Experimental demonstration of an intelligent control plane with proactive spectrum defragmentation in SD-EONs

Xu Zhang; Lei Guo; Weigang Hou; Siqi Wang; Qihan Zhang; Pengxing Guo; Ruijia Li

doi:10.1364/OE.25.024837

⁴ houweigang@cse.neu.edu.cn

1. Introduction

The utilization efficiency of network resources is always one of the main concerns of optical network operators. Recently, elastic optical networks (EONs) [1] have been proposed to more efficiently utilize network spectrum resources because the spectrum can be divided into frequency slots which have finer granularity than fixed-grid networks (e.g. wavelength division multiplexing (WDM) networks). This advantage owes to the introduction of both optical orthogonal frequency division multiplexing (OOFDM) [2,3] and flexible transceivers [4,5], via which, we can dynamically change the modulation format and the transmission rate of the optical signal without hardware modifications. For example, by adjusting the drive signal, the transmitter is able to flexibly switch the modulation format including binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), 8-ary quadrature amplitude modulation (8-QAM), and 16-QAM. To fully take the advantage of EONs, someone needs to dynamically and globally re-arrange dispersing unoccupied frequency slots into available spectrum blocks, i.e., spectrum defragmentation. Therefore, it is essential to facilitate high-efficient spectrum defragmentation in EONs by using a centralized control rather than a distributed way.

It is well known that the software-defined networking (SDN) [6,7] with OpenFlow [8] allows operators to control network by using a centralized controller. SDN breaks the vertical integration by separating the control plane from the underlying data forwarding plane. In a SDN-based network, network devices are only responsible for the data forwarding, and control strategies are implemented in a logically centralized controller. OpenFlow is commonly defined as an open standard protocol, and it plays an important role in SDNs. It is able to provide a communication interface between the control plane and the data plane. In addition, most modern switches (network nodes that use circuit-switched technology need to be extended) can be abstracted as the common set of functions. These functions are explained in terms of data flow, i.e., the data can be identified, characterized, and manipulated based on flow table. Therefore, the combination of SDN and EONs (i.e., SD-EONs) can improve the utilization of network resources [9–14] by using software programming. Previously, there were efforts to achieve dynamic lightpath provisioning [15–18], and someone utilized the spectrum defragmentation [19–22] to improve the spectral efficiency. However, the existing approaches dealing with spectrum fragmentation are mostly reactive strategies. In other words, when the controller is aware of the necessity of the spectrum fragmentation, it reconfigures the routing and spectrum to realize the online defragmentation. Due to frequent network reconfiguration, the processing latency of the control plane becomes very high, thus worsening the overall network performance.

To alleviate the processing pressure generated by the reactive strategy, the proactive strategy should be better. As shown in Fig. 1, we compare reactive and proactive strategies in terms of the spectrum defragmentation in SD-EONs. It can be seen that the proactive strategy globally schedules the spectrum resource for the purpose of reducing spectrum fragmentation by collecting the underlying network status in real time. Instead, the reactive strategy merely implements online spectrum defragmentation to improve the network performance. It is worth mentioning that, in addition to spectrum fragmentation under spectral continuity constraints, the fragmentation generated by inefficient resource allocation algorithm is more easily neglected. The reactive strategy ignores the impact of inefficient resource allocation algorithm, which will lead to more reconfiguration operations. However, the proactive strategy not only reorganizes spectrum fragmentation generated by the continuity constraint but also reduces the spectrum fragmentation by rationalizing the allocation of resources. Unlike reactive strategies and general proactive strategies, our proactive solution carries more business by increasing the connectivity of available spectrum slots. Therefore, the control plane using the proactive spectrum defragmentation will enhance the intelligence of SD-EONs.

Fig. 1 Reactive strategy and proactive strategy in SD-EONs.

Download Full Size | PDF

For this end, we tend to design an intelligent control plane with the SEC-RMLSA algorithm. It owns a proactive spectrum defragmentation strategy in SD-EONs. More specifically, we first elaborate on the overall network architecture, OpenFlow protocol extensions supporting the establishment of optical paths, and system model design. Then, we formulate an integer linear programming (ILP) model to solve the problem of the proactive spectrum defragmentation. Next, we propose a SEC-RMLSA algorithm to improve the utilization efficiency of network resources, and illustrate the process of the algorithm in detail. Moreover, we develop a routing application with the SEC-RMLSA by extending the RYU controller [23], and implement our solution in a semi-practical SD-EON testbed. Finally, the overall feasibility and efficiency of the proposed SEC-RMLSA are experimentally verified. In the six-node eight-link network (n6s8) topology mentioned in Ref [29], our heuristic algorithm’s value is close to ILP result, and the average convergence ratio between the SEC-RMLSA and ILP is about 94.37%. More importantly, all the proposed methodologies are demonstrated based on a small topology in an OpenFlow-based control platform. In addition, the performance of the SEC-RMLSA scheme under different traffic loads is also quantitatively evaluated based on the simulated NSFNET network in terms of the spectrum efficiency, network throughput, blocking probability and end-to-end delay, compared with another commonly proactive provisioning scheme [24]. Specifically, a more advanced reactive algorithm [20] called fragmentation-aware routing and spectrum assignment (FA-RSA) is also compared with the SEC-RMLSA algorithm. It is beneficial for us to verify the feasibility and effectiveness of overall system.

The contributions of this work can be summarized as follows,

1) To maximize the spectral efficiency and the connectivity of all available spectrum slots, we formulate an ILP model to tackle the problem of static network planning and obtain the corresponding optimal solution.
2) To improve the performance of optical networks, we propose a novel algorithm called SEC-RMLSA that can achieve the proactive spectrum defragmentation within acceptable running time.
3) We develop an intelligent control application for SEC-RMLSA strategy, and we make extensions of OpenFlow protocol for accomplishing the seamless operation between the controller and the optical data plane.
4) We implement and experimentally demonstrate the proposed system in a semi-practical SD-EON testbed.

