Open Access
Add to BibSonomy
Abstract
Space division multiplexing elastic optical network (SDM-EON) enables high-capacity transmission, in which the network nodes should provide high switching flexibility while limiting the complexity and costs of nodes. Architecture on demand (AoD) nodes can meet these requirements but the slow configuration time of optical backplane in AoD nodes makes it difficult to serve latency-sensitive requests. In this paper, we propose a hybrid optical backplane based on micro-electromechanical systems (MEMS) and semiconductor optical amplifier (SOA) switches to provide fast configuration time for AoD nodes. Moreover, we propose quantitative measures of node switching flexibility in SDM-EONs and of link configuration speed in AoD nodes. Based on the hybrid backplane architecture and the measurement approaches, we propose a flexibility and fragmentation aware routing, spectrum and core allocation algorithm and an AoD synthesis algorithm. Simulation results show that the hybrid AoD nodes with support of spatial lane change can reduce network blocking probability. The AoD based on this hybrid backplane structure can improve the network performance by 32.8% compared to the AoD based on the traditional MEMS. Compared with traditional reconfigurable optical add/drop multiplexers (ROADMs), the hybrid AoD nodes can control the number of wavelength selective switch (WSS) ports.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
In recent years, with the rapid growth of Internet traffic which challenges the optical backbone infrastructure, elastic optical network (EON) has been proposed to be a spectrally efficient solution to fulfill the demand of disparate traffic granularity, raising the utilization of spectral resources and network capacity. By exploiting fiber capacity to the full, space division multiplexing (SDM) over multi-core or multi-mode fiber would allow the optical transport network to keep pace with traffic growth beyond the petabit per second level, which would be a promising technology for the future bandwidth-hungry applications [1,2].
In the meantime, colorless, directionless and contentionless (CDC) operation of reconfigurable optical add/drop multiplexers (ROADMs) has become a commercial reality for EON. Such switching node architectures can also be upgraded to route signals between different spectral and spatial dimensions in SDM-EON. As a key transmission technology of SDM-EON, super channel (SpCh), which consists of several optical carriers allocated in the spatial or spectral domains, can enable both higher spectral efficiency and cost saving by component sharing and dense electronic integration [3–5]. Spatial lane change (SLC) technology in SDM can alleviate the impact of spectral continuity constraints by allowing connections to use different spatial dimension indices on each link along the light path while still using the same spectral range [3,6]. However, spatial lane change is supported at the cost of more complex interconnection and higher port-count wavelength selective switches (WSS) in ROADM. Accordingly, the cost of this technology needs to be balanced against the benefit of routing flexibility. It is necessary to evaluate the relationship between the transmission technologies and their influences on the flexibility.
To reduce the number of building modules in optical nodes, [7] introduced a novel optical node architecture called an architecture on demand (AoD) node. As an important composition of AoD, an optical backplane, which is usually implemented using a large-port count three-dimensional microelectromechanical system (3D-MEMS), interconnects input/output ports and building modules in an arbitrary manner. However, the slow switching times, typically in milliseconds, make it only suitable for traffic flows that are not latency sensitive. On the contrary, semiconductor optical amplifiers (SOAs)-based optical switching circuits offer the prospect of ease of control, low operating voltage, inherent optical gain, broadband performance and most importantly, the capability to reconfigure in nanoseconds. Large SOA-based switch fabrics can be achieved by cascading switch elements [8]. Integrated SOA-based optical switches have been demonstrated with various port counts by several research groups [9–12]. By using this new SOA-based optical switches for the backplane of AoD node, the configuration speed can be greatly improved. The small size of SOA-based switches can also bring more benefits to AoD node fabrication. In addition, new routing and resource allocation algorithm are required to accommodate these new features introduced by AoD nodes and SOA-based switches.
In this paper, we elucidate the flexibility of switching nodes quantitatively in SDM-EONs and propose a new hybrid AoD backplane architecture based on MEMS and SOA switches, which supports spatial lane change while controlling the required number of WSS ports within a reasonable range. Flexibility and fragmentation aware routing, spectrum and core allocation algorithms are proposed based on this architecture to reduce network blocking probability. We also present simulation results to show the performance of our proposed architecture and algorithms for latency-sensitive requests in different networks.
This paper is organized as follows. In the next section we consider the related works about optical modules, transmission technologies and node architectures for SDM-EONs. In Section 3. we propose a MEMS & SOA switch-based architecture for AoD backplane composition. Section 4. presents flexibility measurement approach and flexibility of optical fibers to distinguish different connections in our proposed architecture. In Section 5, we propose fragmentation and flexibility aware routing, spectrum and core allocation algorithms and synthesis technique for hybrid AoD nodes. Section 6. presents simulations and results. Conclusions are presented in Section 7.
2. Related works
2.1 Traditional optical node architectures and transmission technology
Reconfigurable optical add-drop multiplexers (ROADMs) have been deployed as key building blocks in many networks, which can switch the direction of wavelength-division-multiplexing(WDM) signals and add/drop wavelength to/from any direction. Numerous ROADM architectures for SDM networks have been proposed [13–18] . The fundamental idea behind these architectures is to balance the trade-off between ROADM cost, power consumption and routing flexibility. With the widespread use of multi-core fiber (MCF), it is necessary to suppress the increase in the number of WSS ports due to the increase in fiber cores when designing the architecture.
To improve bandwidth utilization and reduce device complexity, switching strategies with different granularities in SDM network have been proposed. In order of decreasing flexibility and increasing hardware efficiency, three SDM switching strategies have been evaluated [1,5]: independent switching (InS), fractional joint switching (FrJoS), and joint switching (JoS), as it is shown in Fig. 1. Spatial super-channels have been achieved in these switching strategies. Therefore, multi-core fiber fan-out is needed to separate multiple cores into groups of spatial channels. Fig. 1 shows switching node architectures based on route-and-select (R&S) architecture with three nodal degrees ($D=3$) in SDM network with four spatial dimensions ($K=4$). The spatial dimensions are fiber cores here. Note that only connectivity from "East" is shown. Spectrum selective switches (SSS) switch spatial super-channels as groups of G spatial dimensions’ sub-channels, where G is a divisor of K. The three switching strategies can be defined as cases where G takes different values, i.e., $G=1$ for InS strategy and $G=K$ for JoS strategy. In Fig. 1 (a), each spatial dimension is routed independently whereas all spatial dimensions are routed together in Fig. 1 (c). FrJoS, as an intermediate solution, routes groups of G spatial dimensions jointly. Spectrum selective switches have replaced fixed-grid wavelength selective switches to support elastic spectrum allocation for finer granularity. More ports provided by SSSs enable the ability to establish connections between different spatial domains, i.e., spatial lane changes (SLC) are supported.
Higher routing flexibility is enabled with the support of SLC technology [6,19,20], as shown in Fig. 2(a). Spectral efficiency is higher in this scenario. As a comparison, traditional SDM-ROADM without SLC support introduces the spatial continuity constraint, i.e., spectrum should be allocated on the same spatial channel index along the end-to-end routing path, as shown in Fig. 2(b).
