CN110275437B - SDN network flow dominance monitoring node dynamic selection system and method thereof - Google Patents
SDN network flow dominance monitoring node dynamic selection system and method thereof Download PDFInfo
- Publication number
- CN110275437B CN110275437B CN201910489576.1A CN201910489576A CN110275437B CN 110275437 B CN110275437 B CN 110275437B CN 201910489576 A CN201910489576 A CN 201910489576A CN 110275437 B CN110275437 B CN 110275437B
- Authority
- CN
- China
- Prior art keywords
- node
- flow
- monitoring
- component
- forwarding path
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention discloses a dynamic selection system and a dynamic selection method for an SDN network flow dominant monitoring node.A SDN control plane comprises a forwarding calculation module, a node updating module and a path prediction module; the SDN data layer comprises a network resource module; the dynamic selection method comprises two time sequence stages of dominant monitoring node pre-screening and dominant monitoring node dynamic updating, wherein the dominant monitoring node is only operated when the system is in cold start, and the dominant monitoring node is adaptively operated in a closed-loop self-feedback mode after the system is started. The invention is sent from the angle of selecting the monitoring node, and the switch with the densest flow traversal is preferentially selected as the flow monitoring node, so that the purposes of improving the non-redundancy rate of flow collection and reducing the flow monitoring overhead are achieved while the maximum capture of flow statistical information is realized.
Description
Technical Field
The invention belongs to the technical field of network security, and particularly relates to a dynamic selection system and method for SDN network flow advantage monitoring nodes.
Background
As a new network architecture, software Defined Networking (SDN) enables centralized management of network resources by decoupling the control plane and data plane of network devices. The mode of separating the control rights of the IP network equipment and uniformly managing the network by the controller can shield the difference of the bottom heterogeneous network and effectively reduce the dependence of network management on the bottom network equipment. Although the SDN has a significant refinement effect on improving resource utilization efficiency and improving network management, in order to implement such a centralized management manner, the SDN controller needs to obtain a global network view that can reflect the overall state of the current network in time. With the view, the SDN controller can implement decisions such as network resource allocation, traffic forwarding path selection, network anomaly analysis and the like.
Through network traffic monitoring, the SDN controller may obtain a relatively clear view of the global network state. However, due to the randomness of network topology changes and the dynamics of the traffic itself, timely acquisition of a global network view requires that the SDN controller must be able to apply continuous traffic monitoring to the network devices. This demanding requirement for maximizing traffic capture typically increases the communication overhead and information processing delay of the SDN network, making it unusable on the premise that the monitoring resources are limited. In fact, after the SDN controller and the switch establish a connection through the secure channel, all monitoring traffic is forwarded to the SDN controller during the continuous process of collecting traffic information from the corresponding switch in real time, which usually causes congestion of the secure channel under the condition of limited bandwidth, so that the corresponding control decision cannot be dispatched to the switch in time. Meanwhile, as the monitoring traffic information is uninterruptedly converged to the SDN controller to wait for processing, the processing delay of the SDN controller on the non-traffic monitoring information is inevitably increased.
In order to capture as much traffic information as possible with as little communication overhead and information processing delay as possible, SDN traffic monitoring schemes based on overhead weakening techniques such as traffic sampling, traffic passive Push (Push), and traffic active retrieval (Pull) are currently emerging. (1) Most discussed in SDN traffic monitoring based on traffic sampling is the use of sketches to allocate traffic monitoring resources and manage traffic monitoring tasks; the traffic sampling only obtains partial network characteristic traffic, thereby reducing the traffic monitoring overhead, but simultaneously, all the conditions of traffic change cannot be obtained, so that the network global view is not completely and accurately obtained. (2) In the Push-based SDN flow monitoring, an SDN controller can passively receive flow statistical information pushed by a switch to grasp the condition of an active flow, and can obtain flow change information as much as possible with less overhead under normal conditions; however, the method has the problems of harsh requirements on software and hardware of the switch and still generates larger communication and calculation overhead when the flow rate changes frequently, and is rarely adopted at present. (3) In the Pull-based SDN traffic monitoring, an SDN controller actively retrieves traffic information by sending a flow statistics collection request to a switch; the method has the advantages of controllable flow capture scale of a sampling method and sensitive flow change of a Push method, and avoids the harsh requirements of the Push method on the software and hardware of the switch; even in the case of frequent traffic changes, as much traffic monitoring information as possible can be captured with acceptable communication overhead and information processing delay through Flow Statistics Collection (FSC) requests.
Although Pull-type technical solutions are most concerned at present, the fact that network traffic distribution generally complies with the "20/80" rule is not fully considered in implementation of such solutions, and a bottleneck is faced for further reducing the SDN traffic monitoring information redundancy rate and monitoring overhead.
Disclosure of Invention
The invention aims to: the invention aims to solve the defects in the prior art and provides a dynamic selection system and a method for monitoring nodes with SDN network flow advantages. The method for dynamically selecting the dominant monitoring node innovatively provided by the invention comprises the steps of pre-screening and dynamic updating of the dominant monitoring node.
The technical scheme is as follows: the invention discloses a dynamic selection system for SDN network flow advantage monitoring nodes, which comprises a control plane and a data plane; the data plane comprises a network resource module, and the network resource module sends the collected network topology information and the updated available bandwidth and residual bandwidth threshold value to the control plane in the form of a data packet through an OpenFlow1.3 protocol; the control plane comprises a forwarding calculation module, a node updating module and a path prediction module, wherein the forwarding calculation module calculates an optimal forwarding path set according to information sent by the data plane and sends the optimal forwarding path set to the node updating module; according to the optimal forwarding path set, the node updating module calculates the frequency of node flow, and pre-screens out an advantageous monitoring node set; according to the superior monitoring node provided by the node updating module, the path prediction module predicts a new forwarding path in a future period by using the monitored real forwarding path set, and sends a prediction result to the node updating module to update the monitoring node again.
Further, the network resource module comprises a network information sensing component and a bandwidth state updating component, the network information sensing component sends the traffic change information to the bandwidth state updating component, and then the network information sensing component and the bandwidth state updating component respectively send the network topology information, the available bandwidth and the residual bandwidth threshold value information to the forwarding calculation module; the forwarding calculation module comprises a shortest path calculation component and a forwarding path calculation component, the shortest path calculation component calculates front k shortest paths between a source node and a destination node according to network topology information and sends the result to the forwarding path calculation component, and the forwarding path calculation component calculates an optimal forwarding path set according to the front k shortest paths, an available bandwidth and a residual bandwidth threshold value and sends the optimal forwarding path set to the node updating module; the node updating module comprises a node flow frequency calculating component and an advantage monitoring node management component, wherein the node flow frequency calculating component calculates a high-frequency node according to an optimal forwarding path set during cold start, or gives node frequency updating information according to a potential forwarding path set during non-cold start, and then sends the node frequency updating information to the advantage monitoring node management component; the advantage monitoring node management component pre-screens out the advantage monitoring nodes according to the flow frequency of the high-frequency nodes during cold start, or updates the nodes with low flow polling communication cost into the advantage monitoring nodes according to node frequency updating information during non-cold start, and then sends the advantage monitoring nodes to the path prediction module; the path prediction module comprises a flow forwarding monitoring component, a forwarding path management component and a forwarding path prediction component, wherein the flow forwarding monitoring component sends a real forwarding path set monitored by the dominant monitoring node in the period to the forwarding path management component, the forwarding path management component provides all the existing real forwarding path sets as historical forwarding path sets to the forwarding path prediction component, the forwarding path prediction component sends the predicted potential forwarding path sets to a node flow frequency calculation component in the node updating module in a data packet mode, and finally the dominant monitoring node used for monitoring the flow in the next period is calculated by the dominant monitoring node management component to form a closed-loop operation system.
