Design of Optical Tunnel Switching Networks for Big Data Applications

: In this paper, we proposed large-scale optical tunnel switching networks based on the Torus topology network with WSS (Wavelength Selective Switch) for future big data applications. All nodes of the large-scale optical tunnel switching networks use WSS switch modules, and the communications between nodes use multiple λ s (wavelengths), where a tunnel is established with a wavelength which can be reused. The widely used MEMS (Micro-Electro-Mechanical Systems) and LCoS (Liquid Crystal on Silicon) technologies are all millisecond-level switching speeds, so the frame size of the optical frame switch is very large, and this will reduce switching performance. Therefore, they are only suitable for optical tunnel switching networks design, but are not suitable for optical frame switch design. This multi-plane Torus topology network architecture not only increases network throughput, but also has fault tolerance to increase network reliability. When the tra ﬃ c is changed, the number of tunnels between nodes can be scheduled in time to balance the load tra ﬃ c and avoid tra ﬃ c loss. Therefore, it can not only schedule the number of tunnels in time to balance the load tra ﬃ c, in order to avoid tra ﬃ c loss, but also because the channel is ﬁxedly established, this does not generate any bu ﬀ er delay, and this because of the transmission using optical transmission unlimited speed, so it is a good choice for future big data applications that require high speed, high bandwidth and low latency.


Introduction
Currently, the companies focus on seven Artificial Intelligence (AI) applications, including chat bots, process advancement and fraud detection, face recognition, market segmentation, sentiment analysis and call center assistants. The Xiaomi AI speaker has six radio microphones for all-round radio and listening to voice commands. In addition to voice-activated music or listening to audio books, you can also ask for exchange rates, weather, etc., or ask it to calculate math problems. These applications require high speed, high bandwidth, low latency network systems.
Due to the success of streaming video and music service modes, as well as the increase in network bandwidth, streaming game services will be launched in the future. In the past, manufacturers failed to promote streaming games, partly because of the "delay" problem. So the spread of data centers and the utility of the new generation of mobile network 5G will help to meet these challenges.
Unmanned aerial vehicles (UAVs) are very mature in product development, but they have not been widely applied so far, mainly because of the lack of a systematic management platform, resulting in the inability to effectively supervise a large number of drones.
The accuracy of traffic jam forecasting on specific road sections has now reached 90%, which allows driving to effectively avoid traffic jams, but sometimes it may cause alternative road traffic jams, different times. Therefore, a scheduling algorithm must be used to control the usage of every node and simulate its performance, as shown in [12]. Its performance is related to the frame size. Because of the smaller frame with the better performance, therefore, high-speed switching elements, such as SOA switching elements, are required. The optical tunnel switching network is only used for wavelength tunnel scheduling, and does not require high-speed switching elements. Therefore, the performance of the optical tunnel switching network is necessary that the average throughput of traffic reaches to 100% at least. However, the performance of the optical frame switching network ensures that average throughput is less than 100%, because the aggregation algorithm is adopted. This is because, when much traffic is sent to the same node, some traffic will be dropped. In the optical tunnel switching network, when a lot of traffic gets sent to the same node, we can schedule the wavelength tunnel to avoid traffic loss. The advantages of the optical tunnel switching network is that it does not require high-speed switching elements and flexible channel scheduling to avoid data loss when the traffic volume is changed.
The brief descriptions of other sections are presented as follows. In Section 2, we describe the operations of a multi-plane Torus topology optical tunnel switching network. In Section 3, a two-plane Torus topology optical tunnel switching network with 25 nodes is presented, and its performance is discussed. The three-plane Torus topology optical tunnel switching network with 49 nodes is presented, and its performance is also discussed in Section 4. In Section 5, performance evaluation, traffic management and applications of optical tunnel switching networks are discussed. Finally, some conclusions are made in Section 6.

Torus Optical Tunnel Switch Network
The structure of the wavelength selective switch based on LCoS technology is shown in Figure 1 (Provided by Accton Technology Corp.). The SDN technology is used to allocate a wavelength of the input port to the specific output port. For example, λ 1 , λ 7 , λ 11 and λ 20 of input port 1 are assigned to transfer to output port 3, and so on.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 16 and simulate its performance, as shown in [13]. Its performance is related to the frame size. Because of the smaller frame with the better performance, therefore, high-speed switching elements, such as SOA switching elements, are required. The optical tunnel switching network is only used for wavelength tunnel scheduling, and does not require high-speed switching elements. Therefore, the performance of the optical tunnel switching network is necessary that the average throughput of traffic reaches to 100% at least. However, the performance of the optical frame switching network ensures that average throughput is less than 100%, because the aggregation algorithm is adopted. This is because, when much traffic is sent to the same node, some traffic will be dropped. In the optical tunnel switching network, when a lot of traffic gets sent to the same node, we can schedule the wavelength tunnel to avoid traffic loss. The advantages of the optical tunnel switching network is that it does not require high-speed switching elements and flexible channel scheduling to avoid data loss when the traffic volume is changed. The brief descriptions of other sections are presented as follows. In Section 2, we describe the operations of a multi-plane Torus topology optical tunnel switching network. In Section 3, a twoplane Torus topology optical tunnel switching network with 25 nodes is presented, and its performance is discussed. The three-plane Torus topology optical tunnel switching network with 49 nodes is presented, and its performance is also discussed in Section 4. In Section 5, performance evaluation, traffic management and applications of optical tunnel switching networks are discussed. Finally, some conclusions are made in Section 6. The structure of the wavelength selective switch based on LCoS technology is shown in Figure  1 (Provided by Accton Technology Corp.). The SDN technology is used to allocate a wavelength of the input port to the specific output port. For example, λ1, λ7, λ11 and λ20 of input port 1 are assigned to transfer to output port 3, and so on. Figure 2 shows a multi-plane Torus topology optical switching network. The X-axis has X nodes and the Y-axis has Y nodes, so there is a total of X × Y nodes. In order to improve the throughput, Z planes are used.

