CN108183826A - A kind of Multiscale Fusion network simulation duty mapping method under isomerous environment - Google Patents

Abstract

The invention discloses a kind of Multiscale Fusion network simulation duty mapping methods under isomerous environment, solve deployment issue of the virtual network topology under the computing cluster environment of isomery, and the method step includes：Read in heterogeneous computing environment；Edge router in topology and host node mark are virtualized into map section for lightweight, are fusion virtualization map section by remaining node label；According to each Server throughput threshold value, mapping lightweight virtualization map section interior joint is virtualized with lightweight；Calculate the load balance parameter of remaining each server, fusion map section interior joint distributed using multilevel scheme partitioning algorithm, judge server whether redundancy, optimized using different optimization algorithms according to result and rationally mapped.The present invention ensures that the load balance between computing cluster simultaneously reduces telecommunication, the performance of large scale network emulation is promoted, while be with good expansibility to large-scale network topological and scalability, available for every network research and Experimental Network.

Description

Translated fromChinese

一种异构环境下的多尺度融合网络仿真任务映射方法A Multi-Scale Fusion Network Simulation Task Mapping Method in Heterogeneous Environment

技术领域technical field

本发明涉及网络仿真技术领域，特别是涉及一种异构环境下的多尺度融合网络仿真任务映射方法。The invention relates to the technical field of network simulation, in particular to a multi-scale fusion network simulation task mapping method in a heterogeneous environment.

背景技术Background technique

当前，基于虚拟化的云平台成为网络仿真的主流支撑平台：与传统网络模拟技术相比，该技术能够提供更加逼真的仿真环境，与实物测试床相比，该技术能够以非常低的成本轻易地仿真出大规模网络。网络与信息系统安全评估平台是网络安全评估与计算机系统安全评估的有力支撑，网络仿真技术是整个平台的基石。面向大规模、高逼真网络仿真需要，基于云平台与虚拟化的仿真技术已成为趋势。全虚拟化以KVM为代表，KVM虚拟化技术依赖物理CPU和内存，是硬件级别的，功能强大。轻量级虚拟化中以DOCKER容器作为代表，利用LXC实现了类似KVM的功能，能提供给用户更多的计算资源。DOCKER路由器和KVM路由器各有优劣，总的来说，KVM吞吐量高，延时短，性能稳定，但由于KVM路由器在保证正常工作的情况下启动个数受内存大小和逻辑CPU个数所限制，所以一个计算节点能够启动的KVM路由器个数较少；而DOCKER路由器启动速度快，不受内存大小和CPU个数的限制，一个高性能计算节点上能同时运行数千个DOCKER路由器，且在多个DOCKER工作的情况下，本身能做到负载均衡，但性能受启动个数而影响，且吞吐量小，延时稍长。Currently, the cloud platform based on virtualization has become the mainstream support platform for network simulation: compared with traditional network simulation technology, this technology can provide a more realistic simulation environment, and compared with physical test bed, this technology can easily simulate large-scale networks. The network and information system security assessment platform is a strong support for network security assessment and computer system security assessment, and network simulation technology is the cornerstone of the entire platform. Facing the needs of large-scale and high-fidelity network simulation, the simulation technology based on cloud platform and virtualization has become a trend. Full virtualization is represented by KVM. KVM virtualization technology relies on physical CPU and memory. It is hardware-level and powerful. Lightweight virtualization is represented by DOCKER containers, and LXC is used to implement functions similar to KVM, which can provide users with more computing resources. Docker routers and KVM routers have their own advantages and disadvantages. Generally speaking, KVM has high throughput, short delay, and stable performance. Therefore, the number of KVM routers that can be started by a computing node is small; while the DOCKER router starts quickly and is not limited by the memory size and the number of CPUs. Thousands of DOCKER routers can run on a high-performance computing node at the same time, and In the case of multiple dockers working, it can achieve load balancing, but the performance is affected by the number of startups, and the throughput is small and the delay is slightly longer.