The rest of the paper is organized as follows. Section 2 discusses the overall network architecture, system model, SEC-RMLSA algorithm design, and procedure of dynamic path creation. Then, system implementation, experimental demonstration, and performance evaluation are reported in Section 3. Finally, Section 4 summarizes the paper.

2. Network architecture and system model

2.1 Network architecture

The system architecture of SD-EONs is shown in Fig. 2(a). It consists of control plane and data plane. The data plane includes bandwidth-variable wavelength selective switches (BV-WSSs) and edge routers (ERs). They are interconnected with optical fiber, and there is no ability to decide how to handle network traffic. They are only responsible for transmitting the high-speed optical stream according to the decisions made by the control plane. In the data plane, Figs. 2(b)-2(d) show the detailed structures of BV-WSS, OF-AG and ER, respectively. BV-WSS can switch the variable-size spectrum with a granularity at 12.5 GHz, and bandwidth-variable transponders (BV-Ts) are able to set up lightpaths with various bandwidths for client traffic by allocating enough number of frequency slots. In a real network, ERs are usually deployed at some edges of SD-EONs to access the client traffic. As shown in Fig. 2(d), an ER includes several BV-Ts and a spectrum multiplexer/demultiplexer (MUX/DEMUX). If a new flow that is assembled from a bundle of IP packets with same destination is received by the ER, and it does not match any of the existing flow entries, the ER is able to send the first packet of the incoming flow to the controller by using Packet-In message. Moreover, ER can also add, remove, and modify flow entries based on the Flow-Mod message. In SD-EONs, both BV-WSS and ER are required to support the OpenFlow protocol and conduct a cross-connection operation based on the flow entry. Therefore, an OpenFlow Agent (OF-AG) is attached to each of them. Each OF-AG communicates with the controller using extended OpenFlow protocol. More specifically, it can automatically send a vendor-specific command (e.g. Transaction Language 1) to configure the cross-connect of corresponding ports of optical elements by parsing the optical flow entry. The control plane of SD-EONs consists of a number of software modules and an application manager as the coordinator. The application manager is the core component of the control plane, which is able to allow all loaded applications to run in an orderly and stable manner. It is worth noting that a traffic engineering application running the SEC-RMLSA algorithm is programmed in the control plane, which can improve the efficiency of spectrum utilization by enhancing spectral connectivity. In addition, the Packet-In event is extended as an input to the algorithm, which contains not only the source/destination addresses, but also the bit rate of each optical flow. Meanwhile, the extended Flow-Mod message carrying the RMLSA results from the controller can establish a flow entry of optical cross-connection, including input/output ports, central frequency (CF), slot width (SW), and modulation format (MF), as shown in Fig. 2(e). For the flex-grid EON, the CF is defined by $193.1 T H z + x \times 0.00625 T H z$ , where $x$ is a positive or negative integer including 0 and 0.00625 is the nominal CF granularity in THz. The SW is defined by $12.5 G H z \times y$ , where $y$ is a positive integer and 12.5 is the SW granularity in GHz. Any combination of frequency slots is allowed as long as no two frequency slots overlap. In this paper, the MF mainly includes BPSK, QPSK, 8-QAM, and 16-QAM.

Fig. 2 (a) Network architecture of SD-EONs. Detailed structures of (b) Bandwidth-variable wavelength selective switch, (c) OpenFlow-Agent, (d) Edge router, and (e) Optical flow entry.

Download Full Size | PDF

2.2 System model

We model the SD-EONs by a graph $G (V, E)$ , in which $V = {v_{1}, v_{2}, ..., v_{M}}$ denotes the set of network elements and $E = {e_{1}, e_{2}, ..., e_{L}}$ is the set of bidirectional edges between network elements. There are $N$ frequency slots (FSs) on each fiber link, and the bandwidth of each FS is $F$ GHz. $R$ represents the set of all service requirements. Every service requirement $r \in R$ can be composed of a 4-tuple: $< s, d, b, t >$ , where $s$ is an IP address of source client, $d$ is an IP address of destination client, $b$ is the bandwidth requirement, and $t$ is the required service duration. $P_{r} = {p_{1}, p_{2}, ..., p_{K}}$ represents a set of $K$ candidate lightpaths for the request $r$ . Based on the selected lightpath of a traffic request, the request will have the corresponding CF $c f_{r}$ , SW $s w_{r}$ , and MF $m f_{r}$ . Then, let $U_{e}^{r} = {u_{e, 1}^{r}, u_{e, 2}^{r}, \dots, u_{e, N}^{r}}$ denote the status set of FSs for the link $e$ of the request $r$ , where $e \in E$ and $u_{e, n}^{r} (n = {1, 2, \dots, N})$ is a binary variable, taking 0 if the $n_{t h}$ spectrum slot of the link $e$ is occupied by the request $r$ , otherwise, taking 1. $U_{p}^{r}$ represents the status set of FSs for the candidate lightpath $p$ of the request $r$ , where $\forall p \in P_{r}$ . Thus, $U_{p}^{r} = \underset{e \in p}{\cap} U_{e}^{r}$ . Finally, we define $U_{e} = {u_{e}^{1}, u_{e}^{2}, \dots, u_{e}^{N}}$ as the status set of FSs for the link $e \in E$ , where $u_{e}^{n} (n = {1, 2, \dots, N})$ is a binary variable, taking 0 if the $n_{t h}$ spectrum slot of the link $e$ is not available, and 1 otherwise. Using the network model above, our objective is to maximize the spectral efficiency and the connectivity of all available spectrum slots, which can proactively reduce the generation of spectrum fragmentation. High available spectrum consecutiveness produces less spectrum fragmentation, because it provides more available continuous spectrum slots for unsolicited traffic requests. Therefore, SD-EONs owning the high connectivity can accommodate more traffic requests. Mathematically, our problem can be formulated as Eq. (1).

