Credible dynamic load balancing method of data center network based on SDN (Software Defined Network)

Abstract

The invention discloses a credible dynamic load balancing method of a data center network based on an SDN (Software Defined Network) in the technical field of computer networks and communication, which combines the flexibility and efficient data processing capability of the software defined network and introduces a credibility mechanism into a particle swarm algorithm. The credible and dynamic load balancing of the internal flow of the data center can be realized. The method is characterized in that on the basis of a traditional particle swarm algorithm, the path credibility is introduced to serve as a fitness function, the algorithm can guide the motion direction of particles according to the path credibility in the search process, and therefore when link load balancing is carried out, not only is single performance indexes concerned, but also the reliability of the algorithm is improved. And a plurality of key performance indexes such as packet loss rate, throughput, time delay and jitter of the link are comprehensively considered. By comprehensively considering multi-dimensional indexes in a fitness function, the optimized particle swarm optimization algorithm can finally obtain a conclusion of improving the stability and efficiency of data transmission.

Description

Technical Field

[0001] The present invention relates to a trusted dynamic load balancing method for a data center network based on SDN, and belongs to the technical field of computer networks and communications. Background Art

[0002] In today's digital age, data centers serve as the core infrastructure for cloud computing, big data processing, and internet services. Their network performance and efficiency directly impact the quality of business operations and user experience. With the explosive growth of data traffic and the increasing diversity of service demands, data center networks (DCNs) face unprecedented challenges, particularly how to efficiently and reliably manage and allocate network resources to achieve low latency, high throughput, and high availability for data transmission.

[0003] Software Defined Networking (SDN), a revolutionary network architecture, separates the control and data planes, opening up new possibilities for flexible management and intelligent scheduling of data center networks. SDN allows network administrators to programmatically control network behavior through a centralized controller, enabling dynamic configuration and optimization of network resources, thus enhancing network programmability and flexibility.

[0004] In typical network environments, link load balancing algorithms are relatively simple, primarily aiming to ensure even traffic distribution across multiple links and prevent overloading of specific links. Furthermore, typical link load balancers typically perform basic health checks, such as confirming link status through ping tests, and are sometimes statically configured based on pre-set rules or policies, with less frequent dynamic adjustments.

[0005] In contrast, link load balancing algorithms within data centers are more complex and must cope with the challenges of high-speed data flows and massive traffic volumes within data centers. Data center networks typically consist of high-performance switches, routers, servers, storage devices, and security equipment. They are characterized by high bandwidth, low latency, high reliability, and scalability to support large amounts of data processing and transmission needs, and ensure stable system operation through redundant design. This requires algorithms that can automatically adjust based on real-time network conditions, such as identifying and bypassing congested paths to optimize overall performance. Considering that data centers host numerous mission-critical applications, their load balancing solutions place particular emphasis on high availability and redundant design to ensure that the system can maintain normal operation even if some components fail.

[0006] Most current literature on link load balancing in SDN data centers focuses solely on a few key metrics, such as bandwidth utilization and latency, while ignoring other equally important factors, such as node packet loss rate, node throughput, link jitter, and link delay. However, in actual application scenarios, optimizing only a single or a few metrics often cannot fully meet the needs of complex network environments. For example, in large-scale data centers, high node packet loss rates can lead to reduced data transmission reliability, seriously impacting business continuity; in high-performance computing environments, insufficient node throughput can limit the system's overall processing power and affect the efficiency of computing tasks; and in real-time communication applications, increased link jitter and link delay can lead to reduced audio and video quality, impacting the user experience. Therefore, to achieve more comprehensive and efficient link load balancing, it is necessary to comprehensively consider multiple metrics when designing and evaluating optimization strategies to ensure optimal network performance and user experience in different application scenarios.

[0007] In summary, this paper aims to propose an innovative, trusted dynamic link load balancing technology for SDN-based data center networks. By using a multi-metric trust function as the fitness function within a particle swarm optimization algorithm, the load balancing algorithm can determine the ultimate trusted optimal path, thereby resolving the challenges inherent in existing technologies. Summary of the Invention

[0008] The purpose of the present invention is to overcome the shortcomings of the prior art, such as uneven link load, slow response and insufficient reliability.