发明内容Contents of the invention

为兼顾轻量级虚拟化与全虚拟化在仿真规模、仿真逼真性方面的各自优势，本发明基于OpenStack云平台，提出一种异构环境下的多尺度融合网络仿真任务映射方法，将所要仿真的网络拓扑分为轻量级虚拟化映射区和融合虚拟化映射区，并针对不同的映射区，分别借助优化的多级图划分工具METIS对网络拓扑进行合理映射。In order to take into account the respective advantages of lightweight virtualization and full virtualization in terms of simulation scale and simulation fidelity, the present invention proposes a multi-scale fusion network simulation task mapping method in a heterogeneous environment based on the OpenStack cloud platform. The network topology of the network is divided into a lightweight virtualization mapping area and a fusion virtualization mapping area, and for different mapping areas, the optimized multi-level graph division tool METIS is used to map the network topology reasonably.

所述异构环境下的多尺度融合网络仿真任务映射方法包括以下步骤：The multi-scale fusion network simulation task mapping method under the heterogeneous environment comprises the following steps:

S1：读取所需仿真的虚拟网络拓扑T(R,E)，E为路由之间的链路用E(R_i,R_j)表示，R为路由节点集合R＝{R₁,…,R_i,…R_N}，N为拓扑路由器个数，节点权值用W(R_i)表示，链路权值用W_E(R_i,R_j),R_j∈R表示；S1: Read the virtual network topology T(R,E) to be simulated, E is the link between routes represented by E(R_i ,R_j ), R is the set of routing nodes R={R₁ ,…, R_i ,…R_N }, N is the number of topological routers, the node weight is represented by W(R_i ), the link weight is represented by W_E (R_i ,R_j ),R_j ∈ R;

S2：将异构计算环境的服务器集群中CPU、内存与吞吐量均一致的服务器分为一组，服务器组数为Sn，依次提取每组异构计算环境参数,包括服务器组数Sn，每组计算节点的CPU核数，CPU_i,i＝1,2，…，Sn，内存大小Memory_i,i＝1,2，…，Sn，吞吐量阈值Throughput_i,i＝1,2，…，Sn以及每组计算环境的个数Num_i,i＝1,2，…，Sn，服务器的总个数S2: Divide the servers with the same CPU, memory and throughput in the server cluster of the heterogeneous computing environment into one group, the number of server groups is Sn, and sequentially extract the parameters of each group of heterogeneous computing environments, including the number of server groups Sn. Calculate the number of CPU cores of the node, CPU_i , i=1,2,...,Sn, memory size Memory_i ,i=1,2,...,Sn, throughput threshold Throughput_i ,i=1,2,...,Sn And the number Num_i of each computing environment, i=1, 2,..., Sn, the total number of servers

S3：将虚拟网络拓扑中的非终端路由器节点划分为融合虚拟化映射区T’(R,E)，T’中路由个数为N_T’，其余节点划分为轻量级虚拟化映射区T”(R,E)，T”中路由个数为N_T”；S3: Divide the non-terminal router nodes in the virtual network topology into a converged virtualization mapping area T'(R,E), the number of routes in T' is NT_' , and the remaining nodes are divided into a lightweight virtualization mapping area T "(R,E), the number of routes in T" is N_T" ;

S4：根据各服务器的吞吐量阈值Throughput_i，使用多级图划分算法将T”(R,E)划分为T”＝{T”₁,…,T”_i,…,T”_M}，其中，M为轻量级虚拟化映射区所使用的服务器个数，T”_i＝{R_i1,R_i2,……R_ixi}，x_i为T”_i中的路由器个数，且T”_i中所有节点的链路带宽和不超过承载此部分拓扑的第i个服务器的吞吐量阈值Throughput_i，即用轻量级虚拟化仿真T”(R,E)中节点；S4: According to the throughput threshold Throughput_i of each server, use the multi-level graph partition algorithm to divide T”(R,E) into T”={T”₁ ,…,T”_i ,…,T”_M }, where , M is the number of servers used in the lightweight virtualization mapping area, T”_i = {R_i1 , R_i2 ,…R_ixi }, x_i is the number of routers in T”_i , and The link bandwidth sum of all nodes in T”_i does not exceed the throughput threshold Throughput_i of the i-th server carrying this part of the topology, namely Emulate nodes in T”(R,E) with lightweight virtualization;

