelength, so as to obtain the three intensity traces required for the calculation of the frequency chirp temporal evolution [32]. The traces were acquired through a 20 GHz real-time sampling oscilloscope with sampling rate of 50 GS/s. Both the sampling rate and the oscilloscope bandwidth imposed a limitation of the modulation rate that was chosen to be 5 Gbps so as to enable accurate reconstruction of the temporal signal shapes.

The experimental result is shown in Fig. 3, where both the temporal intensity and the temporal frequency chirp are reported. From the instantaneous chirp behavior in Fig. 3(b), it is quite evident how the adiabatic chirp term, proportional to the temporal intensity modulation, is clearly predominant with respect to the transient term (proportional to the derivative of the intensity logarithm), as described in details in [33]. An average adiabatic chirp of 12.5 GHz was measured for a bias current of 11 mA and temperature set to 35 °C.

 

Fig. 3 Temporal intensity of the generated 5 Gbps OOK signal through direct modulation of the VCSEL (a) and resulting temporal frequency chirp measured through the method reported in [32] (b), for a peak-to-peak voltage of the applied data signal equal to 0.7 V, a bias current of 11 mA and temperature equal to 35 °C.

Download Full Size | PDF

3. VCSEL-based transponder for intra-data center networks

The proposed software-defined VCSEL-based transponder is depicted in Fig. 4, which also incorporates technical details referred to the experimental demonstration explained in Section 5. A bit pattern generator (BPG) produces a binary data stream that can exhibit a bit rate of either 25 or 10 Gbps. A bias-tee is used to couple the bias current (I_bias) with the data signal, so as to obtain the driving signal needful to directly modulate the intensity of the laser. The employed VCSEL, fully characterized in the previous section, is a high-speed die laser supplied by Vertilas, operating in the C-band. The die VCSEL was probed through RF microprobe and light was coupled out through a cleaved SMF pigtail in front of the VCSEL output, so as to limit optical reflections. To emulate a coarse WDM (CWDM) scenario, a CWDM filter (CWDMF) has been placed at the laser output. An optical power monitor at the transmitter (Tx) side is employed to check the power at the filter output. At the receiver (Rx) side, direct-detection is exploited through a photo-receiver (PRx) that drives a BER tester (BERT). In a CWDM scenario the transponder architecture has to be scaled up considering a similar Tx and Rx of those shown in Fig. 4, for each wavelength channel. Then, a device controller module is included in the device to process NETCONF messages (e.g., for the device configuration) and to control the bias current (as explained in Section 5). The device controller is connected to a centralized controller, which decides the transponder configuration settings and it is notified about an outage. In future implementations, this device controller module could be externalized at the switch level and shared by multiple transponders in order to reduce the cost and power consumption. More on the controllers and NETCONF messages will be given in the next section. Similarly, the module for temperature control and monitoring will be detailed in the experimental demonstration of the whole system including both data and control plane (Section 5).

 

Fig. 4 VCSEL-based transponder and setup for the experimental demonstration. Tx: transmitter; Rx: receiver; BPG: bit pattern generator; SMF: single-mode fiber; CWDMF: coarse WDM filter; PRx: photo-receiver; BERT: BER tester.

Download Full Size | PDF

A preliminary transponder testing was conducted to analyze its performance at both 10 Gbps and 25 Gbps. Figure 5 shows the BER performance of the transponder for the two data rates together with the optical signal spectrum for a temperature of 25 °C, 35 °C and 45 °C respectively. The temperature influences the laser central emission frequency of about 20 GHz/°C. The general trend is a worsening of the performance by increasing both the temperature and the bit rate. However, for 10 Gbps the BER curve at 35 °C performs slightly better than the one at 25 °C and this is due to the chosen reference bias current (8 mA for both BER measurements) that represents the optimum working condition in that particular case. All the reported cases perform fine for bias current in the range 7.5 mA–11.5 mA. The data peak-to-peak voltage was 0.7 V which represents a tradeoff condition for all the considered cases.

 

Fig. 5 Performance of VCSEL-based IM-DD for 10 Gbps (a,b) and 25 Gbps (c,d) OOK. Signal spectrum for temperature equal to 25, 35 and 45 °C (a,c) and corresponding BER curves versus the received optical power (b,d).