M a x i m i z e \sum_{r \in R} \sum_{e \in E} [\frac{b}{s w_{r} \times F} \times (\frac{\sum_{n = 1}^{N - 1} u_{e}^{n} \cdot u_{e}^{n + 1}}{\sum_{n = 1}^{N} u_{e}^{n} - \sum_{n = 1}^{N - 1} u_{e}^{n} \cdot u_{e}^{n + 1}} \times \frac{\sum_{i = 1}^{N} u_{e}^{n}}{N})]

Here, the first multiplier represents the spectral efficiency, and the multipliers left are defined as the available spectrum consecutiveness of link $e$ . Their product indicates the comprehensive performance of the spectral efficiency and connectivity, which is exactly our objective of the proactive spectrum defragmentation.

Constraints: Notations and variables for problem constraints: $α_{p}^{r}$ is binary variable, taking 1 if the lightpath $p$ ( $\forall p \in P_{r}$ ) is used for the request $r (\forall r \in R)$ , and 0 otherwise. $φ_{r, r'}$ is binary variable, taking 1 if the index of starting FS for the request $r$ is less than that of the request $r'$ , and 0 otherwise. $S_{r}$ is the index of starting FS allocated to the request $r$ , while $E_{r}$ is the index of ending FS.

\sum_{p \in P_{r}} α_{p}^{r} = 1, \forall r \in R

s w_{r} = ⌈ \frac{b_{r}}{m f_{r} \times 2 F} ⌉, \forall r \in R

\sum_{n = 1}^{N} (1 - u_{e, n}^{r}) = s w_{r} \cdot α_{p}^{r}, \forall r \in R, \forall e \in P

φ_{r, r'} + φ_{r', r} = 1, \forall r, r' \in R

S_{r} - S_{r'} > s w_{r'} \cdot φ_{r', r}, i f φ_{r', r} = 1

S_{r'} - S_{r} > s w_{r} \cdot φ_{r, r'}, i f φ_{r, r'} = 1

E_{r} = S_{r} + s w_{r} - 1, \forall r \in R

E_{r'} - S_{r} \geq (s w_{r} + s w_{r'}) \cdot (φ_{r, r'} + 2 - α_{p}^{r} - α_{p'}^{r'}) - 1, i f φ_{r, r'} = 1

E_{r} - S_{r'} \geq (s w_{r'} + s w_{r}) \cdot (φ_{r', r} + 2 - α_{p'}^{r'} - α_{p}^{r}) - 1, i f φ_{r', r} = 1

u_{e}^{n} = \prod_{r \in R} u_{e, n}^{r}, \forall e \in E, n = {1, 2, \dots, N}

Equation (2) ensures the uniqueness of routing. Equation (3) calculates the slot width of the traffic request. Equation (4) ensures the continuous spectrum allocation of a request. Equations (5)-(7) ensure that the starting spectrum slots of the different services are allocated in the correct direction. Equation (8) calculates the index of ending FS, and Eqs. (9) and (10) ensure that the occupied spectrum on the common links of different services cannot overlap. Equation (11) calculates the spectrum occupied status of each link. Note that, for our optimal resource allocation problem, the objective function (see Eq. (1)) has $R \cdot L \cdot (N - 1) \cdot N$ variables, and the constraints (see Eq. (2) and Eq. (4)) totally have $(K + R \cdot N \cdot L)$ variables. In addition, for the whole network graph, our problem approximately has $[R \cdot L \cdot (N - 1) \cdot N + (K + R \cdot N \cdot L)] \cdot M^{2}$ variables in total. For example, if we assume that $M = 10, L = 90, N = 10^{2}, R = 100$ and $K = 5$ , then this problem will have 9,000,000,500 variables, which is more than the NP-hard problem mentioned in [25]. Therefore, our problem is also a NP-hard problem. Since it can be proved that the problem is NP-hard problem, we also design a novel heuristic algorithm called SEC-RMLSA. The detailed process is described in the Algorithm 1.

2.3 SEC-RMLSA algorithm

We design a novel routing, modulation level and spectrum allocation algorithm called SEC-RMLSA to improve the utilization efficiency of network resources. In this subsection, we first present the main procedure of the SEC-RMLSA algorithm shown in Algorithm 1. We then illustrate the operation of the algorithm by using a simple example. It is worthy to note that we have made an extension to the OpenFlow protocol to support SD-EONs. The major extensions include both Packet-In and Flow-Mod messages. The Packet-In is extended to carry the bit rate of each incoming optical flow, and the request bandwidth is the input parameter of the SEC-RMLSA algorithm. In turn, the Flow-Mod message is extended to carry results of the SEC-RMLSA algorithm to set up an optical flow entry including actions (e.g. Add/Remove/Modify flow entry), input and output ports, CF, SW, and MF. Firstly, we calculate K-shortest path (KSP) [30] candidates based on real-time network status, where the edge weight of the virtual network graph is the value of the distance divided by the number of residual spectral slots. The purpose of this weight design is to maximize the load balancing of the network. Next, we calculate the total distance of each candidate path to select the suitable modulation format [26,27]. After determining the modulation format, the required spectral slot width is calculated from the requested bandwidth. Then, the set of spectral slots of all links in a path is carried out the bitwise-and operation, which is to satisfy the constraint of spectral continuity. We count the number of available spectrum blocks in the path, and the available spectrum blocks refer to the continuous spectrum slots that are greater than or equal to the requested slot width. Finally, it is assumed that the requested spectrum slots are allocated to the available spectrum blocks in the path, respectively, and then the objective function values of the traces are calculated, and the spectral block with the largest objective function value is selected. Similarly, repeat the above steps for all K paths, and finally select the optimal path and spectrum block to determine the CF of the service request. According to the pseudo code shown in Algorithm 1, line 5 runs the KSP algorithm. Dijkstra’s algorithm has a worse case time complexity of $O (M^{2})$ , but using a Fibonacci heap it becomes $O (L + M \log M)$ . Since KSP’s algorithm makes $K M$ calls to the Dijkstra at the worst case, the time complexity KSP becomes $K M (L + M \log M)$ . The worst time complexity occurs from lines 9 to 38, which is $K M + K N^{2} + N$ . Therefore, the total time complexity of SEC-RMLSA algorithm is approximate $O (K M^{2} \log M + K N^{2} + K M L)$ , which is polynomial.