S5：剔除已分配完的服务器，将剩下的服务器集群中CPU、内存与吞吐量均一致的服务器分为一组，服务器组数为S，依次提取每组异构计算环境参数,包括服务器组数S，每组计算节点的CPU核数，CPU_i,i＝1,2，…，S，内存大小Memory_i,i＝1,2，…，S，吞吐量阈值Throughput_i,i＝1,2，…，S以及每组计算环境的个数Num_i,i＝1,2，…，S，服务器的总个数S5: Eliminate the allocated servers, divide the remaining servers in the server cluster with the same CPU, memory and throughput into one group, the number of server groups is S, and extract the parameters of each group of heterogeneous computing environments in turn, including server groups Number S, the number of CPU cores of each computing node, CPU_i , i=1,2,...,S, memory size Memory_i ,i=1,2,...,S, throughput threshold Throughput_i ,i=1, 2, ..., S and the number of each computing environment Num_i , i=1, 2, ..., S, the total number of servers

S6：计算剩余Ns-M个服务器的负载平衡参数S6: Calculate the load balancing parameters of the remaining Ns-M servers

判断服务器是否冗余，即满足若冗余则执行步骤S7，若不满足则执行步骤S8；Determine whether the server is redundant, that is, satisfy If it is redundant, execute step S7, and if not, execute step S8;

S7：根据各服务器的负载平衡参数LB_i，用多级图划分算法将T^’(R,E)划分为T’＝{T’₁,…,T’_i,…,T’_n}，其中n为刚大于N_Ti时，所用服务器的个数，采用全映射优化算法对结果进行优化，并将结果采用全虚拟化进行仿真；S7: According to the load balancing parameter LB_i of each server, divide T^' (R, E) into T'={T'₁ ,...,T'_i ,...,T'_n } with a multi-level graph partitioning algorithm, where n is When it is just larger than N_Ti , the number of servers used is optimized by the full-map optimization algorithm, and the result is simulated by full virtualization;

S8：根据各服务器的负载平衡参数LB_i，用多级图划分算法将T’(R,E)划分为T’＝{T’₁,…,T’_i,…,T’_n}，其中采用弹性映射优化算法对结果进行优化，并将结果采用全虚拟化与轻量级虚拟化融合的方式仿真；S8: According to the load balancing parameter LB_i of each server, divide T'(R,E) into T'={T'₁ ,...,T'_i ,...,T'_n } using a multi-level graph partitioning algorithm, where The elastic mapping optimization algorithm is used to optimize the results, and the results are simulated by the fusion of full virtualization and lightweight virtualization;

上述步骤中涉及的i，j没有特定物理意义且相互无关联，只代表所属范围内的自然数。The i and j involved in the above steps have no specific physical meaning and are not related to each other, and only represent natural numbers within the scope.

进一步的，步骤S7所述的全映射优化算法(即非终端路由器全虚拟化映射优化算法)包括以下步骤：Further, the full mapping optimization algorithm described in step S7 (i.e. non-terminal router full virtualization mapping optimization algorithm) includes the following steps:

S7.1、寻找映射当前结果的服务器中映射任务超出负载平衡的服务器T’_i(T’_i∈T’)；S7.1. Find the server T'_i (T'_i ∈ T') whose mapping task exceeds the load balance among the servers that map the current result;

S7.2、依次提取T’_i中映射的节点R_i，R_i的选取以T’_i中存在远程链路且点权最小的节点为先；S7.2, sequentially extract the nodes R_i mapped in T'_i , the selection of R_i is based on the node with the remote link in T'_i and the node with the smallest point weight first;

S7.3、在未达负载平衡的服务器T’_j(T’_j∈T’)中匹配与R_i存在链路的节点R_j；S7.3. Match the node R_j that has a link with R_i in the server T'_j (T'_j ∈ T') that has not reached load balance;

S7.4、将节点R_i迁移至未达负载平衡的服务器T’_j中；S7.4. Migrate the node R_i to the server T'_j that has not reached load balance;

S7.5、若所有T’_i中映射的节点与其余未达负载平衡的服务器中的所有节点均不存在链路，则将T’_i中存在远程链路且点权最小的节点R_i迁移至未达负载平衡的服务器T’_j中。S7.5. If there is no link between all nodes mapped in T'_i and all nodes in the rest of the servers that have not reached load balance, migrate the node R_i that has a remote link in T'_i and has the smallest point weight To the server T'_j which is not load balanced.