The network information perception component, the bandwidth state updating component, the shortest path calculation component, the forwarding path calculation component, the node flow frequency calculation component, the advantage monitoring node management component, the flow forwarding monitoring component and the forwarding path management component all use an OpenFlow protocol and are implemented in a Ryu controller, and the forwarding path prediction component is implemented in a machine learning model Ryu controller.
The invention also discloses a dynamic selection method of the SDN network flow dominant monitoring node dynamic selection system, which mainly comprises the steps of dominant monitoring node pre-screening and dynamic updating, and the specific steps are as follows:
(1) The network information perception component respectively sends the topology information and the flow change information of the bottom physical network to the shortest path calculation component and the bandwidth state updating component;
(2) The shortest path calculation component calculates the first K shortest path sets according to the network topology information and sends the results to the forwarding path calculation component;
(3) The bandwidth state updating component sends the available bandwidth and the residual bandwidth threshold value to the forwarding path calculation component in the form of a data packet according to the traffic change information;
(4) The forwarding path calculation component calculates an optimal forwarding path set by combining the information obtained in the step (2) and the step (3) and sends the result to the node flow frequency calculation component;
(5) The node flow frequency calculation component sends the result to the dominant monitoring node management component by calculating and screening high-frequency nodes on the forwarding path;
(6) The advantage monitoring node management component preliminarily screens out the advantage monitoring nodes according to the node frequency and sends the advantage monitoring nodes to the flow forwarding monitoring component;
(7) The flow forwarding monitoring component sends the real flow forwarding path set of the current period monitored by the dominant monitoring node to the forwarding path management component;
(8) The forwarding path prediction component predicts a potential forwarding path set according to a historical forwarding path set provided by the forwarding path management component and sends the potential forwarding path set to the node flow frequency calculation component;
(9) The node flow frequency calculation component updates the node frequency according to the potential forwarding path set, and the result is provided to the dominant monitoring node management component in the form of node frequency update information;
(10) And (4) updating the advantageous monitoring node set by the advantageous monitoring node management component according to the node frequency updating information, and performing flow forwarding monitoring in the next period in the step (7).
In the process, firstly, based on the shortest path set of hop count and link information such as available bandwidth, a forwarding path selection algorithm based on bandwidth is utilized to calculate an optimal forwarding path set of flow, and on the basis, an advantage monitoring node set is pre-screened according to the frequency of nodes flowing through; secondly, in order to adapt to network traffic fluctuation caused by traffic peak values, user habits and the like, a neural network model based on time series in deep learning is utilized, and through learning of a historical traffic forwarding path in a previous monitoring period, the internal characteristics of sequence type traffic forwarding and routing paths are captured, the flow-through nodes of traffic forwarding are predicted, dynamic updating of an advantage monitoring node set is completed, even if traffic fluctuation occurs in a future monitoring period, for example, due to the fact that bandwidth resources are reasonably distributed to adapt to forwarding path change caused by the fact that the traffic scale is increased, follow-up traffic forwarding can still flow through the advantage monitoring node set at the maximum probability. Meanwhile, the SDN flow monitoring scheme realized based on the method is the same as most of the existing Pull-type schemes, does not need to add extra functions (such as not changing the matching structure of an OpenFlow protocol) to a switch, and is very easy to deploy to the existing SDN network system.
Further, the network information sensing component in the step (1) is an SDN controller that collects network topology related information through a Link Layer Discovery Protocol (LLDP), and stores the network topology related information in a network x; obtaining a first K shortest path sets by adopting a Yen algorithm in the step (2); the network topology related information includes mapping relationship between ports and links between switches, information of access hosts and link information.
Further, the residual bandwidth threshold in the step (3) is set by a network administrator according to the actual network scale and the flow fluctuation rule; the available bandwidth is calculated by an SDN controller in a passive measurement mode, the available bandwidth capacity of a link is set as C, the controller inquires about the number of bytes N (t) at t moment in a port counter of a switch in a time period gamma, and the currently used bandwidth B of the link is represented as:
the SDN controller calculates the available bandwidth of the entire path according to the available bandwidth of each link in each path in the K path sets, and therefore calculates the available bandwidth B' of the entire path according to (1):
B′＝min (C-B)∈K′ (C-B)，K′∈K (2)
further, the detailed steps in the step (4) are as follows:
(4.1) integrating the first K shortest paths based on the hop count and the updated available bandwidth information of the network link by the SDN controller, selecting the shortest paths in the first K shortest paths as a first forwarding path, and taking the other K-1 paths as alternative paths;
(4.2) if the available bandwidth of the forwarding link changes, namely a new active flow enters, comparing the remaining available bandwidth of the forwarding path with the required bandwidth size of the new flow, if the remaining available bandwidth of the forwarding path is larger than the required bandwidth, turning to the step (4.3), otherwise, executing the step (4.4);
(4.3) if the residual bandwidth of the link is in the range of the residual bandwidth threshold after the new flow shares the bandwidth, turning to the step (4.5), otherwise, executing the step (4.4);
(4.4) reselecting an alternative path as a forwarding path of the new traffic, and turning to (4.2);
(4.5) if the path is an alternative path, the controller needs to send a new flow table item to the switch on the path first, and then submit the switch node traversed by the new flow table item to the node flow frequency calculation component; otherwise, directly submitting the result to the node flow frequency calculation component.
Wherein, the algorithm 1 used by the optimal forwarding path calculation component:
further, in the step (5), the node flow frequency calculation component calculates the use frequency of the node related to the forwarding path of each flow through the SDN controller, selects a switch node with a higher use frequency in all paths, selects a switch node with the smallest actual monitoring cost for the traffic of a single polling switch that is not monitored, temporarily stores the nodes in an array form, and sends the nodes to the dominant monitoring node management component in a result form of a high-frequency node.