Download Full Size | PDF

4. NETCONF protocol

NETCONF is an SDN protocol used to control (e.g., configuration) and manage (monitoring) network devices. Network devices can be described through the standard YANG data modeling and controlled through NETCONF messages accordingly [22]. NETCONF is based on remote procedure calls (RPC) communication model. NETCONF natively operates in a client-server way, with the centralized controller as a client and the device controller as a server. NETCONF peers (server and client) enclose requests and responses inside <rpc> and <rpc-reply> messages, respectively [21]. Main NETCONF operations are shown in Fig. 6. Initially, the centralized controller (client) becomes aware of the type of the device through the exchange of Hello messages (not shown) between the centralized and the device controllers. The type of device (i.e., transponder) can be defined through YANG models [22]. In this paper, we rely on a YANG model for transponders proposed in [27]. The model generically assumes the possibility to set several transmission parameters such as the bit rate, the modulation format, the baud rate, and others. Given that in an intra-data center scenario, the transponder has to be simpler than in long haul, some of these parameters could be fixed and, as admitted by the YANG model, are not configurable by an external entity (e.g., we consider OOK as the only supported modulation format).

 

Fig. 6 NETCONF message exchange.

Download Full Size | PDF

Then, the centralized controller may retrieve information about the transponder (e.g., the supported bit rate values), the full (or partial) state of the device by sending the Get message (Fig. 6(1)) to the device controller: e.g., the list of supported modulation formats (OOK in our case) and bit rates. The Edit-conf message (Fig. 6(3)) allows the centralized controller to configure the device (e.g., the bit rate) for accommodating a service. NETCONF also provides a way for the centralized controller to be notified in case specific events occur. To this purpose, the centralized controller needs to subscribe to the event by sending the Create-subscription message (Fig. 6(5)) to the device controller. Then, the device controller sends a Notification message (Fig. 6,(7)) to the centralized controller if the event occurs: e.g., if a monitored parameter falls within a critical range.

 

Fig. 7 (a) Edit-conf message capture for creating a connection, (b) create-subscription message capture and (c) outage-record message capture.

Download Full Size | PDF

The aforementioned messages are the basic operations defined in the NETCONF and YANG standards [21,22]. In addition, NETCONF and YANG allow defining new functions to control/manage the device. We also include an RPC message named Outage-record (whose content will be shown in the next section) to enable or disable the recording of the outage time of a specific connection. The outage time is defined as the lapse of time the transmission experiences a BER above an outage threshold. When the outage record is enabled, the related Connection-outage notification message reports the threshold value and two time instants: when the BER exceeds the threshold (outage begins) and when it returns below the threshold (outage ends). This way, the centralized controller can be aware of the duration of the outage to verify that service level agreements (SLAs) are satisfied. The content of the message will be shown in the experimental demonstration.

5. Experimental demonstration

Experiments are carried out in a testbed integrating data (Fig. 4) and control/management (Fig. 6) planes. Two scenarios have been considered: i) uncooled, emulating a data center with limited temperature control; ii) cooled, emulating a data center with effective temperature control. For both the scenarios, a CWDM 5nm-grid is considered. In particular, the system performance has been tested on a link as shown in Fig. 4, constituting of approximately a 100 m-long SSMF span. To be noted that the length of such fiber span can reach up to about 2 km in case of 1.55µm-VCSEL (like the one employed here) and up to 20 km in case of 1.3µm-VCSEL [12]. As implicitly attested by the chirp characterization shown in Section 2, for the 1.55µm regime the main limiting factor to the maximum fiber reach consists of the quite high adiabatic chirp contribution. Sections 2 and 3 also highlighted how the temperature affects the VCSEL performance (BER). Moreover, the temperature affects the VCSEL central frequency (Fig. 5) and, as a consequence, the spectral position of the signal with respect to the CWDMF (a 3nm-bandwidth Lorentzian-shaped filter centered at 196.2 THz in the reported example), thus causing an extra power loss depending on such detuning. The characterization for temperatures ranging from 25°C to 45°C showed a frequency tuning in the range 196.1−196.5 THz.