具体的，步骤S8所述的弹性映射优化算法(即轻量级虚拟化与全虚拟化融合映射优化算法)包括以下步骤：Specifically, the elastic mapping optimization algorithm described in step S8 (that is, the lightweight virtualization and full virtualization fusion mapping optimization algorithm) includes the following steps:

S8.1、寻找映射当前结果的服务器中映射任务未达负载平衡的服务器T’_i(T’_i∈T’)；S8.1. Find the server T'_i (T'_i ∈ T') whose mapping task does not reach the load balance among the servers of the current mapping result;

S8.2、依次提取T’_i中映射的节点R_i；S8.2, sequentially extracting the nodes R_i mapped in T'_i;

S8.3、在超出负载平衡的服务器T’_j(T’_j∈T’)中匹配与R_i存在链路的节点R_j，R_j的选取以T’_j中存在远程链路且点权最小的节点为先；S8.3. Match the node R j that has a link with R_i in the server T'_j (T'_j ∈ T') that exceeds the load balance. The selection of_{R jdepends} on the existence of a remote link in T'_j and the point weight The smallest node first;

S8.4、将节点R_j迁移至未达负载平衡的服务器T’_i中；S8.4. Migrate the node R_j to the server T'_i that has not reached load balance;

S8.5、若所有T’_i中映射的节点与其余超出负载平衡的服务器中的所有节点均不存在链路，则将超出负载平衡的服务器T’_i中存在远程链路且点权最小的节点R_j迁移至服务器T’_i中。S8.5. If there is no link between the mapped nodes in all_T'i and all nodes in the rest of the servers beyond load balance, then there will be remote links in the servers_T'i beyond load balance with the smallest point weight Node R_j migrates to server T'_i .

本发明的有益效果如下：The beneficial effects of the present invention are as follows:

1、充分利用两种虚拟化的优势映射虚拟网路拓扑，在保证负载平衡的情况下降低了计算集群间的远程通信量，能有效提高异构计算环境下虚拟网络映射的负载均衡度；1. Make full use of the advantages of two types of virtualization to map the virtual network topology, reduce the remote communication traffic between computing clusters while ensuring load balance, and effectively improve the load balance of virtual network mapping in heterogeneous computing environments;

2、在计算节点略微不足的情况下也可完成网络拓扑的映射，算法具有很好的伸缩性；2. In the case of a slight shortage of computing nodes, the mapping of the network topology can also be completed, and the algorithm has good scalability;

3、算法具有很好的可扩展性，适用于大规模异构计算环境以及大规模虚拟网络拓扑，且计算复杂度较低；3. The algorithm has good scalability, is suitable for large-scale heterogeneous computing environments and large-scale virtual network topologies, and has low computational complexity;

附图说明Description of drawings

图1为一种异构环境下的多尺度融合网络仿真任务映射方法流程图。Fig. 1 is a flowchart of a multi-scale fusion network simulation task mapping method in a heterogeneous environment.

图2为本发明实施例所使用的60节点网络拓扑图。Fig. 2 is a 60-node network topology diagram used in the embodiment of the present invention.

图3为60节点拓扑负载平衡度对比图。Figure 3 is a comparison diagram of 60-node topology load balance.

图4为60节点拓扑远程链路通信量对比图。Fig. 4 is a comparative diagram of traffic volume of remote links in 60-node topology.

图5为60节点拓扑各计算节点远程链路通信量对比图。Figure 5 is a comparison diagram of the remote link traffic of each computing node in the 60-node topology.

具体实施方式Detailed ways

下面结合附图和实施例，对本发明的具体实施方式做进一步描述。The specific implementation manner of the present invention will be further described below in conjunction with the drawings and embodiments.

本发明可用于异构计算环境下的网络模拟任务负载平衡，提升网络模拟性能，本发明所述方法包括以下步骤：The present invention can be used for load balancing of network simulation tasks in a heterogeneous computing environment to improve network simulation performance. The method of the present invention includes the following steps:

步骤S1中，读取的虚拟网络拓扑由GT-ITM生成，具有30个路由节点，并对拓扑中每一个度为1的节点链接6个主机节点。In step S1, the read virtual network topology is generated by GT-ITM, has 30 routing nodes, and links 6 host nodes to each node with degree 1 in the topology.

步骤S2中，读入的每组计算环境参数包括服务器组数Sn，每组计算节点的CPU核数CPU_i，内存大小Memory_i，吞吐量阈值Throughput_i以及每组计算环境的个数Num_i，其中i＝1,2，…，Sn。In step S2, each set of computing environment parameters read includes the number of server groups Sn, the number of CPU cores CPU_i of each computing node, the memory size Memory_i , the throughput threshold Throughput_i , and the number Num_i of each computing environment, where i=1, 2, . . . , Sn.