Further, in the step (8), the forwarding path prediction component constructs a network model Sequence-to-Sequence (Sequence 2 Sequence) of an Encoder-Decoder structure according to a historical forwarding path set sent by the forwarding path management component as a data set, forms the network path into a matrix in which nodes are mapped with each other, and fits and conjectures a reasonable order of the nodes in the path. Further, a network model Sequence-to-Sequence (Sequence 2 Seq) of the Encoder-Decoder structure is constructed, namely, a network path is formed into a matrix of mutual mapping among nodes, and a reasonable Sequence of the nodes in the path is fitted and conjectured, wherein the specific construction method comprises the following steps:
the source sequence and the target sequence are two network paths from a source node to a target node, and each network path comprises a set of traversal nodes from the source node to the target node; the forwarding path prediction finds out a potential forwarding path (i.e. a target sequence) meeting the characteristics of a historical forwarding path by learning and extracting the characteristics of a node sequence in the historical forwarding path according to the historical forwarding path (i.e. a source sequence) by using an encoding-decoding structure in a seq2seq model, and the details of the potential forwarding path (i.e. the target sequence) are as follows:
setting vector for source sequence and target sequence separately
It is shown that, among others,
α and β represent the lengths thereof, respectively; d s And D t Data sets respectively representing a source sequence and a target sequence, the number of elements of the data sets being k; θ, ω represents parameters and weights in the neural network, then in the encoding-decoding structure, the representation form of the encoder is:
wherein, the first and the second end of the pipe are connected with each other,
is a data set D s Is one of k elements of (a), θ is a parameter of the encoder;
for the decoder, the task is to follow the source sequence
C and history information that has been generated before, i.e. a set of forwarding paths in a history period
Generating an element N to be generated at time i i I.e., the next hop node in the predicted forwarding path. Let ω be the parameter of the decoder, which is now expressed as:
then, in connection with equations (3) (4), y is expressed for the next network node output in the target sequence as:
wherein, -j represents before the occurrence of the jth element
A subsequence of (i.e.
Further, in the step (9), the node flow frequency calculation component selects a switch node with a higher use frequency according to the switch nodes traversed by all paths in the potential forwarding path set, stores the switch node in an array form, and sends the array to the dominance monitoring node management component in a node frequency updating form, where the array length is the same as the array length temporarily stored in the step (5).
Further, the advantageous monitoring node management component in the step (10) updates the advantageous monitoring node set according to the node frequency update information, and the detailed contents are as follows:
when a new node pair establishes a communication connection, a new active stream s is created j If the traversed switch node is v', the communication cost C incurred when polling the switch node j Is composed of
C j ＝D j (L req +L reph +|s j |L reply ) (6)
Wherein D is j Is the distance between the switch v' and the controller, L req Indicates the request message size, L reph Indicates the reply message header size, L reply Represents a reply message size;
the advantage monitoring node management component updates the advantage monitoring node by using an algorithm 2, and comprises the following steps:
when a new stream s j Traversing the switches v', if the flow is not in the coverage of the monitoring node, calculating the communication cost generated by polling the active flow according to the formula (6), if the communication cost is less than the polled flow and the flow only passes through a specific switch, updating the node in the advantageous monitoring node set, otherwise, continuously using the current monitoring node set to collect flow statistical information to realize flow monitoring, and repeating the process for the next switch node traversed by the flow. And finally, submitting the management result to a flow forwarding monitoring component in a form of an advantageous monitoring node.
Has the beneficial effects that: compared with the prior art, the invention has the following advantages:
(1) The invention is different from the existing Pull technology, and considers the fact that the network flow distribution generally follows the '20/80' rule, the invention is conceived to be sent from the perspective of selecting the monitoring node, and the switch with the densest flow traversal is preferentially selected as the flow monitoring node, so that the purposes of improving the flow collection non-redundancy rate and reducing the flow monitoring overhead are achieved while the flow statistical information is captured maximally.
(2) The invention innovatively provides a dynamic selection method of SDN flow dominant monitoring nodes for ensuring that the flow of a flow monitoring node set can always cover the network flow to the maximum extent, and the dynamic selection method comprises the steps of pre-screening and dynamic updating of the dominant monitoring nodes. Firstly, based on the shortest path set of hop count and link information such as available bandwidth, calculating an optimal flow forwarding path set by using a forwarding path selection algorithm based on bandwidth, and pre-screening an advantage monitoring node set according to the frequency of nodes flowing through; secondly, in order to adapt to network traffic fluctuation caused by traffic peak values, user habits and the like, a neural network model based on time series in deep learning is utilized, and through learning of a historical traffic forwarding path in a previous monitoring period, the internal characteristics of sequence type traffic forwarding and routing paths are captured, the flow-through nodes of traffic forwarding are predicted, dynamic updating of an advantage monitoring node set is completed, even if traffic fluctuation occurs in a future monitoring period, for example, due to the fact that bandwidth resources are reasonably distributed to adapt to forwarding path change caused by the fact that the traffic scale is increased, follow-up traffic forwarding can still flow through the advantage monitoring node set at the maximum probability.
(3) The invention does not need to add extra functions to the switch (such as not changing the matching structure of the OpenFlow protocol), and is easy to be deployed to the existing SDN network system.
In conclusion, the method and the device realize the maximum collection of the flow information in a closed-loop control mode by following the network flow distribution rule and combining a deep learning model, reduce the monitoring cost and are more suitable for being used in the scene with limited monitoring resources.
Drawings
FIG. 1 is an overall frame schematic of an embodiment of the present invention;
FIG. 2 is a flow chart of the method structure and process of the present invention;
FIG. 3 is a network node topology diagram of an example of the present invention;
FIG. 4 is a graph of experimental results of bandwidth surplus ratio for an example of the present invention;
FIG. 5 is a graph comparing experimental results of non-redundant flow ratios for an example of the present invention;
FIG. 6 is a graph comparing the results of the average delay test of the present invention;
fig. 7 is a comparison graph of experimental results of communication bandwidths according to an embodiment of the present invention.
Detailed Description
The technical solution of the present invention is described in detail below, but the scope of the present invention is not limited to the embodiments.
The general process of SDN traffic monitoring (taking Pull class as an example) is as follows: the SDN controller generates an FSC request and transmits the FSC request to each monitoring (switch) node through a control channel; each monitoring node completes flow statistic information collection according to the FSC request and transmits the result to the controller through a control channel; the controller summarizes and analyzes the flow statistical information from each monitoring node. The main overhead of the above procedure is communication overhead (such as bandwidth occupation of control channels) and information processing overhead (such as computational overhead and storage overhead of the controller and the monitoring node). Taking OpenFlow protocol as an example, the FSC request and reply messages for acquiring flow statistics in OpenFlow 1.0 are all over 120 bytes long. When the controller frequently sends out the flow monitoring requirement, on one hand, a large amount of FSC request and reply messages occupy a large amount of bandwidth of a control channel, and larger communication overhead is brought; on the other hand, the frequent collection, summarization and analysis of the convection statistical information respectively occupy more computing resources and storage resources of the switch and the controller, which brings huge information processing overhead and increases the information processing time delay of the basic tasks such as flow forwarding and control decision.
It can be seen from the above that, the larger the number of monitoring (switch) nodes is, the more FSC request reply messages are, and the larger the communication overhead of the control channel is; on the other hand, the more the flow statistical information gathered to the controller, the higher the information redundancy rate, and at the same time, the longer the information processing time delay of the controller is increased; thirdly, the more switch resources occupied by traffic monitoring in the whole network, the longer the information processing delay in traffic forwarding. If a smaller number of monitoring nodes are introduced during SDN traffic monitoring, communication overhead and information processing delay caused by SDN traffic monitoring are inevitably reduced.