Torus Optical Tunnel Switch Network
X-axis has X nodes that need X × (X-1) different wavelengths (λ) if we need to establish full match tunnels. Similarly, Y-axis needs Y × (Y-1) different wavelengths if we need to establish full match tunnels. Therefore, the total number of wavelengths named W is defined:  Figure 2 shows a multi-plane Torus topology optical switching network. The X-axis has X nodes and the Y-axis has Y nodes, so there is a total of X × Y nodes. In order to improve the throughput, Z planes are used.
X-axis has X nodes that need X × (X-1) different wavelengths (λ) if we need to establish full match tunnels. Similarly, Y-axis needs Y × (Y-1) different wavelengths if we need to establish full match tunnels. Therefore, the total number of wavelengths named W is defined: After calculation, we can obtain Therefore, if W = 48, X = 5.42 takes its integer, and the maximum number of nodes is 25 (5 × 5). Similarly, if W = 96, X = 7.3 takes its integer, and the maximum number of nodes is 49 (7 × 7). If we use these wavelengths, we can create direct tunnels. For example, if X = Y = 5, the direct tunnels of each node to the other nodes are only 8 (4 + 4), but the tunnels of each node to the other nodes need 24, so the other 16 tunnels must be transferred through the relay nodes. The total number of tunnels needs 600 (24 × 25). Similarly, if X = Y = 7, the direct tunnels of each node to other nodes is only 12 (6 + 6), but the tunnels of each node to the other nodes need 48, so the other 36 tunnels must be transferred through the relay nodes. The total number of tunnels requires 2352 (48 × 49). These relay tunnels occupy the bandwidth of the direct tunnel, so multiple planes must be used to increase the network throughput. How many planes are needed depend on the number of nodes, and the required throughput that will be discussed in Sections 3 and 4.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 4 of 16 After calculation, we can obtain Therefore, if W = 48, X = 5.42 takes its integer, and the maximum number of nodes is 25 (5 × 5). Similarly, if W = 96, X = 7.3 takes its integer, and the maximum number of nodes is 49 (7 × 7). If we use these wavelengths, we can create direct tunnels. For example, if X = Y = 5, the direct tunnels of each node to the other nodes are only 8 (4 + 4), but the tunnels of each node to the other nodes need 24, so the other 16 tunnels must be transferred through the relay nodes. The total number of tunnels needs 600 (24 × 25). Similarly, if X = Y = 7, the direct tunnels of each node to other nodes is only 12 (6 + 6), but the tunnels of each node to the other nodes need 48, so the other 36 tunnels must be transferred through the relay nodes. The total number of tunnels requires 2352 (48 × 49). These relay tunnels occupy the bandwidth of the direct tunnel, so multiple planes must be used to increase the network throughput. How many planes are needed depend on the number of nodes, and the required throughput that will be discussed in Section 3 and Section 4.  Each module is composed of N servers, Z MUXs (Multiplexer), Z CDMUXs (Cyclic De-Multiplexer), an SDN switch and Z nodes with a node in each plane as shown in Figure 3, where each node is a 5 × 5 WSS. This multi-plane architecture not only increases the throughput of the network, but also increases the stability of the network, and has fault tolerance capability. A module can be Node (x,y) Node (x,y)

SDN Controller
(X+Y-2) λs (X+Y-2) λs   Each module is composed of N servers, Z MUXs (Multiplexer), Z CDMUXs (Cyclic De-Multiplexer), an SDN switch and Z nodes with a node in each plane as shown in Figure 3, where each node is a 5 × 5 WSS. This multi-plane architecture not only increases the throughput of the network, but also increases the stability of the network, and has fault tolerance capability. A module can be connected to N servers. These servers collect data and sort through an SDN switch to aggregate the data to the same module and then transfer it to the multiplexer. The (X + Y − 2) wavelengths are grouped into a DWDM signal and sent to a 5 × 5 wavelength selective switch (5 × 5 WSS). Each node has a 5 × 5 WSS to set the different wavelengths to send to different nodes to establish a tunnel. Different wavelengths in the DWDM signal receiving from different nodes are decomposed into the SDN switch via a CDMUX, and then sent to their corresponding servers. Other planes have the same function to increase network throughput, and when a plane fails, traffic can be routed to another plane to increase network reliability and stability.  Each module is composed of N servers, Z MUXs (Multiplexer), Z CDMUXs (Cyclic De-Multiplexer), an SDN switch and Z nodes with a node in each plane as shown in Figure 3, where each node is a 5 × 5 WSS. This multi-plane architecture not only increases the throughput of the network, but also increases the stability of the network, and has fault tolerance capability. A module can be   Figure 4 shows a two-plane Torus topology optical switching network with 25 nodes. Each module is composed of 16 servers, 2 MUXs, 2 CDMUXs, an SDN switch and 2 nodes, with a node in each plane, as shown in Figure 5, where each node is a 5 × 5 WSS.