Connection setup and management integrating data and control planes have been demonstrated. In both cooled and uncooled scenarios, the centralized controller first creates a new connection by issuing an Edit-conf message (Fig. 7(a)) containing all the parameters to be configured (e.g., 25 Gbps). Once the connection is created, the centralized controller subscribes to the notifications issued by the transponder. Two notifications have been considered in this work: pre-FEC-BER-change reporting a variation of the BER over the outage threshold of 10⁻⁷ (corresponding to a power threshold of −26.6 dBm at the power monitor reported by Fig. 4, that approximately represents 1% of the necessary received power to guarantee BER < 10⁻⁷, net of both splitting ratio variability and insertion loss of SMF connections); connection-outage reporting the connection outage time. Figure 7(b) shows the message sent by the centralized controller to subscribe to the notifications. Then, the controller sets the device to record the outage time in case of outage, specifying the connection id and the outage BER threshold (outage-record message in Fig. 7(c)).

Uncooled scenario: a software was also implemented into the device controller to adapt the bias current based on the transmitter output power observed by the power monitor. Indeed, due to a temperature change, the consequent shift of the carrier frequency induces a power reduction. In particular, once the monitored power falls below a threshold due to the frequency detuning, the device controller triggers a change of I_bias to return the system at above-threshold transmitted power (and consequent lower BER value), possibly avoiding the outage. The threshold has been selected to be −24.5 dBm at the power monitor, corresponding to a BER of 10⁻⁹, thus having enough margins to avoid the outage (such threshold ideally is 1% of the necessary received power to guarantee BER < 10⁻⁹, net of both splitting ratio variability and insertion loss of SMF connections).

In the example shown in Fig. 8(a), the 25 Gbps transmission is optimized at 35 °C and bias current I_bias = 11 mA (yellow spectrum placed at the CWDMF center). A heating up to 45 °C on the uncooled board (magenta spectrum), would not allow successful transmission due to the extra loss induced by the frequency detuning (~8 dB). For that reason, during the heating process, I_bias gets progressively decreased down to 8 mA (red spectrum) by the control system, so as to maintain successful 25 Gbps transmission as attested by BER curves of Fig. 8(b) at 35 °C with bias of 11 mA (yellow circles) and 45 °C with bias of 8 mA (red circles), considering that the received power was > −7 dBm.

 

Fig. 8 Optical spectra and CWDM filter for uncooled example (a) and BER curves for uncooled and cooled examples (b). Experimental validation of the uncooled scenario (c-f); emphasized effect of the bias control on both the transmitted power and the BER performance in the zoom (g-l).

Download Full Size | PDF

In Figs. 8(c)-(f), the monitored power, the VCSEL bias current, the BER value at the receiver and the temperature evolution in time are reported. In detail, a temperature increase at the transmitter from 35 °C to 45 °C with a time duration of about 150 s is emulated through a Peltier Cell. The consequent optical frequency shift causes an optical loss at the CWDMF output. This, in turn, leads to a decrease of the monitored power whose overcoming of the mentioned threshold triggers a decrease of the applied bias current to compensate for the spectral shift. In order to keep the BER performance above the threshold of 10⁻⁹, during the heating process the bias current gets decreased from 11 mA to 8 mA (Figs. 8(c)-(f)). Figures 8(g)-8(l) represents a magnification of the same example to better appreciate the effect of the bias decrease on both the optical power at the filter output and the 'instantaneous' BER performance. In this case no notification is sent to the centralized controller since outage BER threshold (10⁻⁷) is never exceeded thanks to the control action performed locally.

The uncooled scenario was then tested for the extremely critical case of faster temperature variations, namely from 0.7 °C/s up to 3 °C/s. For those cases, no outage threshold was exceeded up to 1.2 °C/s temperature variation. For faster variations ranging from 1.3 °C/s to 3 °C/s as reported by Fig. 9, BER performance degrades over the threshold. The figure shows for each tested case the total outage time duration (Fig. 9(a)), together with the starting and ending times (Fig. 9(b)). The outage is due to the limited response speed of the control system, which actually is not able to update the bias current as fast as needed. The faster the temperature variation, the sooner the outage occurs. The average outage time duration is 9.5 s with a maximum for a temperature variation of 2 °C/s most likely due to the temperature transient when reaching 45 °C. For temperature variation of 3 °C/s the outage duration decreases as the temperature change is much faster than the time needed by the system for suitably reconfiguring the bias current. Since an outage is experienced, the monitoring plane triggers the notifications. In the next subsection, examples of notifications will be shown.