Based on the thought, and meanwhile, the fact that the network traffic distribution generally follows the '20/80' rule is fully considered, the invention provides a basic idea of reducing SDN traffic monitoring overhead by preferentially selecting the switch with the densest traffic traversal as the monitoring node from the viewpoint of selecting the monitoring node, and the idea can realize the improvement of the traffic information collection non-redundancy rate and the reduction of the traffic monitoring overhead on the premise of maximally capturing the traffic statistical information.
As shown in FIG. 1, assume that there are 6 switches S in the network to be monitored i I is more than or equal to 1 and less than or equal to 6 and 8 hosts (H) i I is more than or equal to 1 and less than or equal to 8, the monitoring period is divided into T0-T1 and T1-T2, wherein T0, T1 and T2 are three monitoring moments. In the period from T0 to T1, 4 active flows exist in the network, namely f1 (H1-H6), f2 (H3-H7), f3 (H2-H8) and f4 (H3-H6). Since S1 and S3 cover all active flows, the dominance monitoring node set in this period is { S1, S3} without changing OpenFlow matching items. Even per-flow polling according to this node set saves over 50% communication cost over the original per-flow scheme. In a period from T1 to T2, it is assumed that, due to the influence of time, flow peak value change, user usage habits and other factors, an active flow changes in comparison with the period from T0 to T1, and 6 active flows, namely f1 (H1-H2), f2 (H1-H3), f3 (H2-H4), f4 (H5-H7), f5 (H6-H7) and f6 (H4-H8), appear, at this time, an advantageous monitoring node set is updated to { S2, S4}, and flow monitoring is performed according to the node set, so that a lower information redundancy rate and a lower monitoring overhead can be obtained.
As shown in fig. 2, the system for dynamically selecting an SDN network traffic dominance monitoring node according to this embodiment includes a control plane and a data plane; the data plane comprises a network resource module, and the network resource module sends the collected network topology information and the updated available bandwidth and residual bandwidth threshold value to the control plane in the form of a data packet through an OpenFlow1.3 protocol; the control plane comprises a forwarding calculation module, a node updating module and a path prediction module, wherein the forwarding calculation module calculates an optimal forwarding path set according to information sent by the data plane and sends the optimal forwarding path set to the node updating module; according to the optimal forwarding path set, the node updating module calculates the frequency of node flow, and pre-screens out an advantageous monitoring node set; according to the advantages provided by the node updating module, the path prediction module predicts a new forwarding path in a future period by using the monitored real forwarding path set and sends a prediction result to the node updating module to update the monitoring node again.
The network resource module comprises a network information sensing component and a bandwidth state updating component, the network information sensing component sends flow change information to the bandwidth state updating component, and then the network information sensing component and the bandwidth state updating component respectively send network topology information, available bandwidth and residual bandwidth threshold value information to the forwarding calculation module; the forwarding calculation module comprises a shortest path calculation component and a forwarding path calculation component, the shortest path calculation component calculates front K shortest paths between a source node and a destination node according to network topology information and sends results to the forwarding path calculation component, and the forwarding path calculation component calculates an optimal forwarding path set according to the front K shortest paths, available bandwidth and residual bandwidth threshold and sends the optimal forwarding path set to the node updating module; the node updating module comprises a node flow frequency calculating component and an advantage monitoring node management component, wherein the node flow frequency calculating component calculates high-frequency nodes according to an optimal forwarding path set, or gives node frequency updating information according to a potential forwarding path set, and then sends the node frequency updating information to the advantage monitoring node management component, and the advantage monitoring node management component pre-screens out the advantage monitoring nodes according to the flow frequency of the high-frequency nodes, or updates the nodes with low flow polling communication cost into the advantage monitoring nodes according to the node frequency updating information, and then sends the advantage monitoring nodes to the path prediction module; the path prediction module comprises a flow forwarding monitoring component, a forwarding path management component and a forwarding path prediction component, wherein the flow forwarding monitoring component sends a real forwarding path set monitored by the dominant monitoring node in the period to the forwarding path management component, the forwarding path management component provides all the existing real forwarding path sets as historical forwarding path sets to the forwarding path prediction component, the forwarding path prediction component sends the predicted potential forwarding path sets to a node flow frequency calculation component in the node updating module in a data packet mode, and finally the dominant monitoring node for flow monitoring in the next period is calculated by the dominant monitoring node management component, so that a system running in a closed loop is formed.
The dynamic selection method of the SDN network traffic advantage monitoring node dynamic selection system specifically comprises the following two parts:
(1) The pre-screening mechanism of the advantage monitoring node comprises the following steps: the pre-screening of the dominant monitoring node is only operated when the SDN traffic monitoring system is cold started, and includes two links of link information acquisition and monitoring node selection, which relate to mechanisms such as shortest path calculation, bandwidth state update, forwarding path calculation, and flow frequency calculation, and the specific description is as follows.
(2) The advantage monitoring node dynamically updates the strategy: the advantage monitoring node updating relates to strategies such as forwarding path prediction and monitoring node updating.
The specific implementation steps of the part (1) are described as follows:
a) Link information acquisition
(1) Shortest path computation
At T 0 In the cycle, the SDN controller identifies and manages a network topology structure, such as switch node information, link information, and the like, through information collected by a link discovery protocol LLDP. In the class for implementing the process, the mapping relationship between ports and links between switches, information of an access host, and the like are mainly described, and the information is stored in the network x. Based on the collected bottom layer topology and link information, the network calculates the set of K shortest paths from the source node to the destination node by using a Yen algorithm.
(2) Bandwidth status update
When a network node establishes a new communication connection, underlying physical network resources such as residual bandwidth, position distance and the like are constantly changed, and the parameters affect a forwarding path of final flow, so that the controller needs to acquire network resource information such as updated bandwidth state and send the updated information to other processes in a data packet form.
During operation, the controller obtains the available bandwidth of the path by passive measurement. Assuming the available bandwidth capacity of the link is C, the controller queries the switch port counter for the number of bytes at time t, N (t), over a time period γ, and the currently used bandwidth of the link, B, can be expressed as:
the controller calculates the available bandwidth of the whole path according to the available bandwidth of each link in each path in the K path sets, and therefore calculates the available bandwidth B of the whole path according to (1):
B′＝min (C-B)∈K′ (C-B)，K′∈K (2)
b) Link information acquisition
(1) Forwarding path computation
The controller integrates the calculated first K shortest paths and the updated related data such as network resource information into the network topology information stored in the network, and selects the path with the maximum available bandwidth from the first K shortest paths as the optimal path forwarding flow, and the core algorithm of the controller is shown as algorithm 1. Note that the "maximum" bandwidth does not mean that the value of the available bandwidth of the K links is the maximum, because in the same period, when there is other traffic entering the network domain, when the remaining available bandwidth of the immediate path is very likely to be smaller than that of the alternative path, but as long as the remaining available bandwidth of the current optimal path can still meet the bandwidth resource requirement required by forwarding other traffic, even if its bandwidth is not the maximum (algorithms step1 to step 4), it is still considered as the optimal path, otherwise, one of the computed alternative paths (K-1 paths) is randomly selected to continue forwarding (algorithms step5 to step 9), and the optimal forwarding path set is returned (algorithm step 10). And the information of the selected optimal forwarding path is used for configuring the flow table entry and is issued to each switch node on the path.