Torus Optical Tunnel Switch Network with 25 Nodes
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 16 connected to N servers. These servers collect data and sort through an SDN switch to aggregate the data to the same module and then transfer it to the multiplexer. The (X + Y − 2) wavelengths are grouped into a DWDM signal and sent to a 5 × 5 wavelength selective switch (5 × 5 WSS). Each node has a 5 × 5 WSS to set the different wavelengths to send to different nodes to establish a tunnel. Different wavelengths in the DWDM signal receiving from different nodes are decomposed into the SDN switch via a CDMUX, and then sent to their corresponding servers. Other planes have the same function to increase network throughput, and when a plane fails, traffic can be routed to another plane to increase network reliability and stability.   Figure 4 shows a two-plane Torus topology optical switching network with 25 nodes. Each module is composed of 16 servers, 2 MUXs, 2 CDMUXs, an SDN switch and 2 nodes, with a node in each plane, as shown in Figure 5, where each node is a 5 × 5 WSS. connected to N servers. These servers collect data and sort through an SDN switch to aggregate the data to the same module and then transfer it to the multiplexer. The (X + Y − 2) wavelengths are grouped into a DWDM signal and sent to a 5 × 5 wavelength selective switch (5 × 5 WSS). Each node has a 5 × 5 WSS to set the different wavelengths to send to different nodes to establish a tunnel. Different wavelengths in the DWDM signal receiving from different nodes are decomposed into the SDN switch via a CDMUX, and then sent to their corresponding servers. Other planes have the same function to increase network throughput, and when a plane fails, traffic can be routed to another plane to increase network reliability and stability.   Figure 4 shows a two-plane Torus topology optical switching network with 25 nodes. Each module is composed of 16 servers, 2 MUXs, 2 CDMUXs, an SDN switch and 2 nodes, with a node in each plane, as shown in Figure 5, where each node is a 5 × 5 WSS. Node (x,y)

Torus Optical Tunnel Switch Network with 25 Nodes
Node (x,y)  The X-axis and Y-axis wavelength assignments of each node are shown in Table 1. A total of 40 wavelengths are shared and reused with different planes. Figure 6 shows the first plane tunnel creation of a two-plane Torus topology optical tunnel switching network with 25 nodes. Suppose node N (1, 5) wants to build a tunnel to node N (4, 5), because the two nodes are in the same X-axis loop, and therefore λ 3 (refer to the blue word in Table 1) can establish a direct tunnel. But node N (1, 5) needs to build a tunnel to node N (5, 3), because they are not in the same X-axis loop, therefore it is necessary to establish a relay tunnel; that is, first establish a direct tunnel to node N (5, 5) with λ 4 (refer to the red word in Table 1), and transfer to the SDN switch and then transfer to node N (5, 3) with λ 38 (refer to the red word in Table 1). N (1, 5) also needs to build a direct tunnel to node N (5, 5), and therefore, the tunnel from the node N (1, 5) to the node N (5, 5) is shared with the previous relay tunnel. That is to say, the traffic volume of the two tunnels will be integrated, so the average traffic volume cannot exceed 50%. Of course, the SDN switch at node N (5, 5) must separate the two nodes' traffic according to different IP addresses into the correct servers of the two nodes. That is, the traffic to the node N (5, 5) is left, and the traffic to the node N (5, 3) is transferred out via the SDN switch. Table 1. X-axis and Y-axis wavelength assignment of each node.
However, if each direct tunnel is responsible for transferring a relay tunnel node, there are still 8 nodes that cannot receive traffic from node N (1, 5), so if each direct tunnel is responsible for transferring two relay tunnel nodes, then node N (1, 5) can transmit traffic to any other 24 nodes. However, sharing a direct tunnel between the three nodes reduces the average throughput of traffic to 33% approximately. Therefore, it is necessary to add a plane Torus topology optical tunnel switching network to increase the throughput of traffic. If two planes of the Torus topology optical switching network are used, each plane is responsible for one relay tunnel node, which is equal to two direct tunnels to three nodes, so the throughput of traffic can reach an average of about 66%. Moreover, one module shares two nodes, and one module has only 16 servers. If these servers are evenly distributed to 24 modules, the average throughput of traffic is only about 66%. Therefore, the two-plane Torus topology optical switching network is sufficient to provide 100% of the server's average throughput of traffic. Of course, if the traffic is not evenly distributed, the distribution of the tunnels must be adjusted. However, if each direct tunnel is responsible for transferring a relay tunnel node, there are still 8 nodes that cannot receive traffic from node N (1, 5), so if each direct tunnel is responsible for transferring two relay tunnel nodes, then node N (1, 5) can transmit traffic to any other 24 nodes. However, sharing a direct tunnel between the three nodes reduces the average throughput of traffic to 33% approximately. Therefore, it is necessary to add a plane Torus topology optical tunnel switching network to increase the throughput of traffic. If two planes of the Torus topology optical switching network are used, each plane is responsible for one relay tunnel node, which is equal to two direct tunnels to three nodes, so the throughput of traffic can reach an average of about 66%. Moreover, one module shares two nodes, and one module has only 16 servers. If these servers are evenly distributed to 24 modules, the average throughput of traffic is only about 66%. Therefore, the two-plane Torus topology optical switching network is sufficient to provide 100% of the server's average throughput of traffic. Of course, if the traffic is not evenly distributed, the distribution of the tunnels must be adjusted.

