A collaborative decision-making-based SDN switch migration scheme

By coordinating decision-making among controllers, sharing network status information, and formulating switch migration strategies, the problem of load imbalance in distributed multi-controller scenarios is solved, achieving load balancing and efficient switch migration, thereby improving network response speed and resource utilization.

CN117395201BActive Publication Date: 2026-11-10CHANGSHA UNIVERSITY OF SCIENCE AND TECHNOLOGY
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202310669673.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-07
Publication Date
2026-11-10
Estimated Expiration
2043-06-07

AI Technical Summary

Technical Problem

In distributed multi-controller deployment scenarios, existing switch migration solutions are prone to causing unbalanced controller load, leading to network paralysis or low resource utilization, and the migration effect is not ideal.

Method used

By coordinating decision-making among controllers, sharing real-time network status information, jointly formulating switch migration strategies, optimizing load balancing and response speed, and employing a collaborative decision-making switch migration algorithm, the migration target is selected based on controller load and distance, reducing post-migration flow installation latency.

Benefits of technology

It achieves load balancing in the control plane, improves network response speed and data transmission efficiency, reduces load imbalance and migration frequency in the control plane, and improves system stability and resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117395201B_ABST
    Figure CN117395201B_ABST
Patent Text Reader

Abstract

This invention designs an SDN switch migration scheme based on collaborative decision-making. When a controller experiences overload or needs to hibernate, nearby controllers capable of receiving load form a collaborative decision domain. Controllers within this domain jointly negotiate and formulate a switch migration strategy based on their own load capacity and distance from the switch, until the overloaded controller is no longer overloaded or all switches under the hibernation controller have been migrated. While ensuring appropriate controller load, this scheme aims to maximize the control plane's response speed, load balancing, and overall operating efficiency, achieving a robust, efficient, and environmentally friendly control plane.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of software-defined networking, and more specifically to an efficient switch migration scheme based on collaborative decision-making. Background Technology

[0002] Software-defined networking (SDN) is considered one of the most promising development models for the future Internet because of its advantages such as programmability and centralized management, separating network forwarding and control functions. The SDN control plane typically employs a distributed architecture, deploying multiple controllers to manage network domains such as... Figure 1 As shown. The control plane is primarily responsible for generating flow rules and distributing them to the data plane switches, thereby guiding the switches' packet forwarding behavior. When a switch receives a packet, it first parses its header information, extracts the matching field, and then searches the flow table to find a rule that the packet can match. If the packet correctly matches a rule, the switch processes the packet according to that rule. If no rule is found, the switch needs to request the controller using the Packet-in message in the OpenFlow protocol. When the controller receives the Packet-in message, it determines how to process the packet based on the global network view, sends a Flow-mod message to the switch, and installs a flow table entry. The switch then processes the packet according to this flow table entry.

[0003] Network traffic tends to peak during the day but decreases at night. Furthermore, traffic is also characterized by bursts; when traffic surges rapidly, excessive Packet-in messages from switches can lead to excessively long controller response times. This burstiness and uncertainty can easily cause load imbalances among controllers. Prolonged overload operation of controllers can easily lead to malfunctions and network paralysis. Simultaneously, when overall control plane resource utilization is low, controller loads can be consolidated and controllers can be put into hibernation to save energy. Dynamic switch migration can shorten controller response time by migrating some switches from overloaded controllers, and put some controllers into hibernation when overall control plane resource utilization is low, thereby improving control plane resource utilization and reducing the average controller response time. Therefore, timely switch migration is crucial for improving the scalability and reliability of the SDN control plane. Currently, most existing switch migration schemes migrate several switches from the source controller to a single target controller when the controller is overloaded or lightly loaded. This can easily lead to an unbalanced load on the target controller after migration, or even the inability to find a target controller to take over the migrated switches, as detailed below:

[0004] Option 1 utilizes deep learning to predict future load data based on the controller's historical load data, thus anticipating the timing of controller overload and enabling proactive switch migration. First, based on the current load traffic and predicted future traffic, a set of switches to be migrated is selected from the overload controller. After migration, the overload controller's threshold is reduced below the overload threshold. For each switch in the selected migration set, one or more slave controllers can accept it, provided that the load of the selected migrated switch is less than 10% of the target controller's remaining processing capacity.