(1) Flow-through frequency calculation
For the nodes related to the optimal forwarding path of each flow, the controller selects the switch node with the densest traffic traversal as the monitoring node, namely the node with the highest forwarding traffic frequency. Note that our method requires coverage of all traffic as much as possible, therefore, for the traffic of a single polling switch that is not monitored, we select the switch node with the lowest actual monitoring cost for polling, and flexibly adjust the dominant monitoring node, so as to capture traffic information to the maximum extent under the condition of limited monitoring resources.
The specific implementation steps of the part (2) are described as follows:
a) Predictive model selection
The change of the traffic forwarding path is limited by various factors such as network topology, link state, available bandwidth and traffic scale, and the change rule of the traffic forwarding path needs to be mastered when the traffic forwarding path is predicted. Machine learning can automatically analyze laws from existing data, and is often used for prediction of unknown data. The invention proposes to select a seq2seq model in deep learning in the prediction of a traffic forwarding path, on one hand, a network path is sequence data, and the seq2seq model is a typical sequence analysis technology, adopts a network with an encoding-decoding structure, and has input and output of sequence data; on the other hand, seq2seq realizes gap bridging between a source space and a target space by encoding a variable length sequence into a fixed length encoding vector (in an encoding-decoding structure, encoding is responsible for changing a variable length source sequence into a fixed length vector expression, and decoding is responsible for changing the fixed length vector into a variable length target sequence), and can cope with the phenomenon that the existing value and the predicted value of a traffic forwarding path are not equal in dimension.
b) Seq2seq model-based forwarding path prediction
To predict T i (i is more than or equal to 1) the probability that all nodes in the SDN network appear on the next flow forwarding path in the period, the network path is formed into a matrix of mutual mapping between the nodes, and the matrix is fittedAnd speculate on the reasonable order of the nodes in the path. Let the source sequence and the target sequence be two network paths from the source node to the target node, each of which contains a set of traversal nodes from the source node to the target node. The basic way of forwarding path prediction is to find out a potential forwarding path (i.e. a target sequence) satisfying the characteristics of a historical forwarding path by learning and extracting the characteristics of a node sequence in the historical forwarding path according to the historical forwarding path (i.e. a source sequence) by using a coding-decoding structure in a seq2seq model.
The source sequence and the target sequence are two network paths between a source node and a target node, each of which contains a set of network nodes to be traversed from the source node to the destination node. The problem we consider here is that given an input source sequence, i.e. a historical forwarding path, the node sequence features of the source sequence are mainly learned and extracted by using an encoding-decoding structure in a model, and a target sequence satisfying a constraint condition, i.e. a path set whose prediction output satisfies the historical forwarding path features, is found.
The method comprises the steps that a forwarding path prediction model based on seq2seq takes a traversal node sequence from each source node to a target node in a historical traffic forwarding path set as a source sequence to be input into an encoder, the encoder comprises a plurality of RNN stacks, the encoder can output a hidden state of the encoder at each time point, the encoder compresses ordered forwarding path information into a final hidden state and then sends the final hidden state to a decoder, and the decoder outputs a target sequence according to the hidden state, namely, a future path forwarding node sequence is predicted and output. Through training of historical traffic data, the model can extract sequence features between nodes, namely internal features of a routing path, and the purpose of path discovery is achieved.
Vector for setting source sequence and target sequence respectively
Is shown in the specification, wherein
α and β represent the lengths thereof, respectively; d s And D t Data sets respectively representing a source sequence and a target sequence, the number of elements of the data sets being k; θ, ω represents parameters and weights in the neural network, then in the encoding-decoding structure, the representation form of the encoder is:
wherein the content of the first and second substances,
is a data set D s Is one of the k elements of (a), theta is a parameter of the encoder.
For the decoder, the task is to depend on the source sequence
Represents C and history information that has been previously generated
I.e. the forwarding path set in the history period generates the element N to be generated at the moment i i I.e., the next hop node in the predicted forwarding path. Let ω be the decoder's parameter, then the decoder can be expressed as:
then, in connection with equation (3) (4), for the next network node y output in the target sequence, it can be expressed as:
wherein-j represents the element before the j-th element appears
A subsequence of (i.e.
c) Monitoring node updates
Nodes which are most likely to pass through SN +1 to DN +1 paths in a Ti (i is more than or equal to 1) period can be predicted according to the generated target sequence, the result of recalculation of the node frequency is compared with the monitoring nodes preliminarily screened in the section 4, the dominant monitoring node set is updated, and the nodes in the set can cover the activity flow in the network domain as much as possible.
The core algorithm for monitoring node updates is shown in algorithm 2. When a new node pair establishes a communication connection, a new active stream s is created j If the traversed switch node is v', the communication cost C generated when polling the switch node j Is composed of
C j ＝D j (L req +L reph +|s j |L reply ) (6)
Wherein D is j Is the distance between the switch v' and the controller, L req Indicates the request message size, L reph Indicates the reply message header size, L reply Indicating the reply message size. If the communication cost generated by polling the new flow on the switch is less than the polled flow and the flow only passes through a specific switch, the node is updated to the dominant monitoring node set, otherwise, the current monitoring node set is continuously used for flow monitoring, and the next switch with convenient flow is calculated and compared to update the monitoring node.
Example 1:
in this embodiment, a dynamic selection system of an SDN network traffic dominance monitoring node is implemented and applied to SDN traffic monitoring. The system uses a Mininet simulator to simulate SDN bottom layer physical forwarding equipment, a software switch selects Open VSwitch to support OpenFlow1.3, ryu is used as an SDN controller, and a network topology which is widely used in network research and is generated randomly by a waxman diagram is used, as shown in fig. 3, the network topology comprises 35 nodes, connection is established with a probability p =0.05, namely, the probability of connection between the nodes in the network diagram which is switched randomly is 0.05.
The method comprises the steps of realizing a routing forwarding rule in an Ryu controller according to an optimal path algorithm flow based on bandwidth, issuing the routing forwarding rule to a switch in a flow table form, generating a path set between network nodes by operating Iverf on a host to manufacture and release flow, and collecting a data set.
The SDN network flow advantage monitoring node dynamic selection method runs data 1, namely the data set size:
the collected data sizes are counted according to the path lengths, and when the path lengths (unit: hop) are 5, 10, 15 and 20, the data sizes (unit: group) are 13207, 14694, 14702 and 14792 respectively. It can be seen that: the size of the collected data does not differ much because in this randomly generated network topology, most of the forwarding paths between any random source and destination nodes are closest to 15 hops.