Torus Optical Tunnel Switch Network with 49 Nodes
To increase the capacity of the network, Figure 7 shows a three-plane Torus topology optical tunnel switching network with 49 nodes. Figure 8 shows the module of the three-plane Torus topology optical tunnel switching network with 49 nodes. Each module is composed of 32 servers, 3 MUXs, 3 CDMUXs, an SDN switch and 3 nodes, with a node in each plane, as shown in Figure 8, where each node is a 5 × 5 WSS. Each module consists of three planes corresponding nodes, so 12 λs can establish 6 direct tunnels for the X-axis and Y-axis, respectively. In addition, there are 36 relay tunnel nodes for each module to be established. Similarly, if each plane is responsible for transferring a relay tunnel node, the three direct tunnels are responsible for the traffic of 4 modules, so the average throughput is 75% (3/4). Each module has 36 servers that average to send to 48 modules with an average throughput of only 75% (36/48). Therefore, the three-plane Torus topology optical switching network is sufficient to provide 100% of the average traffic of the server.

Torus Optical Tunnel Switch Network with 49 Nodes
To increase the capacity of the network, Figure 7 shows a three-plane Torus topology optical tunnel switching network with 49 nodes. Figure 8 shows the module of the three-plane Torus topology optical tunnel switching network with 49 nodes. Each module is composed of 32 servers, 3 MUXs, 3 CDMUXs, an SDN switch and 3 nodes, with a node in each plane, as shown in Figure 8, where each node is a 5 × 5 WSS. Each module consists of three planes corresponding nodes, so 12 λs can establish 6 direct tunnels for the X-axis and Y-axis, respectively. In addition, there are 36 relay tunnel nodes for each module to be established. Similarly, if each plane is responsible for transferring a relay tunnel node, the three direct tunnels are responsible for the traffic of 4 modules, so the average throughput is 75% (3/4). Each module has 36 servers that average to send to 48 modules with an average throughput of only 75% (36/48). Therefore, the three-plane Torus topology optical switching network is sufficient to provide 100% of the average traffic of the server.       Figure 9 shows tunnel creation of a three-plane Torus topology optical tunnel switching network with 49 nodes for the first plane. Let X = Y = 7. Therefore, node N (1, 7) and node N (6, 7) can establish a direct tunnel, and node N (1, 7) and node N (7, 2) need to establish a relay tunnel through node N (7, 7). Similarly, node N (1, 3) and node N (7, 3) can establish a direct tunnel, and node N (1, 3) and node N (6, 1) need to establish a relay tunnel through node N (6, 3). Other nodes can also establish direct tunnels and relay tunnels in sequence.    Figure 9 shows tunnel creation of a three-plane Torus topology optical tunnel switching network with 49 nodes for the first plane. Let X = Y = 7. Therefore, node N (1, 7) and node N (6, 7) can establish a direct tunnel, and node N (1, 7) and node N (7, 2) need to establish a relay tunnel through node N (7, 7). Similarly, node N (1, 3) and node N (7, 3) can establish a direct tunnel, and node N (1, 3) and node N (6, 1) need to establish a relay tunnel through node N (6, 3). Other nodes can also establish direct tunnels and relay tunnels in sequence.    Figure 9 shows tunnel creation of a three-plane Torus topology optical tunnel switching network with 49 nodes for the first plane. Let X = Y = 7. Therefore, node N (1, 7) and node N (6, 7) can establish a direct tunnel, and node N (1, 7) and node N (7, 2) need to establish a relay tunnel through node N (7, 7). Similarly, node N (1, 3) and node N (7, 3) can establish a direct tunnel, and node N (1, 3) and node N (6, 1) need to establish a relay tunnel through node N (6, 3). Other nodes can also establish direct tunnels and relay tunnels in sequence.  The three-plane Torus topology optical switching network can greatly increase the capacity of the network. This architecture utilizes 84 λs to provide 1764 servers and establish 2352 tunnels. The architecture in [11] utilizes 40 λs to provide 400 servers and establish 600 tunnels. So, while the number of wavelengths has only doubled, the number of servers that it provides can be increased by more than four times. In the future, if the number of wavelengths can be increased, the number of servers provided can be increased.

Performance Evaluation
For a detailed description of the switch performance, Figure 10 shows the tunnel assignment of a two-plane optical tunnel switching network with 25 nodes. Each node of the 5 × 5 Torus network has 8 direct tunnels and 16 relay tunnels to transfer data to other nodes. According to the uniform distribution of traffic, on average, a direct tunnel allows two relay tunnels (16/8) to share bandwidth, such as node N (5, 5), node N (5, 4) and node N (5, 3) share a direct tunnel from node N (1, 5) to node N (5, 5), as well as node N (1, 4), node N (2, 4) and node N (4, 4) share another direct tunnel from node N (1, 5) to node N (1, 4). If a plane uses only a relay tunnel to share the bandwidth of the direct tunnel, the network needs dual planes. Therefore, we can connect from node N (1, 5) to all nodes on the network at the same time. These three nodes will share the bandwidth of the direct tunnel, and each node will use an average of two-thirds of the bandwidth of the two planes. One SDN switch per node can use 16 wavelength bandwidths for 16 servers (2 × 8). If the 16 wavelength bandwidths of the 16 servers are transmitted to the other 24 node modules on average, the average traffic is also 2/3 of the wavelength bandwidth (16/24).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 9 of 16 The three-plane Torus topology optical switching network can greatly increase the capacity of the network. This architecture utilizes 84 λs to provide 1764 servers and establish 2352 tunnels. The architecture in [11] utilizes 40 λs to provide 400 servers and establish 600 tunnels. So, while the number of wavelengths has only doubled, the number of servers that it provides can be increased by more than four times. In the future, if the number of wavelengths can be increased, the number of servers provided can be increased. Figure 10. Tunnel assignment of a two-plane optical tunnel switching network with 25 nodes.