[0005] Scheme 2 proposes a controller load balancing mechanism based on distributed decision-making, implemented in three phases. First, network load information is collected periodically. When a controller becomes overloaded, a distributed migration decision domain is constructed by considering the load status of its neighboring controllers. This migration decision domain is a set of controllers consisting of the overloaded controller and all its non-overloaded neighboring controllers. Then, the switch with the highest request rate and furthest distance from the overloaded controller is selected as the target switch for migration, considering the costs of data collection, switch migration, and state synchronization. Finally, a migration clock model is established to complete the switch migration and the role transition of the controller.

[0006] However, the existing switch migration solutions mentioned above have the following drawbacks:

[0007] [Option 1] Select a set of switches to be migrated based on the predicted controller load. The selected set of switches can reduce the overloaded controller load value below the overload threshold, and select a single target controller as the migration object based on the remaining processing capacity of the controller. However, when the average load level of the controller is high, this method may result in a large load difference in the control plane after the migration is completed.

[0008] [Option 2] Periodically collect network load information, relocate switches with high request rates that are furthest from the overload controller, and select the target controller by considering the costs of data collection, switch migration, and controller state synchronization. If the average controller load is too high, relocating switches with high request rates may lead to frequent switch migrations, resulting in unsatisfactory migration performance.

[0009] Based on the above, the present invention provides a switch migration scheme based on collaborative decision-making. Summary of the Invention

[0010] The technical problem to be solved by this invention is to design an efficient switch migration method in a distributed multi-controller deployment scenario, which performs switch migration when the controller is overloaded or needs to hibernate, so as to achieve a green and efficient control plane.

[0011] To address the aforementioned technical problems, this invention proposes a switch migration scheme based on collaborative decision-making, employing the following technical solution: designing a switch migration algorithm based on collaborative decision-making, whereby controllers jointly negotiate and formulate a switch migration strategy when the controller experiences overload or needs to hibernate.

[0012] The switch migration algorithm is based on the following idea:

[0013] Within the same network domain, all controllers share their real-time network status information via east-west interface protocols, thereby establishing a global network view. When a controller experiences overload or needs to hibernate, based on a comprehensive understanding of the network's operational status, the controllers collaboratively negotiate and formulate a switch migration strategy. This strategy aims to maximize control plane response speed, load balancing, and overall operational efficiency while ensuring appropriate controller load, thereby achieving a robust, efficient, and environmentally friendly control plane. Specific switch migration procedures for different scenarios are as follows:

[0014] The switch migration process under controller overload scenarios is as follows: Figure 2 As shown in the diagram. In this scenario, there are significant load differences between controllers. An overloaded controller, due to excessive load, causes its managed switches to be unable to respond quickly, and its load should be migrated to a lightly loaded controller in a timely manner. First, the overloaded controller informs nearby controllers of its overload status and sends a switch migration request. Nearby controllers choose to accept or reject the request based on their own load levels. All nearby controllers that accept the request constitute a collaborative decision domain. Subsequently, each controller in the decision domain finds the nearest acceptable switch based on its own load, forming a candidate migration pair for the controller and switch, thus constructing a candidate migration pair set for the entire decision domain. This process is repeated, selecting the nearest candidate migration pair from the candidate migration set and executing the switch migration, then updating the candidate migration pair set until the overloaded controller is no longer overloaded or the candidate migration pair set is empty.

[0015] The switch migration process in controller hibernation scenarios is as follows: Figure 3As shown in the diagram, in this scenario, most controllers have a light load, the average load level of the control plane is low, and computing and storage resources are not fully utilized. In this case, some controllers can be put into hibernation to reduce the energy consumption of the control plane. First, the lightly loaded controllers send switch migration requests to nearby controllers. Nearby controllers choose to accept or reject the request based on their own load levels; all controllers that accept the request form a collaborative decision domain. Then, based on the acceptability of other controllers in the domain, the lightly loaded controller finds the nearest accepting controller for each switch it manages, forming candidate switch controller migration pairs, and thus constructing a candidate migration set for these pairs. Each time, the nearest switch controller pair is selected from the candidate migration set, the switch migration is executed, and the candidate migration set is updated until all switches managed by the lightly loaded controller have been migrated.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. This invention proposes a switch migration scheme based on collaborative decision-making. It calculates the maximum load capacity that a controller can withstand and the average load level of other controllers, so that the load levels of all controllers are relatively balanced after the switch migration is completed, thereby improving the load balancing degree of the control plane.