[0009] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0010] A trusted dynamic load balancing method for a data center network based on SDN is provided, comprising the following steps:

[0011] Obtain network topology information and set link load thresholds based on the network topology information;

[0012] Obtaining a real-time link load of a link, and comparing the real-time link load of the link with a set link load threshold; triggering a load balancing mechanism when the real-time link load of the link exceeds the link load threshold; and recording a link whose real-time link load exceeds the link load threshold as an overloaded link;

[0013] The load balancing mechanism includes:

[0014] Obtain the characteristics of bandwidth-consuming elephant flows;

[0015] Calculate new routing paths through particle swarm optimization;

[0016] The SDN controller sends the updated flow table entries to the switches corresponding to the new routing paths;

[0017] The switch switches the elephant flow from the overloaded link to the new path based on the updated flow table entry, and feeds back the link load status to the SDN controller in real time.

[0018] Furthermore, the calculating of the new routing path by the improved particle swarm algorithm includes: transforming the path credibility function into a fitness function;

[0019] The path credibility function includes link credibility and node credibility, and is expressed as:

[0020]

[0021] is the path credibility from node u to node v, refers to the link credibility from node u to node v, refers to the node credibility of node i on the path, is the existence identification function of the node on the path;

[0022] in, The calculation formula is:

[0023]

[0024] , and is the weight coefficient, Indicates the average bandwidth utilization of the entire link, Indicates the number of segment links contained in the link; Indicates the link delay, Indicates link flapping.

[0025] Furthermore, the node credibility The calculation formula is:

[0026]

[0027] For nodes The packet loss rate, Representative Node The actual throughput, is a node The maximum theoretical throughput, , is a weight factor that satisfies ;

[0028] Among them, the node Packet loss rate The calculation formula is:

[0029]

[0030] node The throughput is calculated as:

[0031]

[0032] ; is the amount of data sent by node i, is the amount of data received by node i.

[0033] Furthermore, the link delay is calculated by sending a probe data packet and calculating the round trip time, the link delay of node i Expressed as:

[0034]

[0035]

[0036] Link jitter of node i Expressed as:

[0037]

[0038] It is The round trip time of a probe packet, is the average of all RTTs, N is the number of probe packets, is the receiving timestamp of the i-th probe packet, is the sending timestamp of the i-th probe packet.

[0039] Furthermore, the bandwidth utilization of the entire link The calculation formula is:

[0040]

[0041] In time interval The change in byte data sent by node i within , The change in byte data received by node j within Representation node arrive Inter-link capacity.

[0042] Furthermore, the calculation of the new routing path by the particle swarm algorithm includes: the speed of particle q The update formula is:

[0043]

[0044]

[0045] is the inertia weight, and is the acceleration factor, and is a random number in the range [0,1], It is a particle The best position in history, is the historical best position of the entire particle group, It is a particle exist Current location at the moment, The random perturbation term introduced for the simulated annealing idea;

[0046] in, Expressed as:

[0047]

[0048] yes The current temperature at the moment, is a random number in the range [-1,1]; By cooling factor Gradually reduce.

[0049] Furthermore, the link load balancing degree is calculated using a first-order exponential smoothing method:

[0050] S t = λ ∙ ∑ 1 N [loa d a (t) - load(t) — ] 2 N +(1 - λ ) ∙ S t - 1

[0051] is the smoothed value of the network load at time t obtained by weighted calculation, is the smoothed value of the previous moment, is a smoothing constant whose value range is [0,1]; N represents the total number of links in the network.

[0052] Furthermore, the network topology adopts a Fat-tree structure with k=4, and the objective function of load balancing is to minimize the maximum link utilization, which includes flow conservation constraints, bandwidth constraints, and non-negativity constraints. The formula is:

[0053]

[0054] Represents a network switch link set, represents the link capacity between nodes i and j; Indicates the network load balancing parameters at time t Greater than threshold When the elephant flow gathers; represents the rth elephant flow The original node, represents the rth elephant flow The destination node, represents the rth elephant flow Bandwidth requirements Indicates that the rth elephant flow is satisfied The traffic on link (i, j) under bandwidth demand.

[0055] Furthermore, the dynamic weight of the particle swarm algorithm Adaptive adjustment based on the fitness function value, the formula is:

[0056]

[0057] and are the maximum and minimum values of the weight, and are the maximum and minimum values of the fitness function, respectively. The particle at the current moment The fitness function value of .