Table 1. Summary of modulation format and path distance.

View Table | View all tables in this article

To show this process more clearly, we use the example shown in Fig. 3 to further illustrate the details of the algorithm. For example, the bandwidth of the service request is 400 Gbps, and it is assumed that each link has 16 frequency slots. Meanwhile, based on the network graph of the controller real-time maintenance, we calculated the K alternative light paths, one of which is 650 km length along with two links. At this point, the occupancy state of spectrum slots for the link 1 and 2 are shown in the Fig. 3. We can choose the high-level modulation format (16-QAM) for this path based on the distance adaptive principle. According to the Nyquist criterion, the number of spectrum slots required for this service is $s w_{r} = ⌈ b_{r} / (m f_{r} \times 2 F) ⌉ = ⌈ 400 / (4 \times 2 \times 12.5) ⌉ = 4$ . Thus, the available spectrum blocks in the path are Block 1 and 2, respectively. Assuming that the traffic is assigned to spectrum 2-5 in spectrum block 1, the spectrum slots 6, 7, 12, 13, 14 and 15 are left. Spectrum slots 6 and 7 form a connection point, and 12-15 generate three connection points. Thus, the total number of connection points is 4, and the number of available spectrum slots is 6. We get the objective function value based on Eq. (1), $O F V_{p}^{B (1)} = 400 / (4 \times 12.5) \times 4 / (6 - 4) \times 6 / 16 = 6$ . Similarly, if the traffic is assigned to spectrum 12-15 in spectrum block 2, then 5 connection points can be obtained according to the unused spectrum slots 2-7. Therefore, the value of objective function can be calculated as $O F V_{p}^{B (2)} = 400 / (4 \times 12.5) \times 5 / (6 - 5) \times 6 / 16 = 15$ . Since $O F V_{p}^{B (2)} > O F V_{p}^{B (1)}$ , the traffic should be assigned to block 2, and then the CF is determined.

Fig. 3 The details of the SEC-RMLSA algorithm.

Download Full Size | PDF

Algorithm 1. Algorithm 1. SEC-RMLSA algorithm

View Table | View all tables in this article

2.4 Procedure for end-to-end dynamic path creation

In conjunction with the network topology of Fig. 2(a), an end-to-end optical path setup procedure is summarized below:

● Step 1: The controller establishes a connection with each OF-AG, and acquires the capability of each network element.
● Step 2: After the connection is established, the controller dynamically maintains the network topology.
● Step 3: The controller establishes an access table based on the address resolution protocol (ARP) request from the IP network to find the ER.
● Step 4: An IP flow (e.g. from Client_1 to Client_2) from the IP network arrives at the ER (ER 1).
● Step 5: If the IP flow does not match any of the flow entries in ER 1, the OF-AG 1 will send the extended Packet-In message to the controller.
● Step 6: The controller runs the SEC-RMLSA algorithm based on the current state of network graph to calculate the optical path for the Client_1, and assigns the input/output ports, CF, SW, and MF. And then, it inserts the optical flow entries into OF-AG 1, 2, 3, 5, and 6, respectively.
● Step 7: Each corresponding OF-AG sends a vendor-specific command (e.g. TL1) to the corresponding optical device (i.e., ER 1, BV-WSS 2, BV-WSS 3, BV-WSS 5, and ER 6) to establish the underlying optical path.
● Step 8: The IP flow from Client_1 arrives at Client_2 along the successfully established light path.

3. Experimental demonstration, results and discussion

To evaluate the feasibility and efficiency of the proposed solution, we set up a testbed based on two experimental topologies: (i) a small topology of 6 optical nodes for the proof-of-concept, shown in Fig. 2(a); (ii) an emulated NSFNET topology for stress testing, shown in Fig. 8(a). (iii) a n6s8 network topology [29] for solving ILP model. The SD-EON testbed is built with virtual machines based on the VMWare Workstation. Since each virtual machine has its own operating system (i.e., Ubuntu system) and virtual hardware resources (i.e., 2 processors, 2GB memory, and network adapter), so we see it as a real node. Each OF-AG is programmed based on Mininet (i.e., OpenvSwitch) and the data plane is simulated with software but not a real optical equipment BV-T or BV-WSS, while the controller is realized with the extended RYU platform running on an Ubuntu server. In addition, the RYU controller communicates with the OF-AG using the extended OpenFlow protocol. The RYU controller is extended to realize the abstraction of the optical layer device resources, the dynamic monitoring of the network state, the maintenance of the traffic engineering database, the computation of the light path performing the SEC-RMLSA algorithm, and the adjustment of the spectral parameters.