[0018] 2. When formulating the switch migration scheme, this invention focuses on the distance between the target controller and the migrated switch, so as to minimize the flow installation delay after the switch migration and effectively improve the network data transmission efficiency.

[0019] 3. Compared with the prior art solution one, the present invention addresses the shortcomings of solution one by having controllers jointly negotiate and formulate a switch migration plan to migrate several switches in the source controller to several target controllers, thereby effectively reducing the load imbalance of the control plane.

[0020] 4. Compared with the prior art solution 2, the present invention addresses the shortcomings of solution 2 by having the controller calculate the acceptable load based on the expected load after the switch migration, thus making the load after the migration appropriate and reasonable, effectively reducing the number of subsequent switch migrations and improving system stability. Attached Figure Description

[0021] To more clearly illustrate the embodiments and technical solutions of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below, wherein:

[0022] Figure 1 This is a schematic diagram of SDN distributed multi-controller deployment in an embodiment of the present invention;

[0023] Figure 2This is a schematic diagram of a controller overload scenario in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the controller sleep scenario in an embodiment of the present invention;

[0025] Figure 4 This is a flowchart of the switch migration algorithm in an embodiment of the present invention;

[0026] Figure 5 This is a flowchart of the collaborative decision domain construction in an embodiment of the present invention;

[0027] Figure 6 This is a flowchart of the switch migration scheme formulation in an embodiment of the present invention;

[0028] Figure 7 This is a flowchart of the candidate migration set update process in an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of switch migration execution in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0032] An abstract structural model of the SDN network in this embodiment is constructed. To facilitate the description of the switch migration scheme, graph theory is used to establish the SDN network model. It is assumed that the SDN network topology is represented by an undirected graph G = (V, E), where V represents the set of nodes and E represents the set of links. The node set V contains M controllers and N switches, and the controller set is C = {c1, c2, ..., c...}. M The set of switches is S = {s1, s2, ..., s}. N Assuming each controller can manage multiple switches, and each switch is managed by only one controller, then controller C... i and switch S j The mapping relationship between them can be represented as:

[0033]

[0034] In an SDN network, the controller's load primarily stems from topology management of the managed network domain, interaction responses from managed switches, and state synchronization with other controllers. Network topology management mainly involves periodically collecting link-state information from managed switches. Assuming d is the number of hops between network devices, and f is the frequency at which the controller collects link-state information, then controller C... i The topology management load can be defined as:

[0035]

[0036] Switch interaction responses primarily involve processing packet-in messages (flow installation requests) sent by the switches under its jurisdiction, as well as exchanging and synchronizing information with the switches. Assuming the information exchange frequency between the controller and the switches is μ, and the switch S... j The rate at which Packet-in messages are sent is λ j Then controller C i The interactive response load can be defined as:

[0037]

[0038] Controller state synchronization primarily involves periodically synchronizing with other controllers in the network to establish and maintain a global network view. Assuming the state synchronization frequency between controllers is η, then controller C... i State-synchronized load can be defined as:

[0039]

[0040] Based on the three types of controller loads mentioned above, any controller C i The load can be expressed as:

[0041]

[0042] Assumption For controller C i The maximum load that controller C can withstand is... i The load state can be defined as:

[0043]

[0044] Where α and β are the controller's sleep and overload state coefficients, respectively, and their values ​​can usually be set to [0.10, 0.40] and [0.75, 0.95], respectively.

[0045] When the controller is under overload or light load, a switch migration is required to offload a certain amount of load, allowing the controller to operate at normal load or enter a sleep state. This invention will migrate the controller C... i The minimum migration load is defined as:

[0046]

[0047] When a controller is overloaded, it sends a switch migration request to other nearby controllers. Upon receiving the request, a nearby controller calculates its own load capacity. This patent comprehensively considers the maximum load a controller can withstand and the average load of the control plane, and adjusts the controller C... i The expected load after switch migration is defined as follows:

[0048]

[0049] Where θ is the weight parameter. This represents the average load on the control plane. Furthermore, the controller C... i The acceptable load capacity is defined as:

[0050]

[0051] After receiving a switch migration request from another controller, the controller selects whether to accept or reject the request based on the available load and responds accordingly.