本实例部署的OpenStack包括一个控制节点、一个网络节点和七个计算节点。七个计算节点中包含三台24核16G吞吐量阈值大约为25000(10^6bit/sec)的计算节点、两台12核16G吞吐量阈值大约为21000(10^6bit/sec)的计算节点以及两台24核32G吞吐量阈值大约为26000(10^6bit/sec)的计算节点。服务器组数Sn＝3，其中，第一组服务器的参数为CPU₁＝24，Memory₁＝16，Throughput₁＝25000，Num₁＝3；第二组服务器的参数为CPU₂＝12，Memory₂＝16，Throughput₂＝21000，Num₂＝2；第三组服务器的参数为CPU₃＝24，Memory₃＝32，Throughput₃＝26000，Num₃＝2；The OpenStack deployed in this example includes a control node, a network node, and seven computing nodes. The seven computing nodes include three 24-core 16G computing nodes with a throughput threshold of about 25,000 (10^6bit/sec), two 12-core 16G computing nodes with a throughput threshold of about 21,000 (10^6bit/sec), and Two 24-core 32G computing nodes with a throughput threshold of about 26000 (10^6bit/sec). The number of server groups Sn=3, wherein the parameters of the first group of servers are CPU₁ =24, Memory₁ =16, Throughput₁ =25000, Num₁ =3; the parameters of the second group of servers are CPU₂ =12, Memory₂ =16, Throughput₂ =21000, Num₂ =2; the parameters of the third group of servers are CPU₃ =24, Memory₃ =32, Throughput₃ =26000, Num₃ =2;

步骤S3中，将虚拟网络拓扑中的终端路由器节点及其相连的主机节点划分到轻量级虚拟化映射区T_l(R,E)，T_l中路由个数N_Tl＝5，其余节点划分到融合虚拟化映射区T_i(R,E)，T_i中路由个数N_Ti＝25；In step S3, the terminal router node in the virtual network topology and its connected host nodes are divided into a lightweight virtualization mapping area_T1 (R, E), the number of routes in_T1 is N_T1 = 5, and the remaining nodes are divided into To the merged virtualization mapping area T_i (R, E), the number of routes N_Ti in T_i =25;

步骤S4中，根据各服务器的吞吐量阈值，将T_l(R,E)划分为T_l＝{T_l1,…,T_li,…,T_lM}，用轻量级虚拟化映射T_l(R,E)中节点。在60节点的虚拟网络拓扑中，终端路由的总吞吐量所以轻量级虚拟化映射区的划分份数M＝1。在大规模虚拟网络拓扑中，轻量级虚拟化映射区的划分份数In step S4, according to the throughput threshold of each server, divide T_l (R, E) into T_l ={T_l1 ,...,T_li ,...,T_lM }, and map T_l ( Nodes in R, E). In a 60-node virtual network topology, the total throughput of terminal routes Therefore, the division number M=1 of the lightweight virtualization mapping area. In a large-scale virtual network topology, the number of divisions of the lightweight virtualization mapping area

步骤S5中，剩余各服务器的负载平衡参数LB₁＝4，LB₂＝4，LB₃＝8，服务器冗余，执行S6。In step S5, the load balancing parameters of the remaining servers are LB₁ =4, LB₂ =4, LB₃ =8, Server redundancy, execute S6.

步骤S6中，将剩余各服务器的负载平衡参数组成的列表(L_Ws＝{4,4,4,4,8,8})作为权重，用多级图划分算法将T_i(R,E)初始划分为T_i＝{T_i1,…,T_ii,…,T_iN}，其中N＝6，划分时每组服务器所占比例然后采用全映射优化算法对结果进行优化，首先，依次判断T_ii中的路由个数N_Ri是否满足将不满足条件的结果T_ii提出，依次提取T_ii中的节点R_i，并寻找节点R_i与的结果T_ij中的节点是否存在链路，若存在链路且点权值为所有存在链路的节点中最小，则将节点R_i迁移至结果T_ij中。若遍历结果，不存在符合迁移条件的节点，则将T_ii节点中存在远程链路且点权值最小的节点R_i迁移至直至的结果T_ij中，直至所有结果均满足将结果采用全虚拟化映射；In step S6, the list (L_Ws ={4,4,4,4,8,8}) composed of the load balancing parameters of the remaining servers is used as the weight, and the T_i (R, E) The initial division is T_i ={T_i1 ,…,T_ii ,…,T_iN }, where N=6, the proportion of each group of servers when dividing Then use the full mapping optimization algorithm to optimize the results. First, judge whether the number of routes N_Ri in T_ii satisfies Propose the result T_ii that does not meet the conditions, extract the node R_i in T_ii in turn, and find the node R_i and Whether there is a link in the node in the result T_ij of the result, if there is a link and the point weight is the smallest among all the nodes with the link, then the node R_i is migrated to the result T_ij . If there is no node that meets the migration conditions in the traversal results, the node R_i that has a remote link and the smallest point weight in the T_ii node is migrated to until Among the results T_ij , until all the results satisfy Apply the result to full virtualization mapping;