However, the complexity and costs of network nodes are significantly increased due to the required higher number of WSS/SSS ports if SLC is supported. Figure 3 illustrates InS and FrJoS strategies (corresponding to different group values $G=1$ and $G=2$) with and without SLC support at an intermediate node with 3 degree [3]. The number of the spatial dimension is assumed to be four ($K=4$). Independent switching strategy ($G=1$) without SLC support, as shown in Fig. 3(a), switches each spatial dimension independently to any output port with the same spatial dimension index (i.e., constrained by spatial continuity). With the support of SLC, spatial continuity constrain is removed and routing flexibility increases sharply, as shown in Fig. 3(b), at the cost of higher port count per WSS/splitter and increasing architecture complexity. Figure 3(c) shows fractional join switching ($G=2$) without SLC support. By binding 2 spatial dimensions, the complexity of the switching architecture is alleviated and the number of WSS declines. However, this switching strategy raises the probability of spectral fragmentation, which will degenerate system performances. Figure 3(d) provides a compromise strategy switching cores by group ($G=2$) with SLC support, which cuts the number of WSS but increases port count. In general, it’s required to support spatial lane change to offer greater flexibility as well as keep the port count of WSS in check.
2.2 SOA-based optical switches
Semiconductor optical amplifiers (SOAs) switches have the advantages of inherent gain, nanosecond switching times, broadband operation, and high ON/OFF extinction ratios, but are relatively power hungry. The accumulation of amplified spontaneous emission (ASE) noise and saturation-induced distortion limits the ultimate size of such switch fabrics [21]. SOA-gate based integrated switch fabrics have been mainly implemented in the broadcast-and-select and wavelength-selective configurations. Each path can be gated by one SOA element and its inherent gain opportunely overcomes the fan-out/fan-in losses while its high ON/OFF extinction ratio ensures excellent crosstalk suppression [22].
Multistage architectures involving cascaded SOA elements certainly enable larger switch fabrics but are still challenging to realize. A scalable photonic interconnection network architecture implementing broadcast-and-select sub-stages is proposed [23]. The first monolithically integrated three-stage 16${\times }$16 SOA-based switch with 480 integrated components is subsequently demonstrated [9], as shown in Fig. 4 (a). The key requirements for large port count optical switch fabrics are addressed noting the need for switches with substantial port counts. Shortly after, an equivalent active-passive monolithically integrated 16${\times }$16 switch is reported for enhanced power efficiency and optical signal-to-noise-ratio (OSNR) [12]. The improved performance and/or further scale-up would require a large reduction in component level excess losses, a more careful design of balancing the summed loss with the SOA gain per stage, and a close examination of SOA designs for linear operation [22]. An 8${\times }$8 SOA-based switch implemented in a three-stage Clos architecture is proposed in [24], which offers on-chip lossless connections with a wide input power dynamic range (IPDR). The feasibility of building a 64${\times }$64 port count SOA switch using the modular architecture as in the 8 x 8 switch block is demonstrated via a physical layer simulator by the same group [8]. Moreover, dynamically reconfigured wavelength-and-space routing has been performed for automated control and path assessment [25,26]. Although it is challenging to monolithically integrate such a large device due to the difficulty of a uniform wafer process, we believe that manufacturing process problems will be overcome, and large SOA-based switch is the trend of fast switching in the future.
Fig. 4. (a) 16${\times }$16 InP-based all-active SOA switch [9]; (b) Architecture on Demand implementation
2.3 Architecture on demand node
To provide greater flexibility by enabling on-demand architecture provisioning, architecture-on-demand (AoD) node is proposed and demonstrated experimentally [27]. This flexible architecture enables on-demand spectrum defragmentation and time multiplexing for subwavelength granularity. Most importantly, functional modules are not hard-wired like in a static architecture but can be interconnected in an arbitrary manner, which makes the node able to provide new functionalities in the future. Figure 4 (b) shows a possible implementation of AoD with an optical backplane that interconnects input ports, output ports and functional building modules. The optical backplane can be implemented with a large port-count optical switch (e.g., micro-electro-mechanical system, semiconductor optical amplifiers based optical switch). The building module can be either a single device such as MUX, DEMUX, wavelength selective switch (WSS), splitter, semiconductor optical amplifier (SOA), or a subsystem composed of several devices. By connecting these modules and subsystems dynamically, AoD can provide a variety of network functions.
Multigranular transmission in the space/frequency/time domain has been demonstrated using AoD nodes and MCFs [28]. They present great flexibility of programmable optical nodes able to switch traffic utilizing the space, frequency and time dimensions with over 6000-fold bandwidth granularity. The flexibility of AoD nodes is firstly defined in [27] based on the entropy of the system. They compare the flexibility of optical modules, traditional static nodes and AoD. The advantage on the flexibility of AoD nodes has been proved both theoretically and experimentally. A flat-structured data center network (DCN) powered by AoD nodes is demonstrated [29]. In this network, clustered DCNs are connected through metro/core networks using all-optical SDM/WDM converters according to the AoD concept. Moreover, the flexibility of AoD nodes can also be used to reduce the number of implemented modules and the power consumption of optical nodes. By using a heuristic algorithm to construct AoD nodes, the total power consumption can be saved by more than 25% [30]. Thanks to the great flexibility to reconfigure the structures dynamically, AoD nodes are future-proof to add more features on demand.
Optical node architectures are evaluated in terms of optical loss, modularity, scalability but optical node flexibility is not widely studied. In [27] a systematic and quantitative measure was proposed based on entropy in information theory, but extensive research on optical node flexibility in SDM-EON network and their application is not studied. Work related to routing and spectrum allocation algorithm taking account of flexibility is also needed.
2.4 Resources allocation problem and synthesis technique for AoD nodes
The routing, spectrum and core assignment (RSCA) problem is one of the important issues of SDM-EONs from the perspective of networking. The RSCA tries to find a suitable route between a given source/destination pair and a proper core in each link of the route for an arriving connection request and allocates the needed bandwidth of the connection in continues frequency slots with/without fixed starting and finishing indexes in all of the selected route links [31,32]. In optical networks, various network services are requested dynamically, and the network traffic changes greatly and rapidly with time. Therefore, it is important to use dynamic approaches to solve the RSCA problem [33]. The routing and spectrum (RSA) problem is equivalent to the routing and wavelength assignment (RWA) problem in traditional WDM networks. However, the RSA problem exhibits different characteristics owing to its flexible resource management.
Spectrum fragmentation is one of the issues that must be challenged in the RSA problem. There are some works that have studied fragmentation problems in EON and have proposed methods to decrease its rate [34–37]. The concept of cross-core virtual concatenation has been presented in SDM networks with the spectrum contiguity constraint relaxed [38]. The authors have considered that spectral-spatial super channels have irregular shapes and their carriers can be distributed over different fiber cores. They minimize the fragmentation of EON-SDM multi-core fibers based on this concept. The work in [39] defines a threshold for distance of light paths. The paths with higher distance than the threshold use the part of the spectrum which its fragmentation is lower.