[0052] like Figure 4 As shown in the figure, one embodiment of the present invention provides a switch migration algorithm based on collaborative decision-making, the specific process of which is as follows:

[0053] The controller predicts the load for the next time period based on its historical load data and anticipates the load status for that period. If it is in an overload or light-load state, it calculates the minimum migration load and sends a switch migration request to establish a collaborative decision domain. If the collaborative decision domain is successfully established, all controllers in the domain jointly negotiate and formulate a migration plan. If the switch migration plan is successfully formulated, the switch migration process is executed. If a controller no longer manages a switch, it goes into hibernation. If the collaborative decision domain fails to be established and the controller is in an overload state, a system alarm is issued, requesting the addition of a new controller. If the controller is in a normal load state, it waits to receive switch migration requests from other controllers and calculates its own acceptable load. If the acceptable load is greater than 0, it replies with a request to accept the request, joins the collaborative decision domain, and participates in formulating and executing the switch migration plan.

[0054] Reference Figure 5-8 The above embodiments specifically include the following operations:

[0055] a. Constructing collaborative decision-making domain operations

[0056] like Figure 5The diagram illustrates the collaborative decision domain construction process. First, the migration controller sends a switch migration request to nearby controllers. Upon receiving the migration request, the nearby controller checks if its acceptable load is greater than 0. If it is, it replies with an acceptance response. If not, it replies with a rejection response. The migration controller continuously receives responses until it receives responses from all controllers or reaches a timeout. Then, it calculates the total acceptable load of all controllers that accepted the request. If the total acceptable load is greater than the minimum migration threshold, the collaborative decision domain is successfully constructed by all controllers that accepted the request. The migration controller then notifies the controllers in the domain that the collaborative decision domain construction was successful. Subsequently, the migration controller sends the load information of all switches under its jurisdiction to all controllers in the decision domain. If the total acceptable load is not greater than the minimum migration threshold, the migration controller sends a collaborative decision domain construction failure message to the controllers that accepted the request.

[0057] b. Develop and implement a switch migration plan.

[0058] like Figure 6 The diagram illustrates the switch migration plan development process. First, each controller in the collaborative decision domain is selected as the migration target. From all switches managed by the source controller, the closest and most acceptable switch is selected as the migration target, generating candidate migration pairs consisting of the target controller and the migration target switch. This creates a candidate migration set for the collaborative decision domain. Then, the closest target controller and migration target switch pair is continuously removed from the candidate migration set and moved to the migration set. The acceptable load of the target controller and the minimum migration load of the source controller are updated. If the target controller cannot accept the load, it is removed from the collaborative decision domain. Subsequently, the candidate migration set is updated. If the candidate set is empty after the update, the switch migration plan development fails. If the source controller is no longer overloaded or no longer manages switches (i.e., the minimum migration load is 0), the switch migration plan development succeeds.

[0059] c. Update candidate migration set operation

[0060] like Figure 7 The diagram illustrates the candidate migration set update process. First, all migration pairs in the entire candidate migration set are iterated over. If a candidate migration pair contains the same switch as the newly added migration pair, the controller in the candidate migration pair reselects the nearest and acceptable switch, and the switch in the candidate migration pair is replaced with this switch. Subsequently, if the target controller to be migrated has a load capacity greater than 0, the controller again selects the nearest and acceptable switch, generating a (target controller, outgoing switch) candidate migration pair, and this migration pair is added to the candidate migration set.

[0061] d. Switch migration execution operation

[0062] like Figure 8 The diagram illustrates the switch migration process, which can be broadly divided into three stages. In the first stage, the target controller's role is changed from "slave" to "equal." First, the source controller sends a start migration message to the target controller to initiate the switch migration. Then, upon receiving the message, the target controller sends a request role change message to the migration switch, instructing it to change the target controller's role to "equal." After changing the target controller's role, the migration switch sends a role change complete message to the target controller. Upon receiving this message, the target controller sends a migration ready message to the source controller, notifying it that the role change is complete.