[0058] Furthermore, the probability that the particle swarm algorithm accepts inferior solutions during the search process is It decreases as the temperature decreases. The specific formula is:

[0059]

[0060] For particles exist The fitness function value at the moment, is the current temperature.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. This invention combines the flexibility of software-defined networks with efficient data processing capabilities. By introducing a credibility mechanism into the particle swarm algorithm, it can achieve reliable and dynamic load balancing of traffic within the data center. At the same time, based on the traditional particle swarm algorithm, it introduces path credibility as a fitness function, allowing the algorithm to guide the movement direction of particles according to the credibility of the path during the search process. Therefore, when performing link load balancing, it not only focuses on a single performance indicator, but also comprehensively considers multiple key performance indicators such as the link's packet loss rate, throughput, latency, and jitter. By comprehensively considering multi-dimensional indicators in the fitness function, the optimized particle swarm algorithm can ultimately improve the stability and efficiency of data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 FIG2 is a flow chart of a load balancing algorithm provided in the first embodiment of the present invention;

[0064] Figure 2 FIG. 1 is a diagram showing the composition of the credibility proposed in the first embodiment of the present invention;

[0065] Figure 3 FIG2 is a flowchart of link load balancing proposed in the first embodiment of the present invention;

[0066] Figure 4 The figure shows the SDN data center network architecture proposed in the first embodiment of the present invention. DETAILED DESCRIPTION

[0067] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Unless there is a conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0068] The term "and / or" in this document simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " in this document generally indicates an "or" relationship between the related objects.

[0069] Example 1:

[0070] This embodiment provides a trusted dynamic load balancing method for a data center network based on SDN, including the following steps:

[0071] First, the centralized control features of the SDN controller are used to collect network topology information, including switches, routers, link status, etc., and a threshold is set for the link load. When the link load reaches this threshold, the load balancing mechanism is triggered.

[0072] Next, the load on each link in the current network is assessed to determine whether the link load exceeds the preset threshold. By monitoring data, it identifies bandwidth-intensive traffic flows and analyzes their characteristics, including source address, destination address, and traffic volume.

[0073] Then the adaptive load balancing algorithm is used to calculate the new routing path, such as Figure 1 As shown, the algorithm aims to transfer the elephant flows from the overloaded links to the backup links. Based on the algorithm calculation results, it determines which elephant flows need to be rerouted and the new routing path.

[0074] Then, the flow table entries are updated in the SDN controller, changing the routing path of the elephant flow to the new path, and the updated flow table entries are sent to the relevant network switches.

[0075] Finally, rerouting and monitoring are implemented, and the switch reroutes the elephant flow according to the new flow table entry.

[0076] After rerouting is implemented, evaluate the effectiveness of link load balancing and continue to monitor the load balancing degree in the network to ensure continuous optimization of network performance.

[0077] like Figure 4 As shown, the data center network in this patent uses a Fat-tree topology with k=4 and is represented as ,in Represent the network switch node set and link set respectively, represents the link capacity between nodes i and j, and the network topology does not change; Indicates the network load balancing parameters at time t Greater than threshold When the elephant flow gathers; Represents the rth elephant flow respectively The source node, destination node, and bandwidth requirements. Indicates that the rth elephant flow is satisfied Traffic on link (i, j) under bandwidth requirements. Taking minimizing the maximum link utilization u as the objective function, the optimal routing mathematical model for the load balancing problem is as follows:

[0078] Objective function:

[0079]

[0080] The above three formulas respectively represent:

[0081] The first formula specifies the traffic conservation law that must be satisfied during traffic scheduling, which states that the traffic entering a node, excluding the source and destination nodes, must be equal to the traffic leaving the node. The second formula considers the bandwidth constraints that must be satisfied during large flow scheduling. The third formula defines that the traffic on each link must be non-negative.

[0082] The load balancing degree of the entire link is calculated using the following formula:

[0083] S t =λ∙ ∑ 1 N [loa d a (t)- load(t) — ] 2 N +(1-λ)∙ S t-1

[0084] in, is the smoothed value of the network load at time t obtained by weighted calculation, is the actual link load variance at time t, is the smoothed value of the previous moment, is a smoothing constant, and its value range is [0,1]. The value of is 0.5. The present invention adopts the first-order exponential smoothing method in the time series prediction method to predict the network load balance at the next moment.