When AoD nodes are implemented for SDM-EONs, the RSCA problem becomes more complicated. The resource allocation should consider not only the resource utilization efficiency, but also the synthesis and reconfiguration of AoD nodes. Reference [40] numerically analyzes RSA solutions in MCF-based EONs that reduce inter-core cross talk of MCFs considering AoD nodes. The proposed RSA strategy also reduces the number of SSSs and amplifiers in AoD nodes. Reference [41] demonstrates and evaluates self-healing capabilities of AoD nodes arising from their flexibility and ability to employ idle components for failure recovery. They propose a routing algorithm which obtains a targeted portion of lightpaths switched at the fiber level to improve efficiency of self-healing by increasing the number of idle components within nodes. The dynamic resource allocation considering the monetary cost of AoD nodes has been studied in [42,43]. An energy-efficient network system that includes a novel energy-efficient AoD node architecture and resource assignment method has been proposed in [44]. The proposed system solves the power-consumption problem by simplifying the implemented building modules based on spatial multiplicity of SDM-EONs.
Although these studies have examined the energy efficiency or monetary cost of AoD nodes, or resources utilization in network, the slow speed of AoD reconfiguration and backplane composition problems have not been fully discussed. The slow switching speed makes AoD nodes hard to serve network requests with strict QoS requirements like latency. QoS and fragmentation aware RSCA algorithm is also needed to cooperate with AoD nodes.
3. Hybrid backplane architecture for AoD nodes
3.1 AoD backplane composition
There are two alternative architectures for backplane composition [45], as shown in Fig. 5. The unidirectional architecture (Fig. 5 (a)) connects $y_U$ optical switches in a unidirectional fashion. The input ports of AoD are connected to the first optical backplane switch and the output ports are connected to the last switch. Successive switch’s input ports are connected to the previous switch’s output ports. The unidirectional architecture is only suitable in a resource dimensioning study carried out before AoD is deployed. Once AoD is operating, the connection of additional backplane switches compromises already established optical links through AoD. As a comparison, the expandable architecture (Fig. 5 (b)) connects $y_E$ optical switches in a bidirectional fashion. The input ports and output ports of AoD are connected to the first optical backplane switch. $2N$ connections ($N$ in each direction) are set between successive backplane switches. This expandable architecture allows us to tailor $y_E$ to the traffic request and the connection of additional backplane switches wouldn’t compromise already established optical links.
3.2 Hybrid MEMS & SOA switch-based backplane architecture
The backplane of AoD has many optical switch implementations. Micro-electro-mechanical systems (MEMS) based optical switches, which have been scaled to hundreds of ports and be of low loss and low power consumption, are already commercially available [46]. However, it takes several milliseconds to switch, which makes them only suitable for long traffic flows that last in seconds. Optical switches with fast switching speed (with reconfiguration times of less than a microsecond to a few microseconds) and with reasonable scalability have been studied and reported [8]. The feasibility of building a 64${\times }$64 port count SOA switch using the modular architecture as in the 8x8 switch block via a physical layer simulator has been demonstrated [8,24]. However, as far as we know, it is not realistic for SOA switches to have large scalability as MEMS. So, we consider combing slow switches and fast switches to exploit their strengths and avoid weakness to offer better network switching functionalities.
Therefore, as shown in Fig. 6, we propose a MEMS & SOA switch (M&S)-based architecture for AoD backplane composition. WSS ports are connected to both MEMS and SOA switches. MEMS and SOA switches are connected in an expandable manner to form the backplane of AoD. The M&S based architecture provides large port count of MEMS for most optical connections to set up, while provides fast switching speed by SOA switches for connections with stringent delay requirements (connection set-up time is mainly considered in the delay requirements). As mentioned before, the traditional ROADM architecture supports spatial lane change at the cost of using WSSs with large port count. As a comparison, by implementing the M&S based backplane architecture, the AoD can set up optical connections dynamically to support spatial lane change while keep the port count of WSSs in check. Moreover, it can also set up some connections in a few microseconds compared to the traditional AoD with MEMS based backplane, which is more suitable for traffic with stringent delay requirements.
4. Flexibility measurement approach
4.1 Flexibility of R&S based switching node architectures
An approach has been proposed to measure flexibility in the context of EON [27]. Here, we consider the influence of some new features introduced by SDM, e.g., spatial lane changes and spatial super-channels, and refine the measurement approach to implement it in SDM-EON.
Consider the problem of switch a signal from any input to any output in Fig. 1(a). Note that switching flexibility is the ability of a system to map inputs to outputs in different ways and across different dimensions, e.g., frequency, time and space [27]. The number of mapping ways is finite. One can associate each distinct map with a different state $s_i$ of the switching node. Let $S={s_1,s_2,s_3,\ldots,s_M}$ be the set of all possible M states. Suppose that for each state $s_i$ there exists a stable value at any instant in time, regardless of the initial conditions, i.e., the system is ergodic. Thus, the system can be modeled as a discrete Markoff process with alphabet $S$ and associated probability $P={p_1,p_2,p_3,\ldots,p_M}$. Therefore, the entropy of the system can be calculated [47] in:
The state probabilities of a device $p_i$ are determined by factors such as traffic load, traffic requirements. However, there is a maximum limit to the entropy set by the properties of the devices itself. We call this maximum entropy the flexibility of the device [27]. Thus, we have :
Obviously, $H(S,P)$ is maximum when all the states are equiprobable, i.e., $p_i=1/M$, where $M$ is the number of different states (also known as the cardinality of $S$). Put $p_i$ into (1), we get
By going back to the definition of entropy [27], if two components $a$ and $b$ with flexibilities $F_a$ and $F_b$ are connected to form a subsystem $(a,b)$, then
with equality only if the components can work independently from one another.Review the switching node architecture in Fig. 1. Switching flexibility is provided in this architecture and hereby calculated. SSS devices at each output may block spectral slots or pass them from one of the inputs or from the A/D module. Determined by the features of SSS devices, a signal cannot be switched simultaneously to several outputs or dropped, i.e., multicasting is not supported. Suppose that SSS has a spectral granularity t times finer than the WSS. Therefore, the switching flexibility of R&S-based node architecture excluding A/D modules is (5) and (6), which supports and does not support lane changes, respectively.
The flexibility of different R&S based switching node architecture with A/D modules has been quantitatively evaluated in our previous work [48]. The results show that the architecture introducing spatial lane changes and implementing InS strategy achieves the best switching flexibility performance.
4.2 Flexibility over times
Equation (3) calculates the flexibility of a system at a fixed instant in time [27]. On this basis, we can get the flexibility of a system that changes its state over a period. Going back to the definition of entropy [47] and Eq. (2), if the system can transition into a new state independently of its previous state, i.e., it is memoryless, and its transition rate is m transitions over a period of time $T$, then
Since traditional ROADM architectures are hard-wired, i.e., they cannot change their states on demand, Eq. (7) is mainly to describe the flexibility of architectures that can be dynamically configured, e.g., AoD architecture. Optical backplane is responsible for architecture configuration in AoD, whose configuration time is related to the flexibility. For example, it takes several milliseconds for MEMS to switch, while the reconfiguration time of some fast optical switches like SOA is a few microseconds. By referring to Eq. (7), AoD equipped with SOA switches is about a thousand times more flexible than AoD equipped with MEMS. In practice, the transition rate m does not only correspond to the inverse of the state switching time. As shown in Fig. 7, the reconfiguration time is the time interval from reception of the reconfiguration signal until the system has stabilized in the new state [27]. There may be a delay in initiating the state switchover if the preprocessing of the reconfiguration signal is required. In general, the flexibility of AoD will benefit from the implementation of fast optical switches.