Establishing a seq2seq neural network prediction model in the prediction model training, wherein the parameters are set as follows:
setting a double-layer GRU as an encoder, and a decoder as a double-layer unidirectional GRU network; the number of the neurons in the hidden layer is 500; setting a single network sequence of a data set as 50, setting the data size as 120, and randomly selecting 100 sequence training models; the experimental environment is a server from an Intel 16 core to a strong E5-2650 CPU, a 32GB memory and 8-path GTX 1080 Ti.
An SDN network flow advantage monitoring node dynamic selection method runs data 2, namely the accuracy of a prediction model:
the data sets are randomly classified according to the proportion of 4. It can be seen that: when the path lengths are respectively 5, 10, 15 and 20, the testing accuracy of the path lengths reaches more than 98%, and when the path length is 15, the model training accuracy is highest, which indicates that the fitting performance of the model is the best at the moment.
TABLE 1 learning model accuracy for waxman stochastic generation graphs based on path length
The SDN network flow advantage monitoring node dynamic selection method runs data 3-bandwidth surplus rate:
based on the established forwarding path prediction model, in the example, dominant monitoring nodes in one period of the SDN network are screened out, and 5 nodes n are selected from the monitoring nodes screened out from the 35 nodes in fig. 3 j (j is more than or equal to 1 and less than or equal to 5), and designing the following flow sending process:
·0～200s：n 1 to n 5 Sending UDP flow with the sending rate of 6Mbps;
·200s～400s：n 2 to n 5 Sending UDP flow with the sending rate of 6Mbps;
400s to 600s: interrupt n 1 、n 2 Sending traffic, n 3 To n 4 Sending UDP flow with the sending rate of 8Mbps until the sending is finished;
·600s～800s：n 1 to n 5 Sending UDP flow with the sending rate of 10Mbps until the sending is finished;
FIG. 4 shows the monitored bandwidth remaining rate of the link in the process (b:)Available bandwidth/bandwidth capacity), as can be seen in the figure, at T 0 In the pre-screening process in the period, the controller updates the utilization condition of the bandwidth resources of the bottom layer in real time, truly reflects the bandwidth surplus rate of a link in the flow sending process, and sends the updated information to other modules.
PFP algorithm and Greedy algorithm are two algorithms commonly used in polling switch nodes to collect traffic information in traffic monitoring, which are compared with the present invention by randomly selecting and releasing traffic between hosts under the node switches in fig. 5, where PFP is the reference for comparison, where each flow is polled on the switch closest to the controller using a polling single request. In the example, the default setting is that the polling request message size is 123 bytes, the reply request message header size is 80 bytes, and the reply request message size of a single stream is 127 bytes.
The SDN network traffic dominance monitoring node dynamic selection method runs data 4-non-redundant traffic ratio:
as shown in fig. 5, it can be seen that: with the PFP algorithm, since each flow goes through a single polling procedure, i.e., a specific single flow (indicated in the polling request) is returned when the switch is polled, and each pair of request/reply messages can only cover the rate of one flow, the redundant information in the flow statistics returned after polling all switches must be the most. Compared with the Greedy scheme, when the flow quantity is fixed, the forwarding path of the flow is predicted through the machine learning model, when the next forwarding node is output, the monitoring node is updated accordingly, the new flow enters the network domain and traverses the monitoring node switch, the controller only polls the switch to output the flow information, the repeated polling of a plurality of switch nodes traversed by the flow can be avoided to a certain extent, the proportion of redundant flow is reduced, and therefore various measurement expenses and time delays caused by non-redundant flow rate indirectly are reduced.
The SDN network flow advantage monitoring node dynamic selection method runs data 5-average time delay:
as shown in fig. 6, it can be seen that: the average delay steadily increases with the amount of traffic. The PFP scheme periodically queries the switch for count information for each port for each flow, which causes the switch to frequently interact with the controller due to its polling of switch nodes at a higher frequency, resulting in a relatively long network delay; when the traffic scale gradually increases, greedy needs to acquire a polling scheme by constructing a weighted set to cover the traffic problem, so that a certain calculation time overhead is caused, and information processing time delay is increased to a certain extent.
The SDN network flow advantage monitoring node dynamic selection method runs data 6-communication cost:
as shown in fig. 7, it can be seen that: PFP schemes are the most costly because each individual flow requires a dedicated pair of request and reply message channels and occupy more network bandwidth as polling frequency increases; the Greedy algorithm is based on the idea of local optimization, the switch polling with the minimum cost benefit is selected each time, but as the number of total active flows is increased, the collected redundant flow is also increased, the network bandwidth burden is increased to a certain extent, and the communication cost in the measurement overhead is increased; compared with the prior art, on one hand, the optimal path algorithm based on the bandwidth provided by the invention realizes network load balance, can reduce the risk of network congestion and other emergency situations to a certain extent, and reasonably distributes network bandwidth resources, and on the other hand, the monitoring node updating strategy realizes that a single polling request retrieves all flow information on the switch, thereby greatly reducing the number of request/reply messages and further reducing the communication cost.
Claims (9)
1. A SDN network flow advantage monitoring node dynamic selection system is characterized in that: including a control plane and a data plane; the data plane comprises a network resource module, and the network resource module sends the collected network topology information and the updated available bandwidth and residual bandwidth threshold value to the control plane in the form of a data packet through an OpenFlow1.3 protocol; the control plane comprises a forwarding calculation module, a node updating module and a path prediction module, wherein the forwarding calculation module calculates an optimal forwarding path set according to information sent by the data plane and sends the optimal forwarding path set to the node updating module; according to the optimal forwarding path set, the node updating module calculates the frequency of node flow, and pre-screens out an advantageous monitoring node set; according to the superior monitoring node provided by the node updating module, the path prediction module predicts a new forwarding path in a future period by using the monitored real forwarding path set and sends a prediction result to the node updating module to update the monitoring node again;
the path prediction module comprises a flow forwarding monitoring component, a forwarding path management component and a forwarding path prediction component, wherein the flow forwarding monitoring component sends a real forwarding path set monitored by the dominant monitoring node in the period to the forwarding path management component, the forwarding path management component provides all the existing real forwarding path sets as historical forwarding path sets to the forwarding path prediction component, the forwarding path prediction component sends the predicted potential forwarding path sets to a node flow frequency calculation component in the node updating module in a data packet mode, and finally the dominant monitoring node for flow monitoring in the next period is calculated by the dominant monitoring node management component to form a closed-loop operation system;
the forwarding path prediction component constructs a network model Sequence-to-Sequence of an encoding-decoding structure according to a historical forwarding path set sent by the forwarding path management component as a data set, the network path is formed into a matrix which is mapped among nodes, and the reasonable Sequence of the nodes in the path is fitted and presumed, wherein the specific construction method comprises the following steps:
the forwarding path prediction component finds out a potential forwarding path meeting the characteristics of the historical forwarding path by learning and extracting the node Sequence characteristics in the historical forwarding path according to the historical forwarding path and by using an encoding-decoding structure in a network model Sequence-to-Sequence, and the detailed contents are as follows:
vector for setting source sequence and target sequence respectively
It is shown that, among others,