Performance Evaluation
For a detailed description of the switch performance, Figure 10 shows the tunnel assignment of a two-plane optical tunnel switching network with 25 nodes. Each node of the 5 × 5 Torus network has 8 direct tunnels and 16 relay tunnels to transfer data to other nodes. According to the uniform distribution of traffic, on average, a direct tunnel allows two relay tunnels (16/8) to share bandwidth, such as node N (5, 5), node N (5, 4) and node N (5, 3) share a direct tunnel from node N (1, 5) to node N (5, 5), as well as node N (1, 4), node N (2, 4) and node N (4, 4) share another direct tunnel from node N (1, 5) to node N (1, 4). If a plane uses only a relay tunnel to share the bandwidth of the direct tunnel, the network needs dual planes. Therefore, we can connect from node N (1, 5) to all nodes on the network at the same time. These three nodes will share the bandwidth of the direct tunnel, and each node will use an average of two-thirds of the bandwidth of the two planes. One SDN switch per node can use 16 wavelength bandwidths for 16 servers (2 × 8). If the 16 wavelength bandwidths of the 16 servers are transmitted to the other 24 node modules on average, the average traffic is also 2/3 of the wavelength bandwidth (16/24). Figure 11. Tunnel assignment of a three-plane optical tunnel switching network with 49 nodes.
Using the X-axis Direct Tunnel (1,5) (1,1) (1,7) (7,7) (1,1) Using the X-axis Direct Tunnel Using the Y-axis Direct Tunnel Figure 10. Tunnel assignment of a two-plane optical tunnel switching network with 25 nodes. Figure 11 shows the tunnel assignment of a three-plane optical tunnel switching network with 49 nodes. Each node of the 7 × 7 Torus network has 12 direct tunnels and 36 relay tunnels to transfer data to other nodes. According to the uniform distribution of traffic, on average, a direct tunnel allows three relay tunnels (36/12) to share bandwidth, such that node N (7, 7), node N (7, 6), node N (7, 5) and node N (7, 4) share a direct tunnel, and the tunnel runs from node N (1, 7) to node N (7, 7). If one plane uses only a relay tunnel to share the bandwidth of the direct tunnel, the network requires three planes. Therefore, we can connect from node N (1, 7) to all nodes on the network at the same time. The four nodes will share the bandwidth of the direct tunnel, and each node uses an average of 3/4 wavelength bandwidth across the three planes. One SDN switch per node can use 36 wavelength bandwidth for 36 servers (3 × 12). If these 36 wavelength bandwidths for 36 servers are transmitted to the other 48 node modules on average, the average traffic is also 3/4 of the wavelength bandwidth (36/48).
Appl. Sci. 2020, 10, x FOR PEER REVIEW 9 of 16 The three-plane Torus topology optical switching network can greatly increase the capacity of the network. This architecture utilizes 84 λs to provide 1764 servers and establish 2352 tunnels. The architecture in [11] utilizes 40 λs to provide 400 servers and establish 600 tunnels. So, while the number of wavelengths has only doubled, the number of servers that it provides can be increased by more than four times. In the future, if the number of wavelengths can be increased, the number of servers provided can be increased. Figure 10. Tunnel assignment of a two-plane optical tunnel switching network with 25 nodes.