4.3 Flexibility of an optical fiber link
Optical fiber links can only transmit signals and cannot switch signals. However, every optical fiber link is connected to two WSSs and the connection status of these WSSs can reflect the flexibility of the link, as shown in Fig. 8. Here we assume the optical fibers are single core single mode fibers. According to the reconfiguration time of optical switches, WSS’s ports can be divided into two categories: ports to MEMS and ports to SOA switches. Because of the slow reconfiguration time, it takes several milliseconds for MEMS to set up a new optical channel, which is unacceptable for some latency-sensitive requests. On the contrary, SOA switches can be reconfigured in a few microseconds, which means they can provide in-time and on-demand architecture construction. Furthermore, WSS’s ports connected to MEMS can be divided into two categories: ports that already have established channels and ports that don’t have established channels. If there are channels that have been established, new requests can multiplex these channels instantly without waiting for the set-up time. Therefore, if the channels to MEMS can be multiplexed, there is nearly no difference between the channels established by MEMS and by SOA switches. The new requests that multiplex channels only lose some degree of directional flexibility, since the established channels cannot make any adjustment.
Consider the flexibility of WSS No.1 and WSS No.2 connected by an optical fiber. Assume that there are $M_{1}$ and $M_{1}^{'}$ ports in WSS No.1 and WSS No.2 respectively that are connected to MEMS and don’t have established channels, $M_{2}$ and $M_{2}^{'}$ ports that are connected to MEMS and already have established channels, $M_{3}$ and $M_{3}^{'}$ ports that are connected to SOA switches. By referring to Eq. (3), the number of connection status that we can established instantly is $M=(M_{2}+M_{3})(M_{2}^{'}+M_{3}^{'})$, so we can define the flexibility of an optical fiber link $F_{link} (S)$:
In Eq. (8), $M=(M_{2}+M_{3})(M_{2}^{'}+M_{3}^{'})$ represents the number of all possible connections in the optical fiber link that can be established instantly. Here we only consider the ports connected to SOA switches and the ports connected to MEMS and have already established channels. Since SOA switches are about a thousand times faster than MEMS, by referring to Eq. (7), the flexibility provided by MEMS is negligible compared with the flexibility provided by SOA when establishing on-demand channels. The ports having established channels are good for the flexibility since they can be multiplexed instantly. The flexibility in Eq. (8) is useful when choosing the best core/mode to establish the channel instantly for latency-sensitive requests in SDM network and AoD architecture. The set-up time of the channel is important in this context.
5. Fragmentation and flexibility aware routing, spectrum and core allocation algorithm and synthesis technique for AoD
5.1 Fragmentation metrics
As the traffic volume is dynamically changing, optical channels are required to be established and destroyed continuously, which results in network fragmentation. Some requests may be blocked due to the lack of continuous or contiguous spectrum resources in congested network. Therefore, it is necessary to carefully design the lightpaths to minimize network fragmentation.
The general idea to calculate the fiber fragmentation $F_{link} (e)$ is to average the fragmentation metrics $F_{sm}(e,k)$ of each spatial mode/core $k\in K$
In each spatial mode/core, the spectrum is divided into a set of free and occupied segments. A free segment denotes a set of contiguous free slots in the fiber, while an occupied segment denotes a set of contiguous allocated ones. Like comb teeth and the space between them, the number of free segments increases with the number of occupied ones, and the spectrum becomes more fragmented. When the average size of free segments decreases, the fragmentation metric should increase as it is harder to fit a new request in smaller available spaces. When the index of last occupied slot decreases, the metric should decrease. The fragmentation metric should reflect these changes accordingly. Here we consider three kinds of fragmentation metric.
External fragmentation (EF) is defined as the ratio between the largest free memory block and the sum of all available blocks in computer science memory management [49,50]. As shown in Eq. (10), for network fragmentation, it is defined as the ratio of the size of the largest free segment over the sum of the sizes of all free segments. $\Gamma (e,k)$ denotes the set of free segments on spatial mode $k\in K$ of link $e\in E$ while $\left | \gamma _{ek} \right |$ is the size of the segment $\gamma _{ek} \in \Gamma (e,k)$. The metric increases as the size of free segment decreases. However, this metric only considers the influence of the largest segment and small segments are ignored. So, it is not accurate sometimes.
The Shannon entropy (SE) was originally applied in information theory to calculate the amount of information available in the message [51]. It accounts for the size of all segments and promotes larger ones by taking the inversion of segment sizes in a logarithm function, as shown in Eq. (11). Thus, the SE overcomes the limitations of the EF, as the EF accounts only for the size of the largest free segment.
The root of sum of squares (RSS) [52,53] in Eq. (12) promotes a network state where a small number of large free segments exists rather than many smaller ones. It is achieved by taking the square root of the sum of squares of free segment sizes, so larger free segments are more influential. The metric increases when the number of free segments increases, or the size of free segments decreases.
Finally, network fragmentation $F_{net}$ is calculated as the average over the fragmentation of all the network’s links.
In Eq. (13), network fragmentation is proportional to the index of the highest allocated slice $s_max$ in the network. The minimization of the highest allocated slice index results in leaving the remaining upper parts of the spectrum available for new requests, which tends to accommodate more requests.
5.2 Fragmentation-aware routing and spectrum allocation algorithm
The network of AoD nodes generally divides the processing of requests into two parts: routing resource allocation and node construction. Let us denote the bold letters as vectors or matrix in the algorithm. We assume that the SDM fiber here is single-mode multi-core fiber. The requests not only have the sources, destinations and bandwidth demands; they also have the delay tolerance. Some of them are latency-sensitive and can’t tolerate MEMS taking hundreds of milliseconds to reconfigure. The general algorithm can be divided into three steps. The first step is to allocate route and spectrum for the request to minimize network fragmentation. After that, many cores in the fiber links may be qualified for the given route and spectrum. The second step is to allocate cores along the route based on the flexibility of each core. The flexibility of the core is a good index to evaluate the reconfiguration time of the node backplanes on both sides of the link. After allocating route, spectrum and core, every node along the route needs to reconfigure its architecture. AoD synthesis technique is to decide whether to multiplex the existing connection or set up a new one.
Algorithm 1. Fragmentation-aware Routing and Spectrum Allocation.