α and β represent the lengths thereof, respectively; d s And D t Representing data sets of a source sequence and a target sequence respectively, wherein the element number of the data sets is k; θ, ω represent the encoder parameters and the decoder weights, respectively, and in the encoding-decoding structure, the encoder is represented by:
wherein the content of the first and second substances,
is a data set D s One of k elements of (a);
for the decoder, the task is to depend on the source sequence
C and history information that has been generated before, i.e. the set of forwarding paths in the history period
Generating an element N to be generated at time i i I.e., the next hop node in the predicted forwarding path, the decoder now is represented as:
then, in connection with equations (3) (4), for the next network node y output in the target sequence, it is expressed as:
wherein-j represents the element before the j-th element appears
A subsequence of (i)
2. The SDN network traffic dominance monitoring node dynamic selection system of claim 1, wherein: the network resource module comprises a network information sensing component and a bandwidth state updating component, the network information sensing component sends flow change information to the bandwidth state updating component, and then the network information sensing component and the bandwidth state updating component respectively send network topology information, available bandwidth and residual bandwidth threshold value information to the forwarding calculation module;
the forwarding calculation module comprises a shortest path calculation component and a forwarding path calculation component, the shortest path calculation component calculates front K shortest paths between a source node and a destination node according to network topology information and sends the result to the forwarding path calculation component, and the forwarding path calculation component calculates an optimal forwarding path set according to the front K shortest paths and available bandwidth and residual bandwidth threshold and sends the optimal forwarding path set to the node updating module;
the node updating module comprises a node flow frequency calculating component and an advantage monitoring node management component, wherein the node flow frequency calculating component calculates a switch node with the highest use frequency as a high-frequency node according to an optimal forwarding path set during cold start, or gives node frequency updating information according to a potential forwarding path set during non-cold start, and then sends the node frequency updating information to the advantage monitoring node management component; the advantage monitoring node management component pre-screens out the advantage monitoring nodes according to the flow frequency of the high-frequency nodes during cold start, and updates the nodes with the polling communication cost smaller than the polled flow into the advantage monitoring nodes according to the node frequency updating information during non-cold start, and then sends the advantage monitoring nodes to the path prediction module.
3. A dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 2, wherein: the method sequentially comprises the following steps:
(1) The network information perception component respectively sends the topology information and the flow change information of the bottom physical network to the shortest path calculation component and the bandwidth state updating component;
(2) The shortest path calculation component calculates the first K shortest path sets according to the network topology information and sends the results to the forwarding path calculation component;
(3) The bandwidth state updating component sends the available bandwidth and the residual bandwidth threshold value to the forwarding path calculation component in the form of a data packet according to the traffic change information;
(4) The forwarding path calculation component calculates an optimal forwarding path set by combining the information obtained in the step (2) and the step (3) and sends the result to the node flow frequency calculation component;
(5) The node flow frequency calculation component sends the result to the dominant monitoring node management component by calculating and screening high-frequency nodes on the forwarding path;
(6) The dominant monitoring node management component preliminarily screens out dominant monitoring nodes according to the node frequency and sends the dominant monitoring nodes to the flow forwarding monitoring component;
(7) The flow forwarding monitoring component sends the real flow forwarding path set of the current period monitored by the dominant monitoring node to the forwarding path management component;
(8) The forwarding path prediction component predicts a potential forwarding path set according to a historical forwarding path set provided by the forwarding path management component and sends the potential forwarding path set to the node flow frequency calculation component;
(9) The node flow frequency calculation component updates the node frequency according to the potential forwarding path set, and the result is provided to the advantage monitoring node management component in the form of node frequency updating information;
(10) And (4) updating the advantageous monitoring node set by the advantageous monitoring node management component according to the node frequency updating information, and performing flow forwarding monitoring in the next period in the step (7).
4. The dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: the network information perception component in the step (1) is that the SDN controller collects network topology related information through a link discovery protocol LLDP and stores the network topology related information in a network x;
obtaining a first K shortest path sets by adopting a Yen algorithm in the step (2);
the network topology related information includes mapping relationship between ports and links between switches, information of access hosts and link information.
5. The dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: the residual bandwidth threshold value in the step (3) is set by a network administrator according to the actual network scale and the flow fluctuation rule; the available bandwidth is calculated by an SDN controller in a passive measurement manner, and if the available bandwidth capacity of a link is E, the controller queries, with a time period γ, the number of bytes N (t) in a switch port counter at time t, then the currently used bandwidth B of the link is represented as:
the SDN controller calculates the available bandwidth of the entire path according to the available bandwidth of each link in each path in the previous K shortest path sets, and therefore calculates the available bandwidth B' of the entire path according to (1):
B′＝min (E-B)∈K′ (E-B)，K′∈K (2)。
6. the dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: the step (4) comprises the following steps in sequence:
(4.1) integrating the first K shortest paths based on the hop count and the updated available bandwidth information of the network link by the SDN controller, selecting the shortest paths in the first K shortest paths as a first forwarding path, and taking the other K-1 paths as alternative paths;
(4.2) if the available bandwidth of the forwarding link changes, namely a new active flow enters, comparing the remaining available bandwidth of the forwarding path with the required bandwidth size of the new flow, if the remaining available bandwidth of the forwarding path is larger than the required bandwidth, turning to the step (4.3), otherwise, executing the step (4.4);
(4.3) if the residual bandwidth of the link is in the range of the residual bandwidth threshold after the new flow shares the bandwidth, turning to the step (4.5), otherwise, executing the step (4.4);
(4.4) reselecting an alternative path as a forwarding path of the new traffic, and turning to (4.2);
(4.5) if the path is an alternative path, the controller firstly issues a new flow entry to the switch on the path, and then submits the switch nodes traversed by the path to the node flow frequency calculation component; otherwise, directly submitting the result to the node flow frequency calculation component.
7. The dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: in the step (5), the node flow frequency calculation component calculates the use frequency of the node related to the forwarding path of each flow through the SDN controller, selects the switch node with the highest use frequency in all paths, selects the switch node with the lowest actual monitoring cost for the traffic of the single polling switch which is not monitored, temporarily stores the nodes in an array form, and sends the nodes in a result form of the high-frequency node to the dominant monitoring node management component.
8. The dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: and (3) the node flow frequency calculation component in the step (9) selects the switch node with the highest use frequency according to the switch nodes traversed by all paths in the potential forwarding path set, stores the switch node in an array form, has the same array length as the array length temporarily stored in the step (5), and then sends the array to the dominant monitoring node management component in a node frequency updating form.