[0085] Among them, the idea of simulated annealing algorithm is added to the particle swarm algorithm to optimize its ability to find the global optimal solution. The specific implementation formula is as follows: ,

[0086]

[0087] in, is the inertia weight, and is the acceleration factor, and is a random number in the range [0,1], It is a particle The best position in history, It is the best historical position of the entire group. It is a particle exist Current location at the moment It is a random perturbation term introduced by the simulated annealing idea and can be expressed as:

[0088]

[0089] in, yes The current temperature at the moment, Is a random number in the range [-1,1]. The temperature update formula is: ,in, is the cooling factor, 0< <1.

[0090] In the optimized particle swarm algorithm, when the new solution is better than the current solution, the new solution is always accepted.

[0091]

[0092] in, For particles exist The algorithm always accepts the transition when the fitness function value at time t is calculated. The algorithm calculates the fitness function value as shown in Formula (X) in the text. When the fitness difference is positive, the probability of accepting the transition decreases exponentially as the difference increases. However, as long as there is a non-zero probability, even inferior solutions may be accepted, especially at high temperatures. As the algorithm runs and the temperature gradually decreases, the algorithm tends to reject poor solutions more and more, concentrating on the well-optimized solutions in the later stages of the search.

[0093] The load balancing algorithm adds a particle swarm algorithm with adaptive weights. The dynamic weight formula adjusted according to fitness is:

[0094]

[0095] in, and are the maximum and minimum values of the weight. The weight in this article changes with the change of the fitness function value. The higher the fitness function value, the higher the credibility of the current path, which means that it is close to or has reached a better solution. Therefore, it is necessary to reduce the dynamic weight to reduce random exploration, stabilize and further optimize the quality of the solution. and are the maximum and minimum values of the fitness function, respectively. The particle at the current moment The fitness function value of .

[0096] In the optimization particle swarm algorithm, the fitness function used is the path credibility function, which is divided into path credibility and node credibility, such as Figure 2 As shown, the specific implementation formula is as follows. The present invention guides the iteration of the particle swarm by using path credibility as the fitness function.

[0097]

[0098] in, is the path credibility from node u to node v, refers to the link credibility from node u to node v, Refers to the node credibility on the path. By calculating the credibility of the link and the credibility of the node, the credibility of each possible scheduling path can be obtained, and then the credibility function is used as the fitness function to constrain and guide the search process of the particle swarm algorithm. Now set the function , when the node exist superior is equal to 1, otherwise it is 0. The specific formula is as follows.

[0099]

[0100] in , and The weight coefficients are used to adjust the weights of link utilization, latency, and jitter based on the actual network path requirements to adjust the impact of different factors on link reliability. In this case, the values of all three are equal, 0.33. Indicates the average bandwidth utilization of the entire link, Indicates the number of segment links contained in the link. The main considerations are the link bandwidth utilization, link delay, and link jitter. The link bandwidth utilization calculation formula is as follows:

[0101]

[0102] in, , Indicates the number of bytes sent and received by each port. Representation node arrive Inter-link capacity.

[0103] The calculation of real-time link delay and jitter often requires additional mechanisms because the OpenFlow protocol itself does not provide these measurements. One-way delay can be calculated by sending probe packets and measuring the round-trip time (RTT). The SDN controller records a sending timestamp when the sending switch generates a probe packet. When the probe data packet arrives at the receiving switch, a receiving timestamp is recorded. The calculation formula for link delay is:

[0104]

[0105]

[0106] The link jitter calculation formula is:

[0107]

[0108] in It is The round trip time of a probe packet, is the average of all RTTs, and N is the number of probe packets.

[0109] Node credibility The calculation formula is as follows:

[0110]

[0111] in, Represents the actual throughput of the node, is the maximum theoretical throughput of the node, and is done this way to standardize nodes of different capabilities. is a node The packet loss rate ranges from 0 to 1. The smaller the value, the better. , is a weight factor that satisfies , to balance the importance of throughput and packet loss rate. The packet loss rate is calculated as follows:

[0112]

[0113] node The throughput is calculated as follows:

[0114]