Algorithm 1 presents the pseudocode of the Fragmentation-aware Routing and Spectrum Allocation algorithm (FRSA). It takes two input variables. $G(u,v,c,f)$ denotes network resources status, e.g., slot $f$ in core $c$ from node $u$ to $v$. $R(u^{'},v^{'},b,t)$ denotes the set of network requests, where $b$ is bandwidth demand and $t$ is delay tolerance. FRSA is to find the appropriate route and slot beginning index. The shortest $k$ paths are calculated as the set of paths $P$. For each path $p_{i}\in P$, the smallest slot index $f_{i}$ is founded constrained by spectrum continuity and contiguity. The spectrum (from $f_{i}$ to $f_{i}+b$) should be available in at least one core on each link along the path $p_{i}$. Since spatial lane change is supported by AoD, many cores may be qualified on the same link. If the path $p_{i}$ and the spectrum are chosen, network fragmentation $F_{net-i}$ can be calculated by referring to one of the fragmentation metrics above. We choose the path $p$ and the spectrum that can minimize network fragmentation. After FRSA, the route and spectrum index can be allocated. The time complexity of the algorithm for finding the k shortest paths in Algorithm 1: is $O(kn^{3})$, where $n$ is the number of nodes in the network, and the time complexity for computing the fragmentation metrics is $O(m|K|N)$, where $m$ is the number of edges in the network, $|K|$ is the number of spatial channels in spatial division multiplexing, and $N$ is the number of slots. Thus the overall time complexity of Algorithm 1: is $O(kn^{3}+m|K|N)$.
5.3 Flexibility-aware core allocation algorithm
Algorithm 2 presents the pseudocode of the Flexibility-aware Core Allocation Algorithm(FCA). It takes three more input variables than FRSA. The routing path p and the slot beginning index $f^{'}$ is from FRSA. $\boldsymbol{A}$ denotes the set of all switching node architecture status in the network, e.g., WSS ports connection status $M_{2}$ and $M_{3}$ in Eq. (8). In FRSA, many cores on the same link may be qualified for the spectrum allocation with the support of spatial lane change. Therefore, flexibility of each candidate core $c_{k} \in c_{ij}$ will be calculated based on the architecture status of the left and right node by referring to Eq. (8). The flexibility is a good index to evaluate the reconfiguration time of AoD architecture. The latency-sensitive requests should be offered with the most flexible cores to set up the connection as fast as possible. On the contrary, the requests that are not latency-sensitive don’t need the flexible cores. So, it would be better to free up the flexible resources for the latency-sensitive requests. After that, route, spectrum and core are allocated and network resources status $\boldsymbol{G}$ is updated. Algorithm 2: needs to calculate the core flexibility within each edge and the overall time complexity is $O(m|K|)$.
Algorithm 2. Flexibility-aware Core Allocation.
5.4 Synthesis algorithm for AoD
The AoD controller executes the algorithm to synthesis a suitable architecture according to the resources allocation results from FRSA and FCA. Based on the most flexible core of each link, the controller needs to decide whether to multiplex the existing connections or set up new connections inside the AoD nodes. The AoD Synthesis Algorithm (SA) is executed in two steps.
Figure 9 shows the first step of SA, which performs latency checking. The requests that are latency-sensitive should multiplex the existing connections or set up connections by SOA switches as much as possible. For the requests that are not latency-sensitive, setting up connections by MEMS is the better choice since it will add more existing connections for the future latency-sensitive requests.
The 2.1 step performs the AoD Synthesis for latency-sensitive requests, as shown in Fig. 10. The first part checks if there are existing connections available for multiplexing. It is the fastest way in the synthesis since AoD backplane doesn’t need to reconfigure its architecture. The second part checks if there are any ports connected to SOA switch available. If so, the connection will be set up by SOA switch by means of the fast reconfiguration. If there are no existing connection and port to SOA switch available, the third part will check the ports connected to MEMS. If there are ports to MEMS available and it can meet the latency requirement, connection will be set up, or the request will be denied. Note that every AoD node along the path needs to execute the synthesis algorithm. Any node that fails to set up the connection will result in the rejection of the request.
The 2.2 step performs the AoD Synthesis for requests that are not latency-sensitive, as shown in Fig. 11. The first part will check if there are ports connected to MEMS available. If so, the connection will be set up by MEMS, which will add more existing connection for the future latency-sensitive requests to multiplex. Since MEMS has many ports, most connections can be set up in this part. The existing connections and the ports connected to SOA switch are valuable resources, so it would be better to set up connections by MEMS for the requests that are not latency sensitive. If the ports to MEMS are used up, the existing connections and the ports to SOA switch will be considered in the following second and third part.
Finally, all AoD nodes along the route will be reconfigured and the requirement will be met.
6. Simulations and results
The simulations are conducted to evaluate the proposed fragmentation and flexibility aware routing and spectrum allocation algorithm and synthesis technique for AoD in SDM-EON. Two networks are used in the simulations: simulation network A topology with 23 nodes and 46 links in Fig. 12(a) and NSF network topology with 14 nodes and 21 links in Fig. 12(b). The average connectivity degree of all nodes in network A and NSF network are 2 and 1.5 respectively. We assume that each network link has one MCF for each direction. MCFs with 7 and 12 cores are considered in different simulations. We set the width of the spectrum slot to 12.5 GHz and the total spectrum resources per core to 4 THz (C band). Therefore, the number of slots per core is 320. We assume that the service time and the interarrival time of the connection requests follow an exponential distribution. Traffic volume data are from the Seattle Internet Exchange [54] and the source and destination nodes are randomly chosen from the simulation networks. Since the data are from IP network, we magnify the traffic volume to simulate requests in optical network and the flow magnification factor are used to evaluate network traffic volume. We assume that every request has the delay tolerance which follows the uniform distribution from 1 millisecond to 200 milliseconds. Requests that have the delay tolerance less than 100 millisecond are regarded as latency sensitive. Further, $k=5$ in the k-shortest path algorithm. Every experiment is conducted in different time frame of the data for 20 times to get the stable result.
Three kinds of switching nodes are evaluated in the experiments: traditional ROADM, AoD based on MEMS and AoD based on M&S. Both ROADM architectures with and without spatial lane change support are considered. Since ROADM cannot build connections on demand and core flexibility in FCA algorithm is not suitable here, core allocation will follow first-fit principle. Since WSS need to be pre-connected on backplane of AoD, the proportion of WSS ports connecting MEMS or SOA switches is also worth studying. In the AoD architecture, different proportions of WSS ports connected to SOA switches are considered. AoD supports both FRSA and FCA algorithms. If WSS ports connecting SOA switches are not sufficient and established connections are not available, latency-sensitive requests will be denied. Thanks to the dynamic synthesis characteristic, spatial lane change is supported by AoD and the port count need is controlled within a reasonable scope.
Single WSS port count of ROADM with and without spatial lane change supported are calculated by referring to Eq. (14) and (15) respectively. $a$ denotes the number of add/drop ports. $d$ denotes the degree of the node. $n$ denotes the number of cores in MCF, which corresponds to the number of WSS in a node. We assume single WSS port count of AoD are calculated by referring to Eq. (16), which means the WSS port count of AoD are between (14) and (15). $\alpha$ denotes the proportion of WSS ports added on $C_{WSS-nLC}$, whose range is between 0 and 100 percent. In our simulations, we assume $\alpha$ is thirty percent. We will check the performance of AoD in this low WSS port count situation. In the AoD based on M&S, we assume that $\beta$ percent of WSS ports are connected to SOA switches, $100-\beta$ percent of WSS ports are connected to MEMS and add/drop modules. We will evaluate the influence of different proportion of WSS ports connected to SOA switches in the AoD based on M&S. Total WSSs port count in a switching node (ROADM or AoD) is calculated by referring to Eq. (17).
