Software-defined wide area network elastic routing method based on service quality perception

Abstract

The invention discloses a software-defined wide area network elastic routing method based on QoS (Quality of Service) awareness, which comprises the following steps of: establishing a joint optimization problem model, realizing flow routing of QoS awareness in a data plane, and simultaneously minimizing control delay of a control plane, and further providing a heuristic QoS awareness elastic routing algorithm because the problem has high complexity. The method provided by the invention is used for realizing efficient flow remapping and routing under the controller failure, and the experimental result based on the real flow data under the real network topology shows that compared with the current method, the method provided by the invention has the advantages that the network throughput is obviously improved, and the network delay and the control delay are effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer networks, and particularly relates to a software-defined wide area network elastic routing method based on service quality perception. Background Art

[0002] Emerging network applications have diverse quality of service (QoS) requirements for different service types (e.g., high throughput and low latency requirements). To meet these diverse needs, software-defined networking (SDN) has been introduced into wide area networks (WANs), namely software-defined wide area networks (SD-WANs). By separating the control plane from the data plane, SDN significantly enhances network flexibility and simplifies the management of WANs. An SD-WAN is typically divided into multiple domains, with an SDN controller deployed in each domain to manage all switches within the domain. These geographically dispersed controllers synchronize regularly to maintain the consistency of the global network view. With the global network state information, the control plane can optimize the routing of flows, thereby quickly responding to traffic fluctuations in the WAN and providing differentiated QoS guarantees for various network services and applications. Specifically, this flexible flow management ability implemented by the controller is usually referred to as path programmability.

[0003] Although SDN brings significant advantages to WANs (e.g., enhanced flexibility, path programmability, and centralized management), the fault tolerance of its control plane still faces severe challenges. SDN controllers usually run in the form of physical servers or virtual machines and may fail due to unexpected events such as natural disasters, hardware / software failures, or power outages. When a controller fails, the switches previously managed by these failed controllers will go offline, resulting in the inability of flows relying on their programmability to be flexibly rerouted, and thus unable to meet diverse QoS requirements. This lack of flexibility leads to a significant decline in QoS performance. Existing experimental results show that controller failures have a major impact on the QoS of SD-WANs. Specifically, controller failures weaken the network's ability to maintain optimal performance, highlighting the necessity of designing elastic routing strategies under controller failures.

[0004] Existing research has proposed enhancing network performance (e.g., restoring path programmability or load balancing) by intelligently remapping offline flows / switches to online controllers. However, these methods may not guarantee satisfactory QoS performance for the following reasons. First, the requirements of data plane QoS are ignored. Existing methods usually do not consider the personalized QoS requirements of each flow in the data plane. Second, control plane QoS is also overlooked. The QoS of the control plane (mainly related to the efficiency of routing policy updates) is not incorporated into the optimization objective. Specifically, the propagation delay between the controller and the switch also greatly affects the timeliness of policy updates. Summary of the Invention

[0005] In view of this, the present invention provides a software-defined wide area network elastic routing method based on quality of service perception, which can jointly maintain the quality of service of the data plane and the control plane under the failure of the controller.

[0006] A software-defined wide area network elastic routing method based on quality of service perception provided by the present invention includes the following steps:

[0007] Step 1: Establish a QoS-aware flow remapping and routing model as shown in the following formula:

[0008]

[0009] where y lk is a binary decision variable that determines whether a flow selects a path; θ l is the throughput of the flow, D lk is the propagation delay of the path; z nml is a binary decision variable that determines whether an offline flow is remapped to an online controller at the switch; G nm is the propagation delay between the switch and the controller; ω1 and ω2 are weight factors, L is the total number of flows in the network, K is the total number of paths for each flow, T is the total number of offline flows, M is the total number of controllers, and N is the total number of switches; the model includes total control capacity constraints, path selection constraints, link capacity constraints, and flow remapping constraints;

[0010] After relaxing y lk and solving the model, a solution set is obtained

[0011] Step 2: Traverse the selection strategy. When the flow f corresponding to the strategy l simultaneously satisfies the total control capacity constraint and the path selection constraint, set y lk = 1. When it does not satisfy simultaneously, let the flow f l select its initial path;

[0012] Step 3: Traverse the y with a value of 1 lk . For the offline flows with non-initial paths among them, form a set χ from all the switches through which they pass;

[0013] Step 4: Traverse the switches s in the set χ i , select an online controller C with a remaining control capacity greater than zero m , remap the offline flow passing through the switch s i to the online controller C m , update the remaining control capacity of C m , and end this process.

[0014] Furthermore, the total control capacity is constrained as shown in the following equation:

[0015]

[0016] where x l is a binary decision variable that determines whether the offline flow f l is mapped to the online controller; β nl is a binary decision variable that determines whether the flow f l flows through the switch s n ; is the remaining control capacity of the controller C m .

[0017] Furthermore, the path selection constraint is that each flow needs to and can only select one path. When the offline flow f l is selected to be remapped to the online controller, it can select one of its available paths {p l2 , …, p lK}, as shown in the following equation:

[0018]

[0019] where x l is a binary decision variable that determines whether the offline flow f l is mapped to the online controller.

[0020] Furthermore, the link capacity constraint is as shown in the following equation:

[0021]

[0022] θ l ·ζ lkq ·y lk ≤ θ q

[0023] where V l is the traffic volume of the flow f l , ζ lkq is the relationship between the path and the link, Cap q is the upper limit of the traffic volume that the link e q can bear, θ q is the utilization rate of the actual traffic volume on the link e q , θ l represents the throughput of the flow f l .

[0024] Furthermore, the flow remapping constraint is as shown in the following equation:

[0025]

[0026] where xl is a binary decision variable that determines whether the offline flow f l is mapped to the online controller; β nl is a binary decision variable that determines whether the flow f l flows through the switch s n ; z nml is the required control resource, and is the remaining control capacity of the controller C m .

[0027] Further, in step 4, the switch s in the set χ is traversed i and an online controller C with a remaining control capacity greater than zero is selected m in the following way: the online controllers are sorted in ascending order of the physical distance from the switch s i , and the online controllers C with a remaining control capacity greater than zero are selected in the order from the front to the back according to the sorting m .

[0028] Further, the way to update the remaining control capacity of C m in step 4 is: let the remaining control capacity be updated to the difference between the original remaining control capacity and the set value

[0029] Further, the set value is 1

[0030] Beneficial effects:

[0031] By establishing a joint optimization problem model, the present invention realizes QoS-aware flow routing in the data plane while minimizing the control delay in the control plane. Since this problem has high complexity, a heuristic QoS-aware elastic routing algorithm is further proposed to achieve efficient flow re-mapping and routing under controller failure. Experimental results based on real traffic data in a real network topology show that the method proposed in the present invention significantly improves network throughput and effectively reduces network delay and control delay compared with the current methods BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a schematic diagram of the processing flow of a QoS-aware software-defined wide area network elastic routing method provided by the present invention DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The following describes the present invention in detail with reference to the accompanying drawings and by way of examples

[0034] A QoS-aware software-defined wide area network elastic routing method provided by the present invention has a processing flow as Figure 1 shown and specifically includes the following steps

[0035] Step 1. Establish a QoS-aware flow remapping and routing model to describe the mapping relationship between offline flows and online controllers in the network, as well as the routing strategy of flows. The model is shown in Equation (1):

[0036]

[0037] where ω1 and ω2 are weight factors, x l is a binary decision variable that determines whether the offline flow f l is mapped to an online controller; represents the remaining control capacity of controller C m ; y lk is a binary decision variable that determines whether flow f l should select path p lk ; V l represents the traffic volume of flow f l ; ζ lkq represents the relationship between the path and the link (ζ lke = 1 means that the path p l of flow f lk passes through link e q ); Cap q represents the upper limit of link e q ; θ q represents the utilization rate of the actual traffic on link e q ; θ l represents the throughput of flow f l ; D lk represents the propagation delay required for path p lk ; z nml is a binary decision variable that determines whether the offline flow f l should be remapped to the online controller C n at switch s m ; G nm represents the propagation delay between switch s n and controller C m . L is the total number of flows, K is the total number of paths for each flow, T is the total number of offline flows in the network, M is the total number of controllers, and N is the total number of switches.

[0038] Generally, it is assumed that an SD-WAN is divided into H domains, and each domain is controlled by an SDN controller. These H controllers are distributed in H different geographical locations, and each controller controls all the SDN switches within its domain. When a controller fails, those controllers that are still working properly are called online controllers. Specifically, the set of online controllers is represented as C = {C1,…,C m ,…,C M}. The failed controllers {CM+1 , …, C H [[ID=3}} The switch controlled by … will go offline and become an offline switch. The set of offline switches is represented as S = {s1, …, s i , …, s N [[ID=7}}. In SD-WAN, there are a total of L flows between every two switches. The set of flows is F = {f1, f2, …, f l , …, f L [[ID=11}}, where {f1, …, f T [[ID=13}} are the offline flows passing through the offline switches, and the remaining flows, i.e., {f T+1 , …, f L [[ID=17}}, are called normal flows. If the flow f l passes through the switch s n , it is denoted as β nl = 1, otherwise β nl = 0. The set of links is E = {e1, e2, …, e q , …, e Q [[ID=29}}. The flow f l has K pre-configured paths, and this set of pre-configured paths is represented as P l = {p l1 , p l2 , …, p lk , …, p lK [[ID=41}}.