Firstly, we conduct experiments based on the small topology in the SD-EONs platform. As shown in Fig. 2(a), the Client_1 in the IP network domain sends the connection request to the Client_2 and Client_3 in another IP network domain. The required bandwidth of the IP flow is 400 Gbps. When the IP flow from Client_1 to Client_2 arrives at the ER 1, the OF-AG 1 sends the Packet-In message to the RYU controller to request the establishment of the optical path because there is no matching flow entry. The controller calculates two paths with a total length of 650 km based on the current virtual network slice. We are able to select the 16-QAM modulation format for this IP flow, shown in Table 1 [28]. Thus, the number of spectrum slots required for this service is 4. After executing the SEC-RMLSA algorithm, we can determine that the central frequency is 193.11875 THz and the final selected optical path is ER 1→BV-WSS 2→BV-WSS 3→BV-WSS 5→ER 6. Similarly, for the IP flow between Client_1 and Client_3, we have chosen the same path but the assigned central frequency is different, i.e., 193.16875 THz. Figure 4 shows printed information from the RYU controller during optical paths setup for the two traffic requests. It is obviously to find that the controller establishes the optical routing information including source/destination IP address, slot width, central frequency, modulation format, optical path information and its total distance.

Fig. 4 Printed information from the RYU controller for the establishment of optical path.

Download Full Size | PDF

Figure 5 presents the whole procedure of establishing flow entries for elastic lightpath setup by using extended OpenFlow protocol. As shown in Fig. 5, 192.168.100.100 denotes the IP address of the RYU controller, while 192.168.100.10 denotes the IP address of the Mininet. 10.0.0.1, 10.0.0.2, and 10.0.0.3 denote the IP address of Client_1, Client_2 and Client_3, respectively. Note that, the RYU controller can dynamically maintain an access table for all clients, and use the abstracted information of device capabilities to build virtual network slices. The client who sent the request for the first time needs to send an ARP request to find which ER the destination client belongs to. After receiving the ARP request, the controller queries the access table. If the access table has the destination client’s information, the controller returns the ARP reply message back to the source client. If not, then broadcast ARP request message to all ERs, the controller will update the access table after receiving the ARP reply message, shown in Fig. 5. After receiving connection request from the ER 1, controller performs the SEC-RMLSA algorithm and then sends extended Flow-Mod messages to the OF-AG 1, 2, 3, 5 and 6. It takes 195.78 milliseconds (ms) during the process of establishing optical path between Client_1 and Client_2. The 130 ms is consumed for the establishment of another optical path between Client_1 and Client_3.

Fig. 5 Wireshark capture of the OpenFlow protocol during lightpath setup.

Download Full Size | PDF

As shown in Fig. 6, we can clearly observe the detailed process of establishing optical path ER 1→BV-WSS 2→BV-WSS 3→BV-WSS 5→ER 6. Five optical flow entries are inserted into the flow tables of OF-AG 1, 2, 3, 5 and 6, respectively. In addition, the source address of the IP flow is 10.0.0.1, while the destination IP address is 10.0.0.2. Meanwhile, we can also find the input/output ports of device along the optical path. For example, in the ER 1, the value ‘00000001’ represents that input port of the IP flow is No. 1, while the output port is No. 2 in the action_output field. It should be noted that the CF, SW and MF are added to the padding field.

Fig. 6 The detailed flow entries of an optical path.

Download Full Size | PDF

Figure 7 illustrates the wireshark capture of the extended Flow-Mod message, which is sent to ER 1. The “Command” field is “OFPFC_ADD” that means to add a new cross-connection flow entry for the establishment of the optical path. The “Hard Timeout = 1000” indicates that the holding time of the optical path is 1000 seconds. The “input port = 0x00000001, action output port = 0x00000002” show that the traffic flow is transmitted from No. 1 to No. 2 port in the optical node. The “source IP address = 0x0a000001, destination IP address = 0x0a000002” represents that the source and destination addresses of the IP flow are 10.0.0.1 and 10.0.0.2, respectively. In addition, as shown in Fig. 7, it can be seen that the CF, SW and MF are encapsulated in the extended Flow-Mod message, and the value of the CF is 193.11875 THz, the required SW is 4*12.5 GHz, modulation level is 4, i.e. 16-QAM.

Fig. 7 Wireshark capture of an extended Flow-Mod message.

Download Full Size | PDF

Then, we carry out a lot of dynamic experiments to stress-test the proposed solution in the SD-EONs platform, and evaluate the overall efficiency of the system. Our experimental topology is based on the NSFNET network shown in Fig. 8(a). Each fiber link is characterized by 100 individual slots of 12.5 GHz each, and the length of the fiber is in kilometers. In this experiment, we consider four different bit rate traffic requests (10 Gbps, 50 Gbps, 100 Gbps, and 400 Gbps) and four modulation formats (BPSK, QPSK, 8-QAM, and 16-QAM). In addition, the flow request is randomly generated between any two nodes, and the modulation format is determined based on the path length shown in Table 1. High-level modulation format is assigned to the short path, while long path can only use low-level modulation format. Note that, the connection request of optical path in a real optical network scenario is not as frequent as in a packet switched network. However, in our experiments, since we want to critically evaluate the performance of the system, our traffic request model is subject to the Poisson distribution, and the holding time of each request follows a negative exponential distribution. Figure 8(b) shows a web-based graphical user interface in the RYU controller, which illustrates the OF-AG connection in the NSFNET topology. And we are able to observe that the IP address of the RYU controller is 192.168.100.100 from the address bar of the web page.

Fig. 8 (a) NSFNET topology for the SD-EONs testbed. (b) The GUI of RYU controller.