[0115] in, i represents the number of bytes sent, Indicates the number of bytes received; Indicates the packet loss count, which is the number of packets lost due to reasons such as queue overflow.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0117] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A trusted dynamic load balancing method for a data center network based on SDN, characterized by: The following steps are involved: Obtain network topology information and set link load thresholds based on the network topology information; Obtaining a real-time link load of a link, and comparing the real-time link load of the link with a set link load threshold; triggering a load balancing mechanism when the real-time link load of the link exceeds the link load threshold; A link whose real-time link load exceeds the link load threshold is recorded as an overloaded link; The load balancing mechanism includes: Obtain the characteristics of the bandwidth-consuming elephant flows; Calculate new routing paths through particle swarm optimization; The SDN controller sends the updated flow table entries to the switches corresponding to the new routing paths; The switch switches the elephant flow from the overloaded link to the new path based on the updated flow table entry, and feeds back the link load status to the SDN controller in real time.

2. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The method of calculating a new routing path by using an improved particle swarm algorithm includes: converting a path credibility function into a fitness function; The path credibility function includes link credibility and node credibility, and is expressed as: ； is the path credibility from node u to node v, refers to the link credibility from node u to node v, refers to the node credibility of node i on the path, is the existence identification function of the node on the path; in, The calculation formula is: ； , and is the weight coefficient, Indicates the average bandwidth utilization of the entire link, Indicates the number of segment links contained in the link; Indicates the link delay, Indicates link flapping.

3. The trusted dynamic load balancing method for a data center network based on SDN according to claim 2, characterized in that: The node credibility The calculation formula is: ； For nodes The packet loss rate, Representative Node The actual throughput, is a node The maximum theoretical throughput, , is a weight factor that satisfies ; Among them, the node Packet loss rate The calculation formula is: ； node The throughput is calculated as: ； Indicates the number of bytes sent Indicates the number of bytes received; Indicates packet loss count; is the amount of data sent by node i, is the amount of data received by node i.

4. The trusted dynamic load balancing method for a data center network based on SDN according to claim 2, characterized in that: The link delay is calculated by sending a probe packet and calculating the round-trip time, the link delay of node i Expressed as: ； ； Link jitter of node i Expressed as: ； It is The round trip time of a probe packet, is the average of all RTTs, N is the number of probe packets, is the receiving timestamp of the i-th probe packet, is the sending timestamp of the i-th probe packet.

5. The trusted dynamic load balancing method for a data center network based on SDN according to claim 2, characterized in that: The bandwidth utilization of the entire link The calculation formula is: ； For the time interval The change in byte data sent by node i within , For the time interval The change in byte data received by node j within Representation node arrive Inter-link capacity.

6. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The calculation of the new routing path by the particle swarm algorithm includes: the speed of particle q The update formula is: ； ； is the inertia weight, and is the acceleration factor, and is a random number in the range [0,1], It is a particle The best position in history, is the historical best position of the entire particle group, It is a particle exist Current location at the moment, The random perturbation term introduced for the simulated annealing idea; in, Expressed as: ； yes The current temperature at the moment, is a random number in the range [-1,1]; By cooling factor Gradually reduce.

7. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The link load balancing degree is calculated using the first-order exponential smoothing method: ； is the smoothed value of the network load at time t obtained by weighted calculation, is the smoothed value at the previous moment, is a smoothing constant whose value range is [0,1]; N represents the total number of links in the network.

8. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The network topology adopts a Fat-tree structure with k=4. The objective function of load balancing is to minimize the maximum link utilization, which includes flow conservation constraints, bandwidth constraints, and non-negativity constraints. The formula is: ； Represents a network switch link set, represents the link capacity between nodes i and j; Indicates the network load balancing parameters at time t Greater than threshold When the elephant flow gathers; represents the rth elephant flow The original node, represents the rth elephant flow The destination node, represents the rth elephant flow bandwidth requirements; Indicates that the rth elephant flow is satisfied Traffic on link (i, j) under bandwidth demand.

9. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The dynamic weight of the particle swarm algorithm Adaptive adjustment based on the fitness function value, the formula is: ； and are the maximum and minimum values of the weight, and are the maximum and minimum values of the fitness function, respectively. The particle at the current moment The fitness function value of .

10. The trusted dynamic load balancing method for a data center network based on SDN according to claim 1, characterized in that: The probability that the particle swarm algorithm accepts an inferior solution during the search process It decreases as the temperature decreases. The specific formula is: ； For particles exist The fitness function value at the moment, is the current temperature.

Images