步骤S7中，若剩余服务器个数不足，假设映射轻量级虚拟化映射区中节点需要的服务器个数M＝2，则剩余各服务器的负载平衡参数组成的列表变为L_Ws＝{4,4,4,8,8}，用多级图划分算法将T_i(R,E)初始划分为T_i＝{T_i1,…,T_ii,…,T_iN}，其中划分时每组服务器所占比例采用弹性映射优化算法对结果进行优化，首先，依次判断T_ii中的路由个数N_Ri是否满足将不满足条件的结果T_ii提出，依次提取T_ii中的节点R_i，并寻找节点R_i与的结果T_ij中的节点R_j是否存在链路，若存在链路且点权值为所有存在链路的节点中最小，则将节点R_j迁移至结果T_ii中。若遍历结果，不存在符合迁移条件的节点，则将的结果T_ij中存在远程链路且点权值最小的节点R_j迁移至直至T_ii中。直至所有结果均满足并将的结果采用全虚拟化的方式映射，将的结果采用映射轻量级虚拟化的方式映射。In step S7, if the number of remaining servers is insufficient, assuming that the number of servers M=2 required for mapping the nodes in the lightweight virtualization mapping area, the list formed by the load balancing parameters of the remaining servers becomes L_Ws ={4, 4,4,8,8}, use the multi-level graph partition algorithm to initially divide T_i (R, E) into T_i ={T_i1 ,…,T_ii ,…,T_iN }, where The proportion of each group of servers when dividing Use the elastic mapping optimization algorithm to optimize the results. First, judge whether the number of routes N_Ri in T_ii satisfies Propose the result T_ii that does not meet the conditions, extract the node R_i in T_ii in turn, and find the node R_i and Whether the node R_j in the result T_ij has a link, if there is a link and the point weight is the smallest among all the nodes with links, then the node R_j will be migrated to the result T_ii . If there is no node that meets the migration conditions in the traversal results, then the The result is that there is a remote link in T_ij and the node R_j with the smallest point weight migrates to T_ii . until all results are met and will The results are mapped in a fully virtualized manner, and the The results are mapped in a light-weight virtualized way.

至此完成异构环境下的多尺度融合网络仿真任务映射方法，实现了异构环境下的多尺度融合网络仿真任务的分配，基于上述步骤映射的虚拟网络可以开展各种测试研究工作。So far, the multi-scale fusion network simulation task mapping method in a heterogeneous environment has been completed, and the distribution of multi-scale fusion network simulation tasks in a heterogeneous environment has been realized. Based on the virtual network mapped by the above steps, various testing and research work can be carried out.

基于本实施例，可展开但不限于以下测试工作：Based on this embodiment, the following test work can be carried out but not limited to:

(1)在虚拟网络拓扑上分别找两台挂在不同终端路由器下的云主机，两台云主机互相执行ping命令，结果表明映射后的网络拓扑仍具备了转发网络流量的能力；(1) On the virtual network topology, find two cloud hosts connected to different terminal routers, and execute the ping command between the two cloud hosts. The result shows that the mapped network topology still has the ability to forward network traffic;

(2)比较本发明提供的异构算法和同构算法的加权负载平衡如图3所示，负载平衡度越接近于零，则越负载平衡。由于同构算法无法根据不同服务器的权值进行拓扑的映射，所以同构映射算法的服务器参数L_Ws只能选取L_Ws＝min(L_Wsi)，本方法的负载平衡度较同构算法好。(2) compare the weighted load balancing of the heterogeneous algorithm provided by the present invention and the homogeneous algorithm As shown in FIG. 3 , the closer the load balance degree is to zero, the more balanced the load is. Since the isomorphic algorithm cannot perform topology mapping according to the weights of different servers, the server parameter L_Ws of the isomorphic mapping algorithm can only be selected as L_Ws =min(L_Wsi ), and the load balance of this method is better than that of the isomorphic algorithm.