[0039] The constraint conditions in the QoS-aware flow remapping and routing model established by the present invention are specifically as follows:

[0040] 1. Total control ability constraint. When the controller fails, the controller cannot manage the offline flows. Therefore, these offline flows cannot be flexibly re-routed to alternative paths to adapt to potential traffic changes, which may lead to a serious reduction in QoS. To solve this situation, the online controller should take over the control of the offline flows to improve the programmability and flexibility of the network. However, the control resources of the controller are limited, so it can only handle a certain number of flows. The control resources required by the controller are determined by the overhead of processing all flow requests within its domain.

[0041] Quantify the available control resources of the controller based on the number of flows that the controller can effectively process without introducing additional delays (e.g., long-tail delays). It should be noted that the control resources required by the controller should not exceed the upper limit of the control ability of the controller, otherwise serious cascading failures may occur. Due to the limited available control resources, only some of the offline flows can be remapped to the online controller, as shown in formula (2):

[0042]

[0043] 2. Path selection constraint. Each flow needs to and can only select one path, as shown in formula (3):

[0044]

[0045] For flow f l , assuming its initial path is p l1 . Each offline flow can only be rerouted to other available paths (paths other than the initial routing path) when it is remapped to other online controllers and regains the lost network programmability. Therefore, the path availability constraint ensures that offline flow f l can select its available paths {p l2 ,…,p lK} when it is selected to be remapped to an online controller, which can be expressed as formula (4) below:

[0046]

[0047] 3. Link capacity constraint. The total traffic load on each link can be calculated as the sum of the traffic demands of all flows passing through that link. Each link has its capacity, and the total traffic load should not exceed this capacity. The link capacity constraint can be expressed using formula (5):

[0048]

[0049] The throughput of each flow is limited by the smallest ratio θ q among all the links it passes through, as shown in formula (6):

[0050]

[0051] 4. Flow remapping constraint. An offline flow may pass through multiple offline switches. To ensure that this offline flow is remapped to an online controller and can be further rerouted to other paths, all the offline switches passed through by this flow should be remapped to online controllers, as shown in formula (7):

[0052]

[0053] Meanwhile, only a limited number of offline flows can be remapped to online controllers to maintain the normal operation of these controllers. Therefore, to ensure that the required control resources do not exceed the control capacity limit of each online controller, as shown in formula (8):

[0054]

[0055] In the prior art, a typical solution for integer programming is to use an integer program optimization solver to obtain the optimal solution of the above QoS-aware flow re-mapping and routing model. However, with the increase in network scale, the solution space may increase significantly, and it may take a long time or even be impossible to find a feasible solution. For this reason, the present invention proposes a heuristic algorithm to solve this problem, thereby improving performance and time complexity.

[0056] Step 2: After relaxing the binary decision variable y lk Solve the QoS-aware flow re-mapping and routing model to obtain a solution set where, is a set formed by the k path selection relationships of f l . The relationships include the selection between the flow and the path and the probability of this selection. Sort the elements in the set in descending order of probability. Establish a set F * and set it to an empty set.

[0057] Step 3: Select a policy from the set to obtain the corresponding flow number l and path number k. where, the selection method of the switch is selected according to the sorting order in the set

[0058] . of.

[0059] Step 4: If the flow f l is not in the set F * , then add the flow f l to the set F * , and execute Step 5; if the flow f l is in the set F * , then execute Step 3.

[0060] Step 5: If both and are satisfied, then it is a feasible path selection policy. Let y lk = 1, and the flow f l selects the path p lk according to the policy, and execute Step 6; if both and are not satisfied, then find the initial path p l of the flow f l1 , let the flow f l continue to select this initial path, let y l1 = 1, and execute Step 6.

[0061] Step 6: If the testing of the policies in the set is not all completed, then execute Step 3; otherwise, exit the loop and execute Step 7.

[0062] Step 7: For all y with values assigned as 1 lk , obtain the corresponding flow number l and path number k; if all y lk tests are completed, exit the loop and end this process. Remap the offline flows that need to change paths. Only after these offline flows are remapped to other online controllers can the paths be changed.