Performance Evaluation
For a detailed description of the switch performance, Figure 10 shows the tunnel assignment of a two-plane optical tunnel switching network with 25 nodes. Each node of the 5 × 5 Torus network has 8 direct tunnels and 16 relay tunnels to transfer data to other nodes. According to the uniform distribution of traffic, on average, a direct tunnel allows two relay tunnels (16/8) to share bandwidth, such as node N (5, 5), node N (5, 4) and node N (5, 3) share a direct tunnel from node N (1, 5) to node N (5, 5), as well as node N (1,4), node N (2, 4) and node N (4, 4) share another direct tunnel from node N (1, 5) to node N (1, 4). If a plane uses only a relay tunnel to share the bandwidth of the direct tunnel, the network needs dual planes. Therefore, we can connect from node N (1, 5) to all nodes on the network at the same time. These three nodes will share the bandwidth of the direct tunnel, and each node will use an average of two-thirds of the bandwidth of the two planes. One SDN switch per node can use 16 wavelength bandwidths for 16 servers (2 × 8). If the 16 wavelength bandwidths of the 16 servers are transmitted to the other 24 node modules on average, the average traffic is also 2/3 of the wavelength bandwidth (16/24). Figure 11. Tunnel assignment of a three-plane optical tunnel switching network with 49 nodes.
To investigate the performance of optical tunnel switching networks, we capture traffic from the Internet core network of Chunghwa Telecom Co., Ltd. in Taiwan (a major provider of telecommunications services). According to the data captured from the core network of a traditional Ethernet switch, the traffic of Ethernet packets is shown in Figure 12, where the X-axis represents various packet sizes. The number of packets of various packet sizes is shown at the bottom of Figure 12, the total number of bytes (TN) of various packet sizes is shown in the middle of Figure 12, and the cumulative total number of bytes (CTN) of various packet sizes is shown at the top of Figure 12. In other words, if the number of packets is equal to Ni for various packet sizes, where i is denoted the packet size. For various packet sizes, the characteristics of TN and CTN are as follows: where j is the maximum packet size, and the maximum j for Ethernet packets is 1538 bytes.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 10 of 16 Figure 11 shows the tunnel assignment of a three-plane optical tunnel switching network with 49 nodes. Each node of the 7 × 7 Torus network has 12 direct tunnels and 36 relay tunnels to transfer data to other nodes. According to the uniform distribution of traffic, on average, a direct tunnel allows three relay tunnels (36/12) to share bandwidth, such that node N (7,7), node N (7, 6), node N (7,5) and node N (7, 4) share a direct tunnel, and the tunnel runs from node N (1, 7) to node N (7, 7). If one plane uses only a relay tunnel to share the bandwidth of the direct tunnel, the network requires three planes. Therefore, we can connect from node N (1, 7) to all nodes on the network at the same time.
The four nodes will share the bandwidth of the direct tunnel, and each node uses an average of 3/4 wavelength bandwidth across the three planes. One SDN switch per node can use 36 wavelength bandwidth for 36 servers (3 × 12). If these 36 wavelength bandwidths for 36 servers are transmitted to the other 48 node modules on average, the average traffic is also 3/4 of the wavelength bandwidth (36/48). To investigate the performance of optical tunnel switching networks, we capture traffic from the Internet core network of Chunghwa Telecom Co., Ltd. in Taiwan (a major provider of telecommunications services). According to the data captured from the core network of a traditional Ethernet switch, the traffic of Ethernet packets is shown in Figure 12, where the X-axis represents various packet sizes. The number of packets of various packet sizes is shown at the bottom of Figure  12, the total number of bytes (TN) of various packet sizes is shown in the middle of Figure 12, and the cumulative total number of bytes (CTN) of various packet sizes is shown at the top of Figure 12. In other words, if the number of packets is equal to Ni for various packet sizes, where i is denoted the packet size. For various packet sizes, the characteristics of TN and CTN are as follows: TN = × and GTN = ∑ ( × ) =1 (4) where j is the maximum packet size, and the maximum j for Ethernet packets is 1538 bytes.   Figure 13 shows the traffic analysis of Ethernet packets. The top of Figure 13 shows the cumulative percentage of packets of various packet sizes, while the bottom shows the cumulative percentage of bytes of various packet sizes. We can see that when the packet size is less than 1300 bytes, the cumulative percentage of bytes is only about 20%, but the cumulative percentage of packets is as high as 68%. The distribution of Ethernet packets of various packet sizes is shown in Figure 14. For example, a packet size group from 100 to 179 bytes is 50.35%, while a packet size group from 1300 to 1538 bytes is 33.21%. This 33.21% of packets has more traffic because these packets are large. The distribution of the traffic model is used to simulate the optical tunnel switching networks. We simulated the performance of the optical tunnel switching networks obtained from a computer. According to the Ethernet data packet distribution shown in Figure 14, the number of data packets transmitted by each server in the optical tunnel switching networks is 10 7 data packets. In our simulation results, the throughput of an optical tunnel switching network with 25 nodes using one plane and two planes under various loads is shown in Figure 15.