Download Full Size | PDF

Next, we compare different performance metrics of the SEC-RMLSA algorithm with the proactive first-fit routing and spectrum allocation (FF-RSA) algorithm. The objective of the FF-RSA algorithm is to minimize the index of the maximum utilized frequency slot [24]. As shown in Fig. 9, we record the spectrum efficiency, network throughput, blocking probability and end-to-end delay, respectively. The results on the spectrum efficiency in Fig. 9(a) indicate that the proposed SEC-RMLSA achieves higher spectrum efficiency. This is because the SEC-RMLSA uses a distance-adaptive modulation scheme, which can dynamically improve the overall spectrum efficiency of the network as the modulation level increasing. The average spectral efficiency of the proposed SEC-RMLSA is maintained at about 3.9 bps/Hz, while the FF-RSA is about 1.9 bps/Hz. As shown in Fig. 9(b), when the number of request is lower than 200, the network throughput of the FF-RSA algorithm is only slightly lower than the SEC-RMLSA. However, when the number of request keeps increasing, FF-RSA falls significantly. This is because the available spectrum slots of the FF-RSA algorithm are reduced due to spectral continuity constraints. Meanwhile, the FF-RSA also produces more spectrum fragmentation with the number of requests increasing. In Fig. 9(c), we also report the blocking probability versus the number of requests. The SEC-RMLSA can reduce the blocking probability significantly comparing to the FF-RSA. The reason for this is that the proposed SEC-RMLSA accommodates more connection requests by reducing spectrum fragmentation, which leads to higher resource occupation rate or lower blocking rate. Thus, the experimental results demonstrate the efficiency of our solutions. Figure 9(d) quantitatively shows the histogram of the end-to-end delay in a serial mode. We can see that the max delay is about hundreds of milliseconds, which consists of the propagation latency between the controller and the OF-AGs, the time latency for flow entry insertion, and the internal processing latency in the controller. Thanks to the introduction of the traffic engineering database, the minimum latency is only tens of milliseconds. Compared with the FF-RSA algorithm, our SEC-RMLSA algorithm does not bring much latency, which verifies its feasibility.

Fig. 9 Experimental results. (a) Spectrum efficiency. (b) Network throughput. (c) Blocking probability. (d) End-to-end delay.

Download Full Size | PDF

Finally, the effectiveness of the SEC-RMLSA proposed is shown in Fig. 10(a). Similar ILP optimization results are obtained for the n6s8 network. Since we have proved that our problem is NP-hard, to obtain the problem solution in a limited time, the adopted topology and the number of requests should be small.

Fig. 10 (a) Available spectrum consecutiveness. (b) Blocking probability.

Download Full Size | PDF

Here, 50 spectrum slots are provided in each fiber link, and the number of requests is set to 5 to 10. Based on the settings above, we build the LINGO model to solve the ILP proposed in section 2.2. Then, by using the Global Solver in LINGO 11.0, the optimal objective function result can be obtained with different number of requests. As can be seen from Fig. 10(a), with the increasing number of requests, both the ILP model and the SEC-RMLSA algorithm illustrate the same trend of the objective result. Intuitively, this trend indicates the correctness of our SEC-RMLSA algorithm. More importantly, the value of our SEC-RMLSA is very close to the ILP model’s value with the average convergence ratio of 94.37%, which also demonstrates the effectiveness of our heuristic algorithm. In other words, our SEC-RMLSA algorithm has a great potential for solving our problem in bigger topology or processing more requests. In addition, a more advanced reactive algorithm, i.e., FA-RSA [20], is also compared with the proactive SEC-RMLSA algorithm based on the NSFNET topology. We perform experiments with dynamic lightpath requests in a SD-EON testbed. Figure 10(b) shows the experimental results. The results on blocking probability indicate that the proposed SEC-RMLSA achieves lower blocking probability. This is because that the SEC-RMLSA proactively reduces the spectrum fragmentation in SD-EONs. With the increasing number of services, the advantage of the high connectivity will be more highlighted from the SEC-RMLSA algorithm. As a result, it is possible to carry more traffic requests.

4. Conclusions

In this paper, we investigated how to enhance the intelligence of control plane with the proactive spectrum defragmentation in SD-EONs. We first presented the system architecture of OpenFlow-based SD-EONs. Then, the SEC-RMLSA scheme was introduced for achieving the proactive spectrum defragmentation based on the proposed architecture. Meanwhile, we developed a routing application with the SEC-RMLSA, and extended OpenFlow protocol to implement a seamless operation between the controller and the optical agent by extending the RYU controller. Finally, all the proposed solutions were verified and demonstrated in the SD-EON testbed. And, we also quantitatively evaluated the performance of overall system under different traffic load scenarios in terms of the spectrum efficiency, network throughput, blocking probability and end-to-end delay. The results indicated that the solution with SEC-RMLSA could utilize the resources of elastic optical network effectively, and improved the service responsiveness, while leading to a reduced blocking probability.

Our future work will focus on the investigation of comprehensive control planes for the heterogeneous multi-granularity optical networks, which are the coexistence networks including fixed-grid and flex-grid, IP + Optical network and multi-vendor equipment. Our current solution cannot well solve the problem of resources allocation in the aforementioned complex networks. Thus, we will design a mix of algorithms to enhance the intelligence of the control plane for this scenario. Finally, for large-scale networks, we will utilize multi-controller solutions to solve the problem of network scalability.

Funding

National Natural Science Foundation of China (NSFC) (61471109, 61401082); Fundamental Research Funds for the Central Universities (N150401002, N161608001); General Armament Department and Ministry of Education United Fund (Grant No. 6141A0224-003).