The blocking probability in the network A and MCFs with 7 cores and 12 cores scenarios are shown in Fig. 13. The vertical axis indicates blocking probability of the requests, and the horizontal axis indicates flow magnification factor for the IP network. Fragmentation metric RSS is used in these simulations. Six kinds of node architectures are evaluated, which include AoD based on M&S of three SOA connection ratios, AoD based on MEMS, ROADM with and without SLC support. Note that ROADM with SLC support provides the fastest switching connections, since the backplane of ROADM is hard wired. However, this architecture requires a huge number of WSS ports, which makes it difficult to achieve.
These figures show that the proposed AoD based on M&S outperforms the AoD only based on MEMS thanks to the quick connection characteristic provided by SOA switches. In this case, latency-sensitive requests can set up connections immediately by SOA switches if there is no connection to multiplex. With the traffic volume increases, the blocking probability gap between ROADM without SLC support and AoD is growing. This is because spatial lane change supported by AoD provides great flexibility to get rid of spatial continuity when allocating routes and cores. By comparing Fig. 13 (a) and (b), the average performance gap between ROADM without SLC support and AoD based on M&S is 16.6% wider in a 7-core network than in a 12-core network, which means AoD based on M&S performs better in low-core network. This is because the flexibility provided by SLC is more important if there are not too many cores to choose from. With the increase of numbers of cores, flexibility of ROADM also increases, and its performance improves. Moreover, AoD based on M&S and ROADM with SLC support perform about the same. Therefore, AoD based on M&S is an effective solution that can replace ROADM to realize SLC in SDM-EONs.
Figure 14 (a) and (b) show the blocking probability in NSF network and MCF with 7 cores and 12 cores respectively. Note that the average connectivity degree in NSF network is relatively low and some key nodes that may cause link congestion are prone to appear in this network. Compared to the network A, the performance gap between AoD nodes and the ROADMs that don’t support SLC widens by 20.5% and 16.3% on the 7-core and 12-core, respectively, in the NSF network. As the average connectivity degree and the number of non-overlapping paths in NSF network decreases, flexibility provided by spatial lane change will become more important. The ability to switch the traffic to another spatial mode allows reducing the blocking probability. From the above four figures, it can be found that as the ratio of WSS ports connected to SOA switches increases, the blocking rate decreases. For such latency-sensitive services, the fast reconfiguration of SOA switches is particularly important. What’s more, by comparing the different proportions of WSS ports connected to SOA switches in AoD based on M&S, it can be concluded that AoD whose 50% to 80% WSS ports are connected to SOA switches has similar performance with ROADM that supports SLC. Since there are still difficulties in the extensive port expansion of SOA switches, it is recommended to control the proportion of WSS ports connected to SOA switches between 20% and 50% to get good results.
The proportion of latency-sensitive requests in blocked requests have been calculated in network A and NSF network in the case of 7-cores, as shown in Fig. 15. AoD based on MEMS, AoD based on M&S and ROADM with SLC support are evaluated. The results show that the use of AoD with a hybrid backplane structure can better reduce the proportion of latency-sensitive services in blocked services, that is, the SOA switching part can better serve latency-sensitive services in the backplane structure. In addition, with the continuous increase of traffic volume, the proportion of latency-sensitive services in the network using AoD nodes is also gradually decreasing. According to the synthesis algorithm for AoD, this is mainly because in the case of low traffic, there is no non-delay-sensitive service to help establish a reusable link first, so if the SOA switch is insufficient, the latency-sensitive requests will be blocked. In the case of high traffic, a reusable link can be established first through non-latency-sensitive requests for subsequent reuse of delay-sensitive requests.
We also evaluate the total number of WSS ports required by ROADM and AoD in different cases. WSS port counts in ROADM and AoD can be calculated by referring to Eq. (14)-(16). Total numbers of WSS ports required by a switching node can be calculated by referring to Eq. (17), and are shown in Fig. 16 (a) and (b) in 7-core and 12-core network respectively. As the nodal degree grows, the number of WSS ports required by ROADM supporting SLC grows rapidly compared with ROADM that does not support SLC. As the number of cores in MCFs increases, the required WSS ports increase more rapidly likewise. These results show that it is unrealistic to use ROADM for SLC in SDM-EONs. Alternative solutions need to find ways to reduce the number of required WSS ports, while still providing similar network performance.
Fig. 16. Total number of WSS ports required by (a) a ROADM in a 7-core network, (b) a ROADM in a 12-core network, (c) network A, (d) NSF network
The total number of WSS ports required by all nodes in the network A and NSF network are shown in Fig. 16 (c) and (d) respectively. As mentioned above, ROADMs that supports SLC require a lot of WSS ports, which is more pronounced in high connectivity degree network like network A. In contrast, the number of WSS ports required for AoD that also supports SLC is greatly reduced, which means it is more suitable to replace ROADM to implement SLC in SDM-EONs.
Three fragmentation metrics EF, SE and RSS are evaluated in network A and NSF network with ROADMs that don’t support SLC and M&S based AoDs, as shown in Fig. 17 (a) and (b). Among these metrics, RSS performs best. This is mainly because it takes account of a small number of large free segments and amplifies their influence in the indicator. It should also be noted that in this experiment, the influence of the fragmentation metrics in the algorithm is still smaller than the influence of the node architecture in the network, especially in low connectivity degree network like NSF network.
7. Conclusions
In this paper, we proposed a MEMS & SOA switch-based architecture for AoD backplane composition. Our proposed architecture can keep the WSS port count in check compared to traditional ROADM that supports spatial lane change in SDM-EON. It can also provide quick configuration to set up new connections for latency-sensitive requests compared to AoD based on MEMS. We discussed the flexibility of R&S based switching nodes and proposed the flexibility of optical fibers to distinguish between different connections in our proposed architecture. Based on these architectures, we proposed a fragmentation-aware routing and spectrum allocation algorithm and a flexibility-aware core allocation algorithm. Three fragmentation metrics were studied, and the results showed that the root of sum of squares metric outperformed the others. The hybrid AoD can reduce network fragmentation and blocking probability by applying the proposed methods compared with other nodes. Moreover, the AoD synthesis algorithm was studied for requests with different latency requirements. We simulated different proportions of WSS ports connected to SOA switches, and it turned out that the higher the proportion, the better the network performance. The AoD based on this hybrid backplane structure can improve the network performance by 32.8% compared to the AoD based on the traditional MEMS. The number of required WSS ports was evaluated for different switching nodes and results showed that the hybrid AoD is more practical than ROADM that supports spatial lane change.
In future work, it may be worthwhile to evaluate our proposed architecture and algorithms for different traffic models. Power consumption and equipment cost will also be discussed compared to traditional ROADM.
Funding
National Natural Science Foundation of China (61821001, 62125103, 61601054).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are available in Ref. [54].