 

Fig. 9 (a) Outage time duration for fast temperature increases in the uncooled scenario. (b) Starting and ending outage time versus the temperature variation speed. (c) Temperature evolution in time for each considered case.

Download Full Size | PDF

It is worth to point out that in general, by making use of tunable VCSELs [34], the maximum range of temperature variation for successful transmission (25°C-45°C for the presented implementation) could be in principle extended and the necessary compensation to the frequency detuning due to the temperature variation could be achieved by tuning the VCSEL emission frequency with no need to change its bias condition. Indeed, bias tuning itself impacts the transmission performance in terms of VCSEL emission power, speed of modulation response (see Figs. 1 and 2) and modulation stability, and this is the main reason for the limited working temperature range demonstrated here. However, tunable VCSELs are presently available only for 10 Gbps transmission [34], even if their further development is expected soon.

Cooled scenario: a soft failure (i.e., a degradation of the VCSEL performance) is emulated by decreasing the emission power of the laser through a variable optical attenuator. The outage BER threshold of 10⁻⁷ is exceeded, thus pre-FEC-BER-change notification is triggered (Fig. 10(c)). Such BER threshold corresponds to a monitored power equal to −26.6 dBm. Both the thresholds are reported in Figs. 10(a) and 10(b) where both the optical power and BER evolution in time are shown. After the notification, the centralized controller can decide to downgrade to 10 Gbps to relax transmission. In principle, such bit rate reconfiguration functionality can be implemented in the uncooled scenario too. BER variation is reported in Fig. 10(b): in the example outage is experienced until the bit rate is reconfigured at 10 Gbps. After that, BER turns below the threshold and the device reports the outage time to the centralized controller (Fig. 10(d)). In the experiment the board temperature was set to 35 °C, I_bias = 8mA and BER curves at both data rates are included in Fig. 8(b) (green circles for the 25 Gbps case, black squares for 10 Gbps). Moreover, T_outage = 15 s + T_R, where T_R is a reconfiguration time of about 4 s during which the data were turned off, which is due to hardware (BPG in this case) reconfiguration latency. 15 s is a sort of “guard time” that the system waits before reconfiguring the transmitted bit rate.

 

Fig. 10 Experimental validation of the cooled scenario: optical power at the monitor (a) and instantaneous BER value at the BERT (b) versus time. (c) Pre-fec-ber-change and (d) Connection-outage notification message capture.

Download Full Size | PDF

6. Conclusion

We proposed and demonstrated a software-defined reconfigurable 10/25 Gbps transponder based on VCSEL for short reach applications, e.g. intra-data center transmission. A centralized controller programs the configuration of the transponder, which can tune its transmission configuration (e.g., bit rate) based on feedback from a monitoring system (e.g., in case of transmission performance degradation). Experiments are carried out in a testbed integrating data and control planes. Cooled and uncooled scenarios are considered and the system performances are tested on a link of 100 m-long SMF span. NETCONF protocol, including YANG model, is exploited for the transponder control and management. The transponder capability to react to physical impairments (i.e. temperature variations, power decrease) has been verified for both uncooled and cooled scenarios and a mechanism of subscription/notifications is exchanged in the control/management plane to take record of an outage and of the outage time in order to verify service level agreements. NETCONF messages for the configuration of transmission were shown, as well as management messages to report monitoring information about affected services. In particular, we proposed messages reporting an excess of BER above the threshold, thus inducing a service outage, and the duration of the outage. In the uncooled scenario, service outage is avoided for slow temperature variations (up to 1.3 °C/s) when the transponder operates at 25 Gbps, while in the cooled scenario dynamic system reconfiguration from 25 Gbps to 10 Gbps allows to recover the service outage due to a permanent signal power decrease.

The presented work represents, to the best of our knowledge, the first integration of NETCONF protocol and YANG modeling, with a VCSEL-based transponder in the context of data centers. Furthermore, the proposed transponder is based on a novel high-speed VCSEL, therefore guaranteeing superior performance and degree of flexibility in terms of bit rate, temperature and bias current working conditions. By the use of a 1.3µm-VCSEL instead of the 1.55µm one employed here, a similar demonstration could be easily extended in the context of access networks and the 25 Gbps capability potentially supports 100G standards based on 4x25Gb/s channels, such as the IEEE 802.3 100GBASE LR4.