References and links

1. M. Jinno, T. Ohara, Y. Sone, A. Hirano, O. Ishida, and M. Tomizawa, “Elastic and adaptive optical networks: possible adoption scenarios and future standardization aspects,” IEEE Commun. Mag. 49(10), 164–172 (2011). [CrossRef]

2. R. Casellas, R. Munoz, J. Fabrega, M. Moreolo, R. Martinez, L. Liu, T. Tsuritani, and I. Morita, “Design and Experimental Validation of a GMPLS/PCE Control Plane for Elastic CO-OFDM Optical Networks,” IEEE J. Sel. Areas Comm. 31(1), 49–61 (2013). [CrossRef]

3. G. Zhang, M. De Leenheer, A. Morea, and B. Mukherjee, “A Survey on OFDM-Based Elastic Core Optical Networking,” IEEE Comm. Surv. and Tutor. 15(1), 65–87 (2013). [CrossRef]

4. L. Liu, R. Muñoz, R. Casellas, T. Tsuritani, R. Martínez, and I. Morita, “OpenSlice: an OpenFlow-based control plane for spectrum sliced elastic optical path networks,” Opt. Express 21(4), 4194–4204 (2013). [CrossRef] [PubMed]

5. L. Liu, H. Choi, R. Casellas, T. Tsuritani, I. Morita, R. Martínez, and R. Muñoz, “Demonstration of a Dynamic Transparent Optical Network Employing Flexible Transmitters/Receivers Controlled by an OpenFlow-Stateless PCE Integrated Control Plane,” J. Opt. Commun. Netw. 5(10), A66–A75 (2013). [CrossRef]

6. D. Kreutz, F. Ramos, P. Verissimo, C. Rothenberg, S. Azodolmolky, and S. Uhlig, “Software-Defined Networking: A Comprehensive Survey,” Proc. IEEE 103(1), 14–76 (2015). [CrossRef]

7. B. Nunes, M. Mendonca, X. Nguyen, K. Obraczka, and T. Turletti, “A Survey of Software-Defined Networking: Past, Present, and Future of Programmable Networks,” IEEE Comm. Surv. and Tutor. 16(3), 1617–1634 (2014). [CrossRef]

8. “OpenFlow switch specification,” http://archive.openflow.org/.

9. R. Muñoz, R. Casellas, R. Martínez, L. Liu, T. Tsuritani, and I. Morita, “Experimental evaluation of efficient routing and distributed spectrum allocation algorithms for GMPLS elastic networks,” Opt. Express 20(27), 28532–28537 (2012). [CrossRef] [PubMed]

10. H. Yang, J. Zhang, Y. Ji, R. Tian, J. Han, and Y. Lee, “Performance evaluation of multi-stratum resources integration based on network function virtualization in software defined elastic data center optical interconnect,” Opt. Express 23(24), 31192–31205 (2015). [CrossRef] [PubMed]

11. Z. Zhu, C. Chen, X. Chen, S. Ma, L. Liu, X. Feng, and S. Yoo, “Demonstration of Cooperative Resource Allocation in an OpenFlow-Controlled Multidomain and Multinational SD-EON Testbed,” J. Lightwave Technol. 33(8), 1508–1514 (2015). [CrossRef]

12. M. Klinkowski, M. Ruiz, L. Velasco, D. Careglio, V. Lopez, and J. Comellas, “Elastic Spectrum Allocation for Time-Varying Traffic in FlexGrid Optical Networks,” IEEE J. Sel. Areas Comm. 31(1), 26–38 (2013). [CrossRef]

13. B. Chatterjee, N. Sarma, and E. Oki, “Routing and Spectrum Allocation in Elastic Optical Networks: A Tutorial,” IEEE Comm. Surv. and Tutor. 17(3), 1776–1800 (2015). [CrossRef]

14. Y. Zhao, R. He, H. Chen, J. Zhang, Y. Ji, H. Zheng, Y. Lin, and X. Wang, “Experimental performance evaluation of software defined networking (SDN) based data communication networks for large scale flexi-grid optical networks,” Opt. Express 22(8), 9538–9547 (2014). [CrossRef] [PubMed]

15. L. Liu, T. Tsuritani, I. Morita, H. Guo, and J. Wu, “Experimental validation and performance evaluation of OpenFlow-based wavelength path control in transparent optical networks,” Opt. Express 19(27), 26578–26593 (2011). [CrossRef] [PubMed]

16. J. Zhang, J. Zhang, Y. Zhao, H. Yang, X. Yu, L. Wang, and X. Fu, “Experimental demonstration of OpenFlow-based control plane for elastic lightpath provisioning in Flexi-Grid optical networks,” Opt. Express 21(2), 1364–1373 (2013). [CrossRef] [PubMed]

17. Y. Zhao, Z. Chen, J. Zhang, and X. Wang, “Dynamic optical resource allocation for mobile core networks with software defined elastic optical networking,” Opt. Express 24(15), 16659–16673 (2016). [CrossRef] [PubMed]

18. Z. Zhu, X. Liu, Y. Wang, W. Lu, L. Gong, S. Yu, and N. Ansari, “Impairment- and Splitting-Aware Cloud-Ready Multicast Provisioning in Elastic Optical Networks,” IEEE/ACM Trans. Netw. 25(2), 1220–1234 (2017). [CrossRef]

19. S. Li, W. Lu, X. Liu, and Z. Zhu, “Fragmentation-aware service provisioning for advance reservation multicast in SD-EONs,” Opt. Express 23(20), 25804–25813 (2015). [CrossRef] [PubMed]

20. Z. Zhu, X. Chen, C. Chen, S. Ma, M. Zhang, L. Liu, and S. Yoo, “OpenFlow-Assisted Online Defragmentation in Single-/Multi-Domain Software-Defined Elastic Optical Networks,” J. Opt. Commun. Netw. 7(1), A7–A15 (2015). [CrossRef]

21. C. Chen, X. Chen, M. Zhang, S. Ma, Y. Shao, S. Li, M. Suleiman, and Z. Zhu, “Demonstrations of Efficient Online Spectrum Defragmentation in Software-Defined Elastic Optical Networks,” J. Lightwave Technol. 32(24), 4701–4711 (2014). [CrossRef]

22. Y. Wang, J. Zhang, Y. Zhao, J. Zhang, J. Zhao, X. Wang, and W. Gu, “Path connectivity based spectral defragmentation in flexible bandwidth networks,” Opt. Express 21(2), 1353–1363 (2013). [CrossRef] [PubMed]

23. “RYU: an OpenFlow controller,” http://osrg.github.io/ryu/.