References
1. P. S. Khodashenas, J. M. Rivas-Moscoso, D. Siracusa, F. Pederzolli, B. Shariati, D. Klonidis, E. Salvadori, and I. Tomkos, “Comparison of spectral and spatial super-channel allocation schemes for sdm networks,” J. Lightwave Technol. 34(11), 2710–2716 (2016). [CrossRef]
2. G. M. Saridis, D. Alexandropoulos, G. Zervas, and D. Simeonidou, “Survey and evaluation of space division multiplexing: From technologies to optical networks,” IEEE Commun. Surv. Tutorials 17(4), 2136–2156 (2015). [CrossRef]
3. M. Yang, Q. Wu, K. Guo, and Y. Zhang, “Evaluation of device cost, power consumption, and network performance in spatially and spectrally flexible sdm optical networks,” J. Lightwave Technol. 37(20), 5259–5272 (2019). [CrossRef]
4. P. Schindler, R. Schmogrow, S. Wolf, B. Bäuerle, B. Nebendahl, C. Koos, W. Freude, and J. Leuthold, “Full flex-grid asynchronous multiplexing demonstrated with nyquist pulse-shaping,” Opt. Express 22(9), 10923–10937 (2014). [CrossRef]
5. J. M. Rivas-Moscoso, B. Shariati, D. M. Marom, D. Klonidis, and I. Tomkos, “Comparison of cd (c) roadm architectures for space division multiplexed networks,” in Optical Fiber Communication Conference, (Optical Society of America, 2017), pp. Th2A–45.
6. M. Klinkowski, P. Lechowicz, and K. Walkowiak, “Survey of resource allocation schemes and algorithms in spectrally-spatially flexible optical networking,” Opt. Switch. Netw. 27, 58–78 (2018). [CrossRef]
7. M. Garrich, N. Amaya, G. S. Zervas, P. Giaccone, and D. Simeonidou, “Architecture on demand: Synthesis and scalability,” in 2012 16th International Conference on Optical Network Design and Modelling (ONDM), (IEEE, 2012), pp. 1–6.
8. Q. Cheng, M. Ding, A. Wonfor, J. Wei, R. V. Penty, and I. H. White, “The feasibility of building a 64× 64 port count soa-based optical switch,” in 2015 International Conference on Photonics in Switching (PS), (IEEE, 2015), pp. 199–201.
9. A. Wonfor, H. Wang, R. Penty, and I. White, “Large port count high-speed optical switch fabric for use within datacenters,” J. Opt. Commun. Netw. 3(8), A32–A39 (2011). [CrossRef]
10. S. Tanaka, S.-H. Jeong, S. Yamazaki, A. Uetake, S. Tomabechi, M. Ekawa, and K. Morito, “Monolithically integrated 8: 1 soa gate switch with large extinction ratio and wide input power dynamic range,” IEEE J. Quantum Electron. 45(9), 1155–1162 (2009). [CrossRef]
11. T. Kirihara, M. Ogawa, H. Inoue, H. Kodera, and K. Ishida, “Lossless and low-crosstalk characteristics in an inp-based 4/spl times/4 optical switch with integrated single-stage optical amplifiers,” IEEE Photonics Technol. Lett. 6(2), 218–221 (1994). [CrossRef]
12. R. Stabile, A. Albores-Mejia, and K. Williams, “Monolithic active-passive 16× 16 optoelectronic switch,” Opt. Lett. 37(22), 4666–4668 (2012). [CrossRef]
13. F.-J. Moreno-Muro, R. Rumipamba-Zambrano, P. Pavón-Marino, J. Perelló, J. M. Gené, and S. Spadaro, “Evaluation of core-continuity-constrained roadms for flex-grid/mcf optical networks,” J. Opt. Commun. Netw. 9(11), 1041–1050 (2017). [CrossRef]
14. A. C. Jatoba-Neto, D. A. Mello, C. E. Rothenberg, S. Ö. Arik, and J. M. Kahn, “Scaling sdm optical networks using full-spectrum spatial switching,” J. Opt. Commun. Netw. 10(12), 991–1004 (2018). [CrossRef]
15. N. K. Fontaine, T. Haramaty, R. Ryf, et al., “Heterogeneous space-division multiplexing and joint wavelength switching demonstration,” in Optical Fiber Communication Conference, (Optical Society of America, 2015), pp. Th5C–5.
16. M. D. Feuer, L. E. Nelson, K. Abedin, X. Zhou, T. Taunay, J. F. Fini, B. Zhu, R. Isaac, R. Harel, G. Cohen, and D. M. Marom, “Roadm system for space division multiplexing with spatial superchannels,” in Optical Fiber Communication Conference/National Fiber Optic Engineers Conference 2013, (Optica Publishing Group, 2013), p. PDP5B.8.
17. L. Nelson, M. Feuer, K. Abedin, et al., “Spatial superchannel routing in a two-span roadm system for space division multiplexing,” J. Lightwave Technol. 32(4), 783–789 (2014). [CrossRef]
18. T. Kuno, Y. Mori, S. Subramaniam, M. Jinno, and H. Hasegawa, “Design and evaluation of a reconfigurable optical add-drop multiplexer with flexible wave-band routing in sdm networks,” J. Opt. Commun. Netw. 14(4), 248–256 (2022). [CrossRef]
19. R. Rumipamba-Zambrano, F.-J. Moreno-Muro, J. Perelló, P. Pavón-Mari no, and S. Spadaro, “Space continuity constraint in dynamic flex-grid/sdm optical core networks: an evaluation with spatial and spectral super-channels,” Comput. Commun. 126, 38–49 (2018). [CrossRef]
20. R. Rumipamba-Zambrano, F.-J. Moreno-Muro, P. Pavón-Marino, J. Perelló, S. Spadaro, and J. Solé-Pareta, “Assessment of flex-grid/mcf optical networks with roadm limited core switching capability,” in 2017 International Conference on Optical Network Design and Modeling (ONDM), (IEEE, 2017), pp. 1–6.
21. Q. Cheng, A. Wonfor, J. Wei, R. Penty, and I. White, “Monolithic mzi-soa hybrid switch for low-power and low-penalty operation,” Opt. Lett. 39(6), 1449–1452 (2014). [CrossRef]
22. Q. Cheng, S. Rumley, M. Bahadori, and K. Bergman, “Photonic switching in high performance datacenters,” Opt. Express 26(12), 16022–16043 (2018). [CrossRef]
23. I. White, E. T. Aw, K. Williams, H. Wang, A. Wonfor, and R. Penty, “Scalable optical switches for computing applications,” J. Opt. Netw. 8(2), 215–224 (2009). [CrossRef]
24. Q. Cheng, A. Wonfor, J. Wei, R. V. Penty, and I. H. White, “Low-energy, high-performance lossless 8× 8 soa switch,” in Optical Fiber Communication Conference, (Optical Society of America, 2015), pp. Th4E–6.
25. Q. Cheng, R. Stabile, A. Rohit, A. Wonfor, R. V. Penty, I. H. White, and K. Williams, “Fast dynamic wavelength and path scheduling in a monolithic 8× 8 switch,” in OFC 2014, (IEEE, 2014), pp. 1–3.