[0063] The second phase involves inserting and deleting empty flow table entries to determine the migration timing. First, the source controller sends a Flow-mod message to the switch to be migrated to add an empty flow table entry that does not match any packets. Then, the source controller sends a Barrier-request message to the switch to confirm whether the previous Flow-mod message has been executed. After confirmation, the switch replies with a Barrier-reply message to the source controller. Next, the source controller sends a Flow-mod message to the migration switch to delete the empty flow table entry. After performing the deletion operation, the switch sends Flow-removed messages to both the source and target controllers. Finally, the source controller sends a migration complete message to the target controller. At this point, the target controller takes over control of the migration switch and receives flow installation requests from it.

[0064] In the third phase, the target controller's role is changed from equal to master. First, the target controller sends a role change request message to the switch. Then, the switch changes the controller's role back to master and sends a role change completion message to the target controller, ending the migration process.

[0065] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for migrating SDN switches based on collaborative decision-making, characterized in that, include: When the source controller is overloaded, the source controller sends a switch migration request to a nearby controller; Nearby controllers choose to accept or reject the request based on their own load capacity, and the controllers that accept the request constitute a collaborative decision domain; After the collaborative decision domain is successfully constructed, all controllers in the domain, based on their own acceptable load capacity and distance to the switches, jointly negotiate and formulate a switch migration plan until the source controller is no longer overloaded or no longer manages the switches. The formulation of the switch migration plan includes: using each controller in the collaborative decision domain as the migration target, each controller in the collaborative decision domain traverses all switches managed by the source controller and selects a switch that it can accept and is closest to; generating candidate migration pairs between the selected controller and the chosen switch; thus generating a candidate migration set for the collaborative decision domain; continuously removing the closest target controller and migration-out switch pair from the candidate migration set and moving them to the migration set; updating the acceptable load capacity of the target controller and the minimum migration-out load capacity of the source controller; if the target controller cannot accept the load, removing the controller from the collaborative decision domain; and updating the candidate migration set.

2. The SDN switch migration method based on collaborative decision-making according to claim 1, characterized in that, All controllers share their real-time network status information through the east-west interface protocol, establish a global network view, and jointly negotiate and formulate switch migration strategies.

3. The SDN switch migration method based on collaborative decision-making according to claim 1 or 2, characterized in that, The construction of the collaborative decision domain includes: the source controller sending a switch migration request to nearby controllers; nearby controllers determining whether the acceptable load is greater than 0, replying with an acceptance response if it is greater than 0, and replying with a rejection response if it is not greater than 0; the source controller continuously receives responses until it receives responses from all controllers or reaches the timeout period; calculating the total acceptable load of all controllers that accepted the request; if the total acceptable load is greater than the minimum migration load, the collaborative decision domain is successfully constructed by all controllers that accepted the request, the source controller notifies the controllers in the domain that the collaborative decision domain has been successfully constructed, and sends the load information of all switches under its jurisdiction to all controllers in the decision domain; if the total acceptable load is not greater than the minimum migration load, the source controller sends a collaborative decision domain construction failure message to the controllers that accepted the request.

4. The SDN switch migration method based on collaborative decision-making according to claim 3, characterized in that, The controller calculates the acceptable load by: calculating the average load of the control plane; and calculating its acceptable load based on the average load of the control plane and the maximum load it can withstand.

5. The SDN switch migration method based on collaborative decision-making according to claim 1, characterized in that, The updated candidate migration set includes: updated candidate migration pairs; if the target controller to be migrated can receive a load greater than 0, the controller will select the nearest and acceptable switch again, generate a candidate migration pair, and add the candidate migration pair to the candidate migration set.

6. The SDN switch migration method based on collaborative decision-making according to claim 5, characterized in that, The update of candidate migration pairs includes: traversing all migration pairs in the candidate migration set; if the switch in the candidate migration pair is the same as the switch in the newly added migration pair, then the controller in the candidate migration pair reselects the nearest and acceptable switch, and replaces the switch in the candidate migration pair with that switch.

Citation Information

Patent Citations

  • Method and device for load balancing of controller

    CN106027410A