24. L. Zhang, N. Ansari, and A. Khreishah, “Anycast Planning in Space Division Multiplexing Elastic Optical Networks With Multi-Core Fibers,” IEEE Commun. Lett. 20(10), 1983–1986 (2016). [CrossRef]

25. W. Hou, R. Zhang, W. Qi, K. Lu, J. Wang, and L. Guo, “A Provident Resource Defragmentation Framework for Mobile Cloud Computing,” IEEE Trans. Emerg. Topics Comput. PP(99), 1–12 (2015). [CrossRef]

26. L. Costa and A. Drummond, “New Distance-Adaptive Modulation Scheme for Elastic Optical Networks,” IEEE Commun. Lett. 21(2), 282–285 (2017). [CrossRef]

27. S. Talebi and G. Rouskas, “On distance-adaptive routing and spectrum assignment in mesh elastic optical networks,” J. Opt. Commun. Netw. 9(5), 456–465 (2017). [CrossRef]

28. C. Rottondi, P. Boffi, P. Martelli, and M. Tornatore, “Routing, Modulation Format, Baud Rate and Spectrum Allocation in Optical Metro Rings With Flexible Grid and Few-Mode Transmission,” J. Lightwave Technol. 35(1), 61–70 (2017). [CrossRef]

29. G. Shen and R. Tucker, “Energy-Minimized Design for IP Over WDM Networks,” J. Opt. Commun. Netw. 1(1), 176–186 (2009). [CrossRef]

30. I. Szczesniak and B. Wozna-Szczesniak, “Adapted and Constrained Dijkstra for Elastic Optical Networks,” in Proceedings of 20th International Conference on Optical Network Design and Modeling, Cartagena, Spain, pp. 1–6 (2016). [CrossRef]

Initialization:
Dynamically maintain the network graph with the edge weight that is the value of the distance divided by the number of residual spectral slots;
When traffic request (s, d, b) arrives, do:
1. src-er, dst-er = get edge router (s,d);
2. if (src-er, dst-er) = Null then:
3.	block request;
4. else:
5.	k-paths = ksp(src-er, dst-er);
6.	if k-paths = Null then:
7.		block request;
8.	else:
9.		for path in k-paths do:
10.			for link = 1 to len(path) do:
11.				calculate the total distance of the path;
12.				path-slots = link 1 & link 2 &…& link (len(path)-1);
13.			end for
14.			select the suitable modulation format mf;
15.			sw = b/(mf2F);
16.			while 0< = start-slot<N-sw + 1 do:
17.				if path-slots[start-slot] is free then:
18.						next-slot = start-slot;
19.						while next-slot<N and path-slots[next-slot] is free do:
20.								next-slot = next-slot + 1;
21.						end while
22.						if next_slot - start_slot > = sw then:
23.								storage the available spectrum blocks in the path to ‘Block’;
24.								start_slot = next_slot;
25.						else:
26.								start_slot = next_slot;
27.						end if
28.				else:
29.							start_slot = start_slot + 1;
30.				end if
31.			end while
32.			for block = 1 to len(Block) do:
33.					calculate the spectral block with the largest objective function value;
34.			end for
35.		select the optimal path and spectrum block;
36.		determine the central frequency;
37.		insert flow entries to related OF-AGs;
38.		end for
39.	end if
40. end if

Initialization:
Dynamically maintain the network graph with the edge weight that is the value of the distance divided by the number of residual spectral slots;
When traffic request (s, d, b) arrives, do:
1. src-er, dst-er = get edge router (s,d);
2. if (src-er, dst-er) = Null then:
3.	block request;
4. else:
5.	k-paths = ksp(src-er, dst-er);
6.	if k-paths = Null then:
7.		block request;
8.	else:
9.		for path in k-paths do:
10.			for link = 1 to len(path) do:
11.				calculate the total distance of the path;
12.				path-slots = link 1 & link 2 &…& link (len(path)-1);
13.			end for
14.			select the suitable modulation format mf;
15.			sw = b/(mf2F);
16.			while 0< = start-slot<N-sw + 1 do:
17.				if path-slots[start-slot] is free then:
18.						next-slot = start-slot;
19.						while next-slot<N and path-slots[next-slot] is free do:
20.								next-slot = next-slot + 1;
21.						end while
22.						if next_slot - start_slot > = sw then:
23.								storage the available spectrum blocks in the path to ‘Block’;
24.								start_slot = next_slot;
25.						else:
26.								start_slot = next_slot;
27.						end if
28.				else:
29.							start_slot = start_slot + 1;
30.				end if
31.			end while
32.			for block = 1 to len(Block) do:
33.					calculate the spectral block with the largest objective function value;
34.			end for
35.		select the optimal path and spectrum block;
36.		determine the central frequency;
37.		insert flow entries to related OF-AGs;
38.		end for
39.	end if
40. end if

Experimental demonstration of an intelligent control plane with proactive spectrum defragmentation in SD-EONs

Abstract

1. Introduction

2. Network architecture and system model

2.1 Network architecture

2.2 System model

2.3 SEC-RMLSA algorithm

2.4 Procedure for end-to-end dynamic path creation

3. Experimental demonstration, results and discussion

4. Conclusions

Funding

References and links

Cited By

Figures (10)

Tables (2)

Equations (11)

Optics Express

Modulation Level	Modulation Format	Subcarrier Bandwidth (GHz)	Subcarrier Capacity (Gbps)	Path Distance (km)
1	BPSK	12.5	25	3000 +
2	QPSK	12.5	50	1500-3000
3	8-QAM	12.5	75	700-1500
4	16-QAM	12.5	100	0-700