26. Q. Cheng, R. Stabile, A. Rohit, A. Wonfor, R. Penty, I. White, and K. Williams, “First demonstration of automated control and assessment of a dynamically reconfigured monolithic 8× 8 wavelength-and-space switch,” J. Opt. Commun. Netw. 7(3), A388–A395 (2015). [CrossRef]
27. N. Amaya, G. Zervas, and D. Simeonidou, “Introducing node architecture flexibility for elastic optical networks,” J. Opt. Commun. Netw. 5(6), 593–608 (2013). [CrossRef]
28. N. Amaya, M. Irfan, G. Zervas, et al., “Fully-elastic multi-granular network with space/frequency/time switching using multi-core fibres and programmable optical nodes,” Opt. Express 21(7), 8865–8872 (2013). [CrossRef]
29. S. Yan, E. Hugues-Salas, V. Rancano, et al., “First demonstration of all-optical programmable sdm/tdm intra data centre and wdm inter-dcn communication,” in 2014 The European Conference on Optical Communication (ECOC), (IEEE, 2014), pp. 1–3.
30. A. Muhammad, G. Zervas, N. Amaya, D. Simeonidou, and R. Forchheimer, “Introducing flexible and synthetic optical networking: Planning and operation based on network function programmable roadms,” J. Opt. Commun. Netw. 6(7), 635–648 (2014). [CrossRef]
31. O. Gerstel, M. Jinno, A. Lord, and S. B. Yoo, “Elastic optical networking: A new dawn for the optical layer?” IEEE Commun. Mag. 50(2), s12–s20 (2012). [CrossRef]
32. F. S. Abkenar and A. G. Rahbar, “Study and analysis of routing and spectrum allocation (rsa) and routing, modulation and spectrum allocation (rmsa) algorithms in elastic optical networks (eons),” Opt. Switch. Netw. 23, 5–39 (2017). [CrossRef]
33. T. Takagi, H. Hasegawa, K.-i. Sato, Y. Sone, B. Kozicki, A. Hirano, and M. Jinno, “Dynamic routing and frequency slot assignment for elastic optical path networks that adopt distance adaptive modulation,” in 2011 Optical Fiber Communication Conference and Exposition and the National Fiber Optic Engineers Conference, (IEEE, 2011), pp. 1–3.
34. I. Stiakogiannakis, E. Palkopoulou, D. Klonidis, O. Gerstel, and I. Tomkos, “Dynamic cooperative spectrum sharing and defragmentation for elastic optical networks,” J. Opt. Commun. Netw. 6(3), 259–269 (2014). [CrossRef]
35. R. Wang and B. Mukherjee, “Provisioning in elastic optical networks with non-disruptive defragmentation,” J. Lightwave Technol. 31(15), 2491–2500 (2013). [CrossRef]
36. F. Cugini, F. Paolucci, G. Meloni, G. Berrettini, M. Secondini, F. Fresi, N. Sambo, L. Poti, and P. Castoldi, “Push-pull defragmentation without traffic disruption in flexible grid optical networks,” J. Lightwave Technol. 31(1), 125–133 (2012). [CrossRef]
37. Y. Yin, H. Zhang, M. Zhang, M. Xia, Z. Zhu, S. Dahlfort, and S. B. Yoo, “Spectral and spatial 2d fragmentation-aware routing and spectrum assignment algorithms in elastic optical networks,” J. Opt. Commun. Netw. 5(10), A100–A106 (2013). [CrossRef]
38. Y. Zhao and J. Zhang, “Crosstalk-aware cross-core virtual concatenation in spatial division multiplexing elastic optical networks,” Electron. Lett. 52(20), 1701–1703 (2016). [CrossRef]
39. B. C. Chatterjee and E. Oki, “Dispersion-adaptive first–last fit spectrum allocation scheme for elastic optical networks,” IEEE Commun. Lett. 20(4), 696–699 (2016). [CrossRef]
40. A. Muhammad, G. Zervas, and R. Forchheimer, “Resource allocation for space-division multiplexing: Optical white box versus optical black box networking,” J. Lightwave Technol. 33(23), 4928–4941 (2015). [CrossRef]
41. M. Džanko, M. Furdek, G. Zervas, and D. Simeonidou, “Evaluating availability of optical networks based on self-healing network function programmable roadms,” J. Opt. Commun. Netw. 6(11), 974–987 (2014). [CrossRef]
42. S. Fujii, Y. Hirota, H. Tode, and T. Watanabe, “Dynamic spectrum and core allocation reducing costs of architecture on demand nodes,” in 2014 The European Conference on Optical Communication (ECOC), (IEEE, 2014), pp. 1–3.
43. S. Fujii, Y. Hirota, T. Watanabe, and H. Tode, “Dynamic spectrum and core allocation with spectrum region reducing costs of building modules in aod nodes,” in 2014 16th International Telecommunications Network Strategy and Planning Symposium (Networks), (IEEE, 2014), pp. 1–6.
44. S. Fujii, Y. Hirota, H. Tode, and T. Watanabe, “On-demand routing and spectrum allocation for energy-efficient aod nodes in sdm-eons,” J. Opt. Commun. Netw. 9(11), 960–973 (2017). [CrossRef]
45. M. Garrich, N. Amaya, G. S. Zervas, J. R. Oliveira, P. Giaccone, A. Bianco, D. Simeonidou, and J. C. R. Oliveira, “Architecture on demand design for high-capacity optical sdm/tdm/fdm switching,” J. Opt. Commun. Netw. 7(1), 21–35 (2015). [CrossRef]
46. V. Aksyuk, S. Arney, N. Basavanhally, et al., “238 x 238 micromechanical optical cross connect,” IEEE Photonics Technol. Lett. 15(4), 587–589 (2003). [CrossRef]
47. C. E. Shannon, “A mathematical theory of communication,” The Bell Syst. Tech. J. 27(3), 379–423 (1948). [CrossRef]
48. S. Ding, S. Yin, Z. Zhang, and S. Huang, “Evaluation of the flexibility of switching node architectures for spaced division multiplexed elastic optical network,” in Optical Fiber Communication Conference, (Optical Society of America, 2020), pp. M2G–2.
49. B. C. Chatterjee, S. Ba, and E. Oki, “Fragmentation problems and management approaches in elastic optical networks: A survey,” IEEE Commun. Surv. Tutorials 20(1), 183–210 (2017). [CrossRef]
50. D. Amar, E. Le Rouzic, N. Brochier, J.-L. Auge, C. Lepers, and N. Perrot, “Spectrum fragmentation issue in flexible optical networks: analysis and good practices,” Photonic Netw. Commun. 29(3), 230–243 (2015). [CrossRef]
51. P. Wright, M. C. Parker, and A. Lord, “Minimum-and maximum-entropy routing and spectrum assignment for flexgrid elastic optical networking,” J. Opt. Commun. Netw. 7(1), A66–A72 (2015). [CrossRef]
52. Z. Ye, A. N. Patel, P. N. Ji, and C. Qiao, “Root mean square (rms) factor for assessing spectral fragmentation in flexible grid optical networks,” in 2014 OptoElectronics and Communication Conference and Australian Conference on Optical Fibre Technology, (IEEE, 2014), pp. 357–358.
53. P. Lechowicz, A. Knapińska, and R. Goścień, “Fragmentation-aware traffic grooming with lane changes in spectrally–spatially flexible optical networks,” Electronics 10(12), 1502 (2021). [CrossRef]
54. “The SIX, The Seattle Internet Exchange,” https://www.seattleix.net/.