[0063] Step 8: If the flow belongs to an offline flow and k≠1, it means that this path is not the initial path of the flow. Find all the switches that the offline flow passes through, denoted as set χ, and then execute Step 9; if the flow does not belong to an offline flow, execute Step 7.

[0064] Step 9: Select switch s from set χ in sequence i . At the same time, sort all the online controllers in set C in ascending order of their physical distance from this switch.

[0065] Step 10: Select controller C from set C m .

[0066] Among them, the selection method of the controller is selected according to the sorting order in set C.

[0067] Step 11: If let z nml =1, remap the offline flow f n passing through switch s l to online controller C m , update the remaining control capacity of the controller and execute Step 12; if not satisfied and not all the controllers in set C have been tested, execute Step 10.

[0068] Step 12: If all the switches in set χ have been tested, execute Step 7; if not all the tests have been completed, execute Step 9;

[0069] The performance of this embodiment was evaluated through experimental simulation. The method proposed by the present invention can increase the average throughput by 1.71%, reduce the average delay by 46.65%, and reduce the average control delay by 62.74%.

[0070] In summary, the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A software-defined wide area network elastic routing method based on quality of service perception, characterized in that Including the following steps: Step 1, establish a QoS-aware flow remapping and routing model as shown in the following formula: where y lk is a binary decision variable that determines whether the flow selects a path; θ l is the throughput of the flow, D lk is the propagation delay of the path; z nml is a binary decision variable that determines whether the offline flow is remapped to the online controller at the switch; G nm is the propagation delay between the switch and the controller; ω1 and ω2 are weight factors, L is the total number of flows in the network, K is the total number of paths for each flow, T is the total number of offline flows, M is the total number of controllers, and N is the total number of switches; the model includes total control capacity constraints, path selection constraints, link capacity constraints, and flow remapping constraints; Set y lk Solve the model after relaxation to obtain a solution set Step 2, traverse the selection strategy. When the flow f corresponding to the strategy l simultaneously satisfies the total control ability constraint and the path selection constraint, let y lk = 1. When not simultaneously satisfied, let the flow f l select its initial path; Step 3: Traverse the y with a value of 1 lk , for the offline flows with non-initial paths among them, form a set χ with all the switches through which they flow; Step 4. Traverse the switches s in the set χ i , and select the online controller C with remaining control capacity greater than zero m . Remap the offline flows passing through the switch s i to the online controller C m , and update the remaining control capacity of C m , and end this process.

2. The software-defined wide area network elastic routing method according to claim 1, characterized in that The total control capacity constraint is as shown in the following formula: where x l is a binary decision variable that determines whether the offline flow f l is mapped to the online controller; β nl is a binary decision variable that determines whether the flow f l flows through the switch s n ; is the remaining control capacity of the controller C m .

3. The software-defined wide area network elastic routing method according to claim 1, wherein The path selection constraint is that each flow needs to and can only select one path. When the offline flow f l is selected to be remapped to the online controller, it can select its available paths {p l2 ,…,p lK}, as shown in the following formula: where x l is a binary decision variable that determines whether the offline flow f l is mapped to the online controller.

4. The software-defined wide area network elastic routing method according to claim 1, characterized in that The link capacity constraint is as shown in the following formula: θ l ·ζ lkq ·y lk ≤θ q Among them, V l is the flow rate of flow f l , ζ lkq is the relationship between the path and the link, Cap q is the upper limit of the flow that link e q can withstand, θ q is the utilization rate of the actual flow on link e q , θ l represents the throughput of flow f l .

5. The software-defined wide area network elastic routing method according to claim 1, wherein The flow remapping constraint is as shown in the following formula: where x l is a binary decision variable that determines whether the offline flow f l is mapped to the online controller; β nl is a binary decision variable that determines whether the flow f l flows through the switch s n ; z nml is the required control resource, and m is the remaining control capacity of the controller C 6. The software-defined wide area network elastic routing method according to claim 1, wherein In step 4, traverse switch s in set χ i Select online controller C with remaining control capacity greater than zero m , in the following way: according to the physical distance between the online controller and switch s i sort them in ascending order, and select online controller C with remaining control capacity greater than zero in the order from front to back according to the sorting m .

7. The software-defined wide area network elastic routing method according to claim 1, wherein The way to update C described in step 4 m for the remaining control ability is: let the remaining control ability be updated to the difference between the original remaining control ability and the set value.

8. The software-defined wide area network elastic routing method according to claim 7, characterized in that, The set value is 1.

Images