Obviously, the two-plane Torus topology optical switching network is sufficient to provide a throughput close to 100% of the average service throughput of the server.
In our simulation results, the throughput of an optical tunnel switching network with 49 nodes in one plane, two planes and three planes under different loads is shown in Figure 16. Obviously, a three-plane Torus topology optical switching network is sufficient to provide a throughput close to 100% of the average service throughput of the server. Appl. Sci. 2020, 10, x FOR PEER REVIEW 11 of 16   Figure 13 shows the traffic analysis of Ethernet packets. The top of Figure 13 shows the cumulative percentage of packets of various packet sizes, while the bottom shows the cumulative percentage of bytes of various packet sizes. We can see that when the packet size is less than 1300 bytes, the cumulative percentage of bytes is only about 20%, but the cumulative percentage of packets is as high as 68%. The distribution of Ethernet packets of various packet sizes is shown in Figure 14. For example, a packet size group from 100 to 179 bytes is 50.35%, while a packet size group from 1300 to 1538 bytes is 33.21%. This 33.21% of packets has more traffic because these packets are large. The distribution of the traffic model is used to simulate the optical tunnel switching networks. We simulated the performance of the optical tunnel switching networks obtained from a computer. According to the Ethernet data packet distribution shown in Figure 14, the number of data packets transmitted by each server in the optical tunnel switching networks is 10 7 data packets. In our simulation results, the throughput of an optical tunnel switching network with 25 nodes using one plane and two planes under various loads is shown in Figure 15.     Figure 13 shows the traffic analysis of Ethernet packets. The top of Figure 13 shows the cumulative percentage of packets of various packet sizes, while the bottom shows the cumulative percentage of bytes of various packet sizes. We can see that when the packet size is less than 1300 bytes, the cumulative percentage of bytes is only about 20%, but the cumulative percentage of packets is as high as 68%. The distribution of Ethernet packets of various packet sizes is shown in Figure 14. For example, a packet size group from 100 to 179 bytes is 50.35%, while a packet size group from 1300 to 1538 bytes is 33.21%. This 33.21% of packets has more traffic because these packets are large. The distribution of the traffic model is used to simulate the optical tunnel switching networks. We simulated the performance of the optical tunnel switching networks obtained from a computer. According to the Ethernet data packet distribution shown in Figure 14, the number of data packets transmitted by each server in the optical tunnel switching networks is 10 7 data packets. In our simulation results, the throughput of an optical tunnel switching network with 25 nodes using one plane and two planes under various loads is shown in Figure 15.  Obviously, the two-plane Torus topology optical switching network is sufficient to provide a throughput close to 100% of the average service throughput of the server.    In our simulation results, the throughput of an optical tunnel switching network with 49 nodes in one plane, two planes and three planes under different loads is shown in Figure 16. Obviously, a three-plane Torus topology optical switching network is sufficient to provide a throughput close to 100% of the average service throughput of the server.  Figure 17 shows the traffic distribution of a two-plane optical tunnel switching network with 25 nodes. The maximum available bandwidth between any two nodes is two wavelengths, and the average available bandwidth is 2/3 wavelengths. Generally, the traffic distribution is uneven, so we can combine it with SDN scheduling to meet our bandwidth requirements.  Figure 17 shows the traffic distribution of a two-plane optical tunnel switching network with 25 nodes. The maximum available bandwidth between any two nodes is two wavelengths, and the average available bandwidth is 2/3 wavelengths. Generally, the traffic distribution is uneven, so we can combine it with SDN scheduling to meet our bandwidth requirements. Method 1: We can use three nodes to share the bandwidth of the direct tunnel through SDN scheduling. For example, from node N (1,5) to node N (5, 5) is 2λ bandwidth, node N (5, 5) uses 4/3 λ bandwidth as well as node N (5, 4), and node N (5, 3) use 1/3 λ bandwidth. This condition can be set on the SDN switch of node N (5, 5). Method 2: We use the wavelengths assigned by neighboring nodes, assuming that the original node goes from node N (1,5) to node N (2,5), node N (3,5), node N (4, 5), and node N (5, 5) uses two wavelengths for communication between nodes. Except for node N (5, 5), which uses 4/3 λ bandwidth, other nodes such as node N (5,4) and node N (5, 3) use less than 2/3 λ bandwidth, as shown in Figure 18. The details of the change in traffic volume are shown in Figure 19 and described in detail below. (1,5) (1,1)