9. The dynamic selection method for the SDN network traffic dominance monitoring node dynamic selection system according to claim 3, wherein: in the step (10), the advantageous monitoring node management component updates the advantageous monitoring node set according to the node frequency update information, and the detailed contents are as follows:
when a new node establishes a communication connection, a new active stream s is generated j If the traversed switch node is v', the communication cost C generated when polling the switch node j Is composed of
C j ＝D j (L req +L reph +|s j |L reply ) (6)
Wherein D is j Is the distance between the switch v' and the controller, L req Indicates the request message size, L reph Indicates the reply message header size, L reply Indicating a reply message size;
when a new active stream s j Traversing the switches v', if the active flow is not in the coverage of the monitoring node, calculating the communication cost generated by polling the active flow according to the formula (6), if the communication cost is less than the polled flow and the active flow only passes through a specific switch, updating the node in the advantageous monitoring node set, otherwise, continuously using the current monitoring node set to collect the statistical information of the active flow to realize flow monitoring, and repeating the next switch node traversed by the flowThe above-mentioned process; and finally, submitting the management result to a flow forwarding monitoring component in a form of an advantageous monitoring node.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910489576.1A CN110275437B (en) | 2019-06-06 | 2019-06-06 | SDN network flow dominance monitoring node dynamic selection system and method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910489576.1A CN110275437B (en) | 2019-06-06 | 2019-06-06 | SDN network flow dominance monitoring node dynamic selection system and method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110275437A CN110275437A (en) | 2019-09-24 |
| CN110275437B true CN110275437B (en) | 2022-11-15 |
Family
ID=67960426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910489576.1A Active CN110275437B (en) | 2019-06-06 | 2019-06-06 | SDN network flow dominance monitoring node dynamic selection system and method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110275437B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111953669B (en) * | 2020-07-30 | 2022-10-25 | 江苏大学 | Tor flow tracing and application type identification method and system suitable for SDN |
| CN113269347B (en) * | 2021-03-31 | 2023-05-30 | 安徽农业大学 | College express network node flow prediction method based on random forest |
| CN113992590A (en) * | 2021-10-25 | 2022-01-28 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Link load balancing method based on software defined network |
| CN113965400B (en) * | 2021-11-01 | 2023-06-30 | 电子科技大学长三角研究院(衢州) | Method for determining flow key points in communication network |
| CN115225561B (en) * | 2022-08-15 | 2022-12-06 | 南京邮电大学 | Route optimization method and system based on graph structure characteristics |
| CN115474215B (en) * | 2022-09-06 | 2023-03-21 | 广东九博科技股份有限公司 | Router-based networking method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015119408A (en) * | 2013-12-19 | 2015-06-25 | Kddi株式会社 | Traffic flow allocation method and device |
| CN105721577A (en) * | 2016-02-15 | 2016-06-29 | 安徽大学 | Server load balancing method for software defined network |
| CN106357537A (en) * | 2016-11-09 | 2017-01-25 | 北京工业大学 | Link monitoring method based on SDN multipath transmission |
| KR101748750B1 (en) * | 2016-02-29 | 2017-06-19 | 부산대학교 산학협력단 | System and Method for Controlling SDN based Traffic aware Bandwidth in Virtualized WLANs |
| CN107959633A (en) * | 2017-11-18 | 2018-04-24 | 浙江工商大学 | A kind of load balance method based on price mechanism in industry real-time network |
| CN108880888A (en) * | 2018-06-20 | 2018-11-23 | 浙江工商大学 | A kind of SDN network method for predicting based on deep learning |
| CN109067758A (en) * | 2018-08-23 | 2018-12-21 | 江苏大学 | A kind of SDN network data transmission intimacy protection system and its method based on multipath |
-
2019
- 2019-06-06 CN CN201910489576.1A patent/CN110275437B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015119408A (en) * | 2013-12-19 | 2015-06-25 | Kddi株式会社 | Traffic flow allocation method and device |
| CN105721577A (en) * | 2016-02-15 | 2016-06-29 | 安徽大学 | Server load balancing method for software defined network |
| KR101748750B1 (en) * | 2016-02-29 | 2017-06-19 | 부산대학교 산학협력단 | System and Method for Controlling SDN based Traffic aware Bandwidth in Virtualized WLANs |
| CN106357537A (en) * | 2016-11-09 | 2017-01-25 | 北京工业大学 | Link monitoring method based on SDN multipath transmission |
| CN107959633A (en) * | 2017-11-18 | 2018-04-24 | 浙江工商大学 | A kind of load balance method based on price mechanism in industry real-time network |
| CN108880888A (en) * | 2018-06-20 | 2018-11-23 | 浙江工商大学 | A kind of SDN network method for predicting based on deep learning |
| CN109067758A (en) * | 2018-08-23 | 2018-12-21 | 江苏大学 | A kind of SDN network data transmission intimacy protection system and its method based on multipath |
Non-Patent Citations (2)
| Title |
|---|
| Multipath Based Privacy Protection Method for Data Transmission in SDN;Na Dong等;《Artificial Intelligence for Communications and Networks. First EAI International Conference》;20190526;第127-138页 * |
| Toward Privacy-Preserving Symptoms Matching in SDN-Based Mobile Healthcare Social Networks;Shunrong Jiang等;《IEEE Internet of Things Journal》;20180129;第5卷(第3期);第1379-1388页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110275437A (en) | 2019-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110275437B (en) | SDN network flow dominance monitoring node dynamic selection system and method thereof | |
| CN108521375B (en) | SDN-based network multi-service flow QoS transmission and scheduling method | |
| Wang et al. | Load-balancing routing in software defined networks with multiple controllers | |
| CN108600102B (en) | Flexible data transmission system based on intelligent cooperative network | |
| CN108259367B (en) | Service-aware flow strategy customization method based on software defined network | |
| CN107005781B (en) | System and method for generating virtual network topology for M2M communications | |
| Su et al. | CeMon: A cost-effective flow monitoring system in software defined networks | |
| CN112491619B (en) | Service customization network resource self-adaptive distribution method based on SDN | |
| CN109688056B (en) | Intelligent network control system and method | |
| Hyun et al. | Knowledge-defined networking using in-band network telemetry | |
| EP3039817A1 (en) | Determination and use of link performance measures | |
| Chen et al. | Multi-controller placement towards SDN based on Louvain heuristic algorithm | |
| Zhang et al. | Efficient flow detection and scheduling for SDN-based big data centers | |
| Zhu et al. | Centralized QoS routing using network calculus for SDN-based streaming media networks | |
| Mohammadi et al. | An adaptive type-2 fuzzy traffic engineering method for video surveillance systems over software defined networks | |
| Wang et al. | PrePass: Load balancing with data plane resource constraints using commodity SDN switches | |
| Jin et al. | Towards joint resource allocation and routing to optimize video distribution over future internet | |
| CN107995121A (en) | Flow statistical method in software defined network based on asterisk wildcard | |
| Zhang et al. | DSOQR: Deep reinforcement learning for online QoS routing in SDN-based networks | |
| Zheng et al. | Application of improved ant colony algorithm in load balancing of software-defined networks | |
| Zhang et al. | Dynamic network service deployment across multiple SDN domains | |
| Mohammadi et al. | Taxonomy of traffic engineering mechanisms in software-defined networks: a survey | |
| Li et al. | CFlow: A learning-based compressive flow statistics collection scheme for SDNs | |
| Lin et al. | Proactive multipath routing with a predictive mechanism in software‐defined networks | |
| Mogyorósi et al. | Adaptive protection of scientific backbone networks using machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |