Data center network bandwidth allocation and flow optimization method based on space game
By applying spatial game evolution laws and multi-time window assignment group-dependent scheduling in data center networks, the uneven resource allocation and QoS fluctuations caused by dynamic traffic are solved, bandwidth and traffic optimization are achieved, and resource utilization is improved. and network stability.
Patent Information
- Application Number
- CN202411345268.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-26
- Publication Date
- 2024-12-06
AI Technical Summary
When facing dynamic traffic, existing data center networks have uneven resource allocation, QoS fluctuations and unpredictable problems, resulting in insufficient resource utilization and reduced fairness, and the inability to achieve bandwidth minimization and traffic optimization.
Using the data center network bandwidth allocation and traffic optimization method based on spatial game, by building a cloud computing data center model, using spatial game evolution laws and multi-time window assigned groups to rely on network bandwidth scheduling to ensure the full utilization of bandwidth and traffic resources, and Reduce QoS fluctuations.
It realizes bandwidth minimization and traffic optimization in dynamic data center networks, improves resource utilization and network stability, ensures fairness and predictability of resource allocation, and reduces time costs.
Smart Images

Figure CN119094475A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to a data center network bandwidth allocation and flow optimization method based on space game. Background Art
[0002] Cloud computing is a type of distributed computing, which refers to the decomposition of huge data computing processing programs into countless small programs through the network "cloud". Then, these small programs are processed and analyzed by a system composed of multiple servers to obtain the results and return them to the user. The basic architecture of the cloud computing network is that each data center is composed of service clusters connected by a cloud network or a data center network. Through virtualization technology, physical servers can allocate their computing power, storage space, memory and other resources to host multiple virtual servers (also known as virtual machines, VMs) with customized resource requirements. Cloud service providers allocate these virtual machines to users (also known as tenants) who need cloud computing services. A tenant can rent multiple virtual machines on different physical servers. At the same time, multiple virtual machines on the same physical server may also belong to different tenants. From the perspective of cloud computing data center flow, dynamic, non-fixed bandwidth and minimum traffic are generally the best choices. At present, the existing bandwidth allocation and traffic optimization methods in data centers have problems such as being unable to respond to real-time changing user needs, unable to guarantee bandwidth minimization, and unable to optimize traffic allocation solutions, resulting in insufficient resource utilization and reduced fairness. In actual applications, the dynamic nature of data center network tenant network traffic may face the following problems:
[0003] 1. Uneven resource allocation: Due to the dynamic nature of traffic, some tenants may generate a lot of traffic at certain times, while less traffic at other times. This imbalance may lead to unreasonable resource allocation. Some tenants may get too much bandwidth resources, while other tenants may be restricted.
[0004] 2. QoS fluctuations: Dynamic traffic may cause large fluctuations in the quality of service (QoS) in the network. When some tenants generate a large amount of traffic, it may affect the network performance and service quality of other tenants, leading to increased latency, bandwidth congestion and other problems, reducing the stability and reliability of the overall network.
[0005] 3. Unpredictability: The dynamic nature of network traffic makes it difficult to predict, which brings challenges to network management and resource scheduling. Difficulties in accurately predicting traffic changes may lead to insufficient or excessive resource allocation, affecting network performance and efficiency. Summary of the invention
[0006] In the two cases of broadband utility consistency and inconsistency, the spatial evolution dynamics law is used to complete the resource allocation problem of data center network bandwidth and traffic. According to the spatial game evolution law, the group-dependent network bandwidth scheduling problem with multiple completion time windows assignment is adopted to ensure the full utilization of network bandwidth and traffic resources of data center network tenants, while keeping QoS fluctuations reduced and improving the stability of the overall network.
[0007] The present invention provides a data center network bandwidth allocation and traffic optimization method based on space game, and the specific steps of the data center network bandwidth allocation and traffic optimization method based on space game are as follows:
[0008] Step 1: Construct a cloud computing data center model framework, assuming that the participants in the game are available physical machines m i , the cloud data center has N different physical machines {m1,m2,…,m n The number of available resources for each physical machine is The network bandwidth is B(m i ), for a cloud computing data center, there are L tasks at each decision moment, and each subtask l needs to be allocated a virtual unit. The resource vector is The maximum response time of the task is RT i ;
[0009] Step 2: Modeling of bandwidth allocation and traffic optimization in spatial game data center networks, defining constraints on network resources: C s Refers to the actual maximum communication capacity of the physical link s. In the presence of ω communication links, VM i1 ,VM i2 ,…,VM iω To represent the communication between VMs, the attached path P i1 ,P i2 ,…P iω , the bandwidth allocation matrix is R k ,The space-time game model is considered from the following four aspects:
[0010] (a) Tenant K’s communication starts from become at this time becomes
[0011] (b) Communication of Tenant K Terminate, at this time Automatic deletion;
[0012] (c) Tenant K adds new communication A new path will be added to the network
[0013] (d) Tenant K’s communication from Remains unchanged, but actual bandwidth and traffic change;
[0014] Step 3: Pre-combination stage. In the data center, the physical machines with available resources are regarded as the main players in the game. For each physical server, by traversing possible virtual machine combination schemes, the different types of virtual machines that each server can carry are determined, while ensuring that the server capacity limit is not exceeded. For each physical machine, the type combination schemes of the virtual machines to be created are counted;
[0015] Step 4: Participant strategy space game generation, first for each physical server m i , generate possible bandwidth and traffic allocation schemes, denoted as O(m i ) This set contains various possible allocation schemes, and then from O(m i ) select the first g combinations {com1,com2,…,com g} and calculate the minimum value minO(m i ), and finally each com g (m i ) into A g (m i );
[0016] Step 5: Expand the directed generation number of the spatial game. First, sort the original allocation strategies according to minO(m1),…,minO(M) to obtain the index sequence {l1,…l M}, and then the participants follow the sorted order {l1,…l M}Select the game strategy in turn, that is, select the bandwidth and traffic allocation scheme of the central network according to the performance indicators of the physical machine;
[0017] Step 6: Solve the optimal strategy for the spatial game. For the physical server m in step 4, calculate the strategy with the maximum utility under the given game. According to the calculated results, select the best strategy for each physical server m and add it to the selected server. Finally, each participant l m Select the optimal bandwidth and traffic allocation matrix A(m) from the strategy set of the spatial game i ).
[0018] Compared with the prior art, the beneficial effects of the present invention are: a method for bandwidth allocation and traffic optimization of data centers based on spatial games is proposed, which provides a new solution to the problems of uneven bandwidth allocation among multiple tenants, traffic instability, QoS fluctuations, etc. in resource allocation of data centers in cloud computing. The spatial dynamic evolutionary game is used to design an allocation strategy matrix including bandwidth allocation and traffic optimization, and spatial evolutionary games for both broadband utility consistency and inconsistency are designed in the dynamic data center network to ensure minimum bandwidth and optimal traffic, while improving the allocation strategy of prediction fairness and resource utilization. At the same time, the group-dependent scheduling problem of the assignment of time windows is completed at the task time level to minimize the time cost. A new method from the perspective of spatial games is provided for subsequent research on bandwidth allocation and traffic optimization of data centers in cloud computing. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 A flowchart of a data center network bandwidth allocation and traffic optimization method based on spatial game for this application;
[0020] Figure 2 Task allocation structure diagram for data center network;
[0021] Figure 3 Figure 1. Competition for link bandwidth resources for data center network traffic. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0023] Game theory is a direction of mathematical theory research. It mainly solves decision-making problems and finds the optimal solution by studying the rational interaction between multiple game participants. In the data center network, the spatial game method is used to design the bandwidth allocation strategy, and the data center network bandwidth allocation strategy with dynamic minimum bandwidth guarantee, full resource utilization, predictability and fairness is studied. Specifically, individuals in the space randomly select new strategies from all strategies. Assuming that the benefits of the strategy are greater than the average benefits of the entire group, then the individual adopts this new strategy with a probability proportional to the difference between the two benefits; otherwise, it remains unchanged. In this spatial evolutionary game mode, the state protocol of the entire group is adjusted as follows:
[0024]
[0025] According to the average kinetics formula:
[0026]
[0027] The spatial game dynamics equation is obtained:
[0028]
[0029] Among them, U represents a population of evolutionary games, τ is the state adjustment protocol, Representative strategy The proportion of each strategy in the population. The change in the proportion of each strategy in the average evolutionary dynamics It is equal to the proportion of individuals with other strategies converted to strategy i minus the proportion of individuals with strategy i converted to individuals with other strategies.
[0030] Example 1
[0031] exist Figure 1 In the cloud computing data center model framework, step 1 represents the optional strategy of multi-tenants to select their own suitable hosts from the available physical machines in the cloud data center to create their own virtual unit types. Step 2 is the modeling of network bandwidth allocation and traffic optimization in the spatial game data center, which describes four data center bandwidth allocation and traffic optimization situations: (a) Tenant K's communication from become at this time becomes (b) Communication of Tenant K Terminate, at this time Automatically deleted. (c) Tenant K adds new communication A new path will be added to the network (d) Tenant K’s communication from The bandwidth and traffic remain unchanged, but the actual bandwidth and traffic change. Step 3: Pre-combination stage. In the data center, the physical machines with available resources are regarded as the main players in the game. Specifically, for each physical server, by traversing the possible virtual machine combination schemes, the different types of virtual machines that each server can carry are determined, while ensuring that the server capacity limit is not exceeded. Step 4: Participant strategy space game generation. First, for each physical server m i , generate possible bandwidth and traffic allocation schemes, denoted as O(m i ) This set contains various possible allocation schemes. The actual needs of users are multifaceted. Assuming that the dimension of the demand is d, the user's minimum resource demand is The minimum requirement for the space generation game is The minimum and maximum normalization method can be used to linearly transform the values of task requirements in the task requirement matrix TR to eliminate the influence of different dimensions on the scheduling results.
[0032] After normalization, each tr in the task requirement matrix TR ik(1≤i≤L,1≤k≤d) will satisfy 0≤tr ik ≤1.
[0033] Step 5: Expand the directed generation number of the spatial game. First, sort the original allocation strategies according to minO(m1),…,minO(M) to obtain the index sequence {l1,…l M}. Then the participants follow the sorted order {l1,…l M}Select the game strategies one by one.
[0034] Step 6: Solve the optimal strategy for the spatial game. The specific detailed steps are as follows: For the last participant l M , calculate the strategy that maximizes utility max[x],argMax y Utility(l M )[x][y]. For l M-1 , select l M The strategy selection with the maximum utility[l M-1 ], that is, argMax x U(l M-1 )[x][max[x]]. For l M , select the strategy that maximizes utility selection[l M ], that is, argMax x [max[x]]. (6-4) M-1 Add to the list of selected physical servers. M-2 To each physical server m of the first participant 1: add up the total amount of resources in the selected physical servers. For each strategy A of physical server m g (m): If strategy A is selected g (m), then calculate its skewness value ske(m). Calculate the total allocated resources equal to strategy A g (m) plus φ. For each i from 1 to L: If tri and A are calculated g If the similarity between (m) is greater than or equal to the threshold θ, then AR is selected. i Calculate the fair deviation value v(A) of each game allocation scheme. Generate a profit matrix for utility calculation. m , select the strategy selection[m] that maximizes utility, i.e. argMax x Utility(m)[x]. Add server m to the list of selected physical servers. Finally, each participant l m The optimal strategy in the selection set selection[m] can be expressed as an allocation matrix A(m i ), while A* ={A * (m1),A * (m2),…,A * (m M )} represents the distribution matrix representation of the subgame Nash equilibrium solution.
[0035] exist Figure 2 The task allocation structure diagram of the data center network can be seen in the figure. The participants in the game are the available physical machines m i , the cloud data center has N different physical machines {m1,m2,…,m n The number of available resources for each physical machine is The network bandwidth is B(m i ), for a cloud computing data center, there are L tasks at each decision moment, and each subtask l needs to be allocated a virtual unit. A resource supply S(m i )=(S1(m i ),S2(m i ),…,S i (m i )),S i ∈N represents the number of virtual unit types required to create task i, and the physical machine resource allocation strategy is the allocation matrix A(m i ).
[0036] exist Figure 3 In the data center network, traffic competes for link bandwidth resources. There are M data links in the data center network, where the bandwidths are C = {C1, C1, …, C M}. On these M data links, there are N data flows F={F1,F2,…,F N}, define the bandwidth vector and allocation vector required by each data flow as D t =D t1 ,D t2 ,…,D tN ; B t =B t1 ,B t2 ,…,B tN , considering the allocation of instantaneous bandwidth demand at a specific moment, define the reference bandwidth vector B 0 , so that each data flow can obtain the minimum bandwidth and provide the minimum QoS guarantee. For the matrix F = (f ij ) N×M Indicates the distribution of each data flow in the DC network. The principle of bandwidth allocation is that if the bandwidth demand of a flow is less than the base bandwidth, its demand is met; otherwise, the bandwidth that meets the bandwidth allocation fairness requirement is allocated. The upper and lower limits of bandwidth allocation are defined as: but
[0037] The invention patent is based on the data center network bandwidth allocation and traffic optimization of spatial games. It models the network in cloud computing data, uses the dynamic evolution method of spatial games to design algorithm bandwidth allocation strategies, and studies data center network bandwidth allocation strategies with dynamic minimum bandwidth guarantee, full resource utilization, predictability and fairness. The spatial evolution game models are used in the two cases of bandwidth utility consistency and inconsistency respectively to study the rules of bandwidth allocation games under bandwidth utility consistency. On this basis, the network bandwidth scheduling problem with group dependencies and multi-completion time window assignments is further studied. The experimental platform uses Google Cluster Trance as the reference data for data center workloads. The experiments have the following
[0038] Assumptions:
[0039] (1) The experiment only considers two data types, namely CPU and memory;
[0040] (2) The user request provides the virtual machine types corresponding to all task types;
[0041] (3) The cloud platform will determine in advance how much resources are available for allocation before each decision.
[0042] Experiments show that the method of the present invention has a significant improvement on bandwidth allocation and traffic optimization of data centers, can achieve dynamic bandwidth minimization of data center network bandwidth allocation, and ensure predictability and fairness of resource utilization.
[0043] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A data center network bandwidth allocation and traffic optimization method based on spatial game, characterized by: The specific steps of the data center network bandwidth allocation and traffic optimization method based on spatial game are as follows: Step 1: Construct a cloud computing data center model framework, assuming that the game participants are available physical machines m i , the cloud data center has N different physical machines {m1,m2,…,m n The number of available resources for each physical machine is The network bandwidth is B(m i ), for a cloud computing data center, there are l tasks at each decision moment, and each subtask l needs to be allocated a virtual unit. The resource vector is The maximum response time of a task is RT i ; Step 2: Modeling of bandwidth allocation and traffic optimization in spatial game data center networks, defining constraints on network resources: C s Refers to the actual maximum communication capacity of the physical link s. In the presence of ω communication links, VM i1 ,VM i2 ,…,VM iω To represent the communication between VMs, the attached path P i1 ,P i2 ,…P iω , the bandwidth allocation matrix is R k ,The space-time game model is considered from the following four aspects: (a) Tenant K’s communication starts from become at this time becomes (b) Communication of Tenant K Terminate, at this time Automatic deletion; (c) Tenant K adds new communication A new path will be added to the network (d) Tenant K’s communication from Remains unchanged, but actual bandwidth and traffic change; Step 3: Pre-combination stage. In the data center, the physical machines with available resources are regarded as the main players in the game. For each physical server, by traversing possible virtual machine combination schemes, the different types of virtual machines that each server can carry are determined, while ensuring that the server capacity limit is not exceeded. For each physical machine, the type combination schemes of the virtual machines to be created are counted; Step 4: Participant strategy space game generation, first for each physical server m i , generate possible bandwidth and traffic allocation schemes, denoted as O(m i ) This set contains various possible allocation schemes, and then from O(m i ) select the first g combinations {com1,com2,…,com g } and calculate the minimum value minO(m i ), and finally each com g (m i ) into A g (m i ); Step 5: Expand the directed generation number of the spatial game. First, sort the original allocation strategies according to minO(m1),…,minO(M) to obtain the index sequence {l1,…l M }, and then the participants follow the sorted order {l1,…l M }Select the game strategy in turn, that is, select the bandwidth and traffic allocation scheme of the central network according to the performance indicators of the physical machine; Step 6: Solve the optimal strategy for the spatial game. For the physical server m in step 4, calculate the strategy with the maximum utility under the given game. According to the calculated results, select the best strategy for each physical server m and add it to the selected server. Finally, each participant l m Select the optimal bandwidth and traffic allocation matrix A(m) from the strategy set of spatial game i ).
Citation Information
Cited By
Continuous verification method and system during data backup
CN120144368A
A Continuity Verification Method and System during Data Backup
CN120144368B