Method 1:
We can use three nodes to share the bandwidth of the direct tunnel through SDN scheduling. For example, from node N (1,5) to node N (5, 5) is 2λ bandwidth, node N (5, 5) uses 4/3 λ bandwidth as well as node N (5, 4), and node N (5, 3) use 1/3 λ bandwidth. This condition can be set on the SDN switch of node N (5, 5).

Method 2:
We use the wavelengths assigned by neighboring nodes, assuming that the original node goes from node N (1,5) to node N (2, 5), node N (3,5), node N (4, 5), and node N (5, 5) uses two wavelengths for communication between nodes. Except for node N (5, 5), which uses 4/3 λ bandwidth, other nodes such as node N (5,4) and node N (5, 3) use less than 2/3 λ bandwidth, as shown in Figure 18. The details of the change in traffic volume are shown in Figure 19 and described in detail below.
Due to traffic changes, the bandwidth requirement of node N (5, 5) exceeds the 4/3 λ bandwidth. If the sum of the bandwidth of node N (4,5), node N (4, 2) and node N (4, 1) is equal to or less than the 1λ bandwidth, we can allocate a λ 3 of node N (4, 5) and change it to node N (5, 5), so that a wavelength of 3λ can be used from node N (1,5) to node N (5, 5), including node N (5,4) and node N (5, 3), which is shown in Figure 19. However, the total bandwidth used by node N (5, 5) for the entire network is still limited to 16λ. Therefore, if the bandwidth requirement of a node exceeds the 2λ (200%) bandwidth (red light), including relay tunnel nodes, then we can allocate the neighbor nodes that are smaller than a λ (100%) bandwidth (green light), including relay tunnel nodes to this busy node (red light) to avoid traffic loss.
The traffic management of a three-plane optical tunnel switching network with 49 nodes is similar to the traffic management of a two-plane optical tunnel switching network with 25 nodes.
nodes. The maximum available bandwidth between any two nodes is two wavelengths, and the average available bandwidth is 2/3 wavelengths. Generally, the traffic distribution is uneven, so we can combine it with SDN scheduling to meet our bandwidth requirements. Method 1: We can use three nodes to share the bandwidth of the direct tunnel through SDN scheduling. For example, from node N (1, 5) to node N (5, 5) is 2λ bandwidth, node N (5, 5) uses 4/3 λ bandwidth as well as node N (5, 4), and node N (5, 3) use 1/3 λ bandwidth. This condition can be set on the SDN switch of node N (5, 5). Method 2: We use the wavelengths assigned by neighboring nodes, assuming that the original node goes from node N (1,5) to node N (2, 5), node N (3,5), node N (4, 5), and node N (5, 5) uses two wavelengths for communication between nodes. Except for node N (5, 5), which uses 4/3 λ bandwidth, other nodes such as node N (5, 4) and node N (5, 3) use less than 2/3 λ bandwidth, as shown in Figure 18. The details of the change in traffic volume are shown in Figure 19 and described in detail below. Due to traffic changes, the bandwidth requirement of node N (5, 5) exceeds the 4/3 λ bandwidth. If the sum of the bandwidth of node N (4, 5), node N (4, 2) and node N (4, 1) is equal to or less than the 1λ bandwidth, we can allocate a λ3 of node N (4, 5) and change it to node N (5, 5), so that a wavelength of 3λ can be used from node N (1, 5) to node N (5, 5), including node N (5, 4) and node N (5, 3), which is shown in Figure 19. However, the total bandwidth used by node N (5, 5) for the entire network is still limited to 16λ. Therefore, if the bandwidth requirement of a node exceeds the 2λ (200%) bandwidth (red light), including relay tunnel nodes, then we can allocate the neighbor nodes that are smaller than a λ (100%) bandwidth (green light), including relay tunnel nodes to this busy node (red light) to avoid traffic loss. (1,5) ( The traffic management of a three-plane optical tunnel switching network with 49 nodes is similar to the traffic management of a two-plane optical tunnel switching network with 25 nodes. If the node's traffic exceeds 200%, then 5: Analyze the traffic of the nodes, and then find the busy tunnel. 7: If there is a neighboring node whose traffic is less than one λ bandwidth, then 8: Assign a λ bandwidth of the neighboring node to the busy node. 10: } 11: } 12: } Figure 20 shows the application architecture of the Torus topology tunnel switching network in the mobile edge computing (MEC) of Telecom 5G. When high-speed, high-bandwidth, low-latency applications are required, the traffic is directly transmitted to the server via the Optical Tunnel Switch (OTS), without being sent to the cloud data center for processing of big data. The architecture can provide up to 1764 servers to establish 2352 tunnels. Of course, the aforementioned architecture can also be used in the data center or telecommunications core network to handle big data. (1,5) (  Scanning traffic for all nodes 3: { 4.

Applications
If the node's traffic exceeds 200%, then 5: Analyze the traffic of the nodes, and then find the busy tunnel. 7: If there is a neighboring node whose traffic is less than one λ bandwidth, then 8: Assign a λ bandwidth of the neighboring node to the busy node. 10: } 11: } 12: } Figure 20 shows the application architecture of the Torus topology tunnel switching network in the mobile edge computing (MEC) of Telecom 5G. When high-speed, high-bandwidth, low-latency applications are required, the traffic is directly transmitted to the server via the Optical Tunnel Switch (OTS), without being sent to the cloud data center for processing of big data. The architecture can provide up to 1764 servers to establish 2352 tunnels. Of course, the aforementioned architecture can also be used in the data center or telecommunications core network to handle big data.

Conclusions
A large-scale optical tunnel switching network using Torus topology networks and wavelength selective switches for future big data applications has been presented. All nodes of a large-scale optical tunnel switching network use WSS switching modules, and communication between nodes uses multiple wavelengths to establish a tunnel, and its wavelength can be reused. Because using a wavelength to establish a tunnel not only simplifies the complexity of the network, but also reduces the latency of the network. When the traffic of one node is increased, the number of tunnels for the node can be increased in time to balance the load traffic to avoid traffic loss. Therefore, many applications of the Internet of Things with big data in the future can be realized. The simulation results show that the Torus topology optical switching network with 25 nodes needs two planes to provide approaching to 100% throughput, and the Torus topology optical switching network with 49 nodes needs three planes to provide approaching to 100% throughput. The advantages of the optical tunnel switching network is that it does not require high-speed switching elements and flexible channel scheduling to avoid data loss when traffic volume is changed. C2TCP (Cellular Controlled Delay TCP) uses TCP to calculate the average delay time with RTT (Round Trip Time) of each packet to control the transmission traffic [13]. Therefore, the multi-plane Torus topology network architecture can also use TCP to calculate the average delay time with RTT of each packet using the delay time of transmitting ACK signals, and then estimate the traffic between nodes, and nodes via the SDN controller, according to the average delay time. When the traffic is changed, the number of network tunnels can be scheduled in time to balance the load traffic to avoid traffic loss.

Conclusions
A large-scale optical tunnel switching network using Torus topology networks and wavelength selective switches for future big data applications has been presented. All nodes of a large-scale optical tunnel switching network use WSS switching modules, and communication between nodes uses multiple wavelengths to establish a tunnel, and its wavelength can be reused. Because using a wavelength to establish a tunnel not only simplifies the complexity of the network, but also reduces the latency of the network. When the traffic of one node is increased, the number of tunnels for the node can be increased in time to balance the load traffic to avoid traffic loss. Therefore, many applications of the Internet of Things with big data in the future can be realized. The simulation results show that the Torus topology optical switching network with 25 nodes needs two planes to provide approaching to 100% throughput, and the Torus topology optical switching network with 49 nodes needs three planes to provide approaching to 100% throughput. The advantages of the optical tunnel switching network is that it does not require high-speed switching elements and flexible channel scheduling to avoid data loss when traffic volume is changed.