(3)比较本发明提供的异构算法和同构算法形成的远程链路，如图4，本方法的远程链路通信量要小于同构算法形成的远程链路通信量。(3) Comparing the remote link formed by the heterogeneous algorithm provided by the present invention and the isomorphic algorithm, as shown in FIG. 4 , the remote link communication volume of this method is smaller than the remote link communication volume formed by the isomorphic algorithm.

上述实验结果表明，本发明通过提供一种异构环境下的多尺度融合网络仿真任务映射方法，实现在异构环境下对虚拟网络拓扑的合理映射，保证计算集群间的负载平衡以及降低计算集群间的远程通信，降低算法运行时间，提升大规模网络仿真的性能，如下表1所示，同时对大规模网络拓扑具有良好的可扩展性，对所仿真的网络拓扑具有良好的可伸缩性，可用于各项网络研究与实验网络。The above experimental results show that, by providing a multi-scale fusion network simulation task mapping method in a heterogeneous environment, the present invention realizes a reasonable mapping of the virtual network topology in a heterogeneous environment, ensures load balance between computing clusters and reduces the load of computing clusters. The long-distance communication among them reduces the running time of the algorithm and improves the performance of large-scale network simulation, as shown in Table 1 below. At the same time, it has good scalability for large-scale network topology and good scalability for the simulated network topology. It can be used in various network research and experimental networks.

表1大规模虚拟网络拓扑各评估参数对比表Table 1 Comparison table of evaluation parameters of large-scale virtual network topology

同构算法isomorphic algorithm异构算法heterogeneous algorithm远程通信量remote traffic351230351230366508366508负载均衡度load balance35.08935.0890.0190.019算法运行时间Algorithm running time11.85411.8542.3932.393

上述具体实施方案仅用于说明本发明，而并非对本发明的限制，在本领域普通技术人员所具备的知识范围内，还可以在不脱离本发明宗旨的前提下做出各种变化。The above-mentioned specific embodiments are only used to illustrate the present invention, not to limit the present invention. Within the knowledge of those skilled in the art, various changes can be made without departing from the gist of the present invention.

Claims (3)

1. a kind of Multiscale Fusion network simulation duty mapping method under isomerous environment, which is characterized in that include the following steps：

S1：The virtual network topology T (R, E) emulated needed for reading, E are the link E (R between routing_i,R_j) represent, R is roadBy node set R={ R₁,…,R_i,…R_N, N be topological router number, node weights W (R_i) represent, link metric W_E(R_i,R_j),R_j∈ R are represented；

S2：The consistent server of CPU, memory and handling capacity in the server cluster of heterogeneous computing environment is divided into one group, clothesDevice group of being engaged in number is Sn, every group of heterogeneous computing environment parameter is extracted successively, including server group number Sn, the CPU of every group of calculate nodeCheck figure, CPU_i, i=1,2 ..., Sn, memory size Memory_i, i=1,2 ..., Sn, throughput threshold Throughput_i, i=The number Num of 1,2 ..., Sn and every group of computing environment_i, i=1,2 ..., Sn, the total number of server

S3：Nonterminal router node in virtual network topology is divided into fusion virtualization map section T ' (R, E), T ' RoadsIt is N by number_T’, remaining node division is lightweight virtualization map section T " (R, E), and it is N that number is route in T "_T”；

S4：According to the throughput threshold Throughput of each server_i, T " (R, E) is divided into T " using multilevel scheme partitioning algorithm={ T "₁,…,T”_i,…,T”_M, wherein M virtualizes server number used in map section for lightweight,_iFor T "_iIn router number,AndT”_iIn all sectionsThe link bandwidth of point and the throughput threshold Throughput no more than i-th of server for carrying this partial-topology_i, i.e.,Emulation T " (R, E) interior joint is virtualized with lightweight；

S5：The server distributed is rejected, by the consistent service of CPU, memory and handling capacity in remaining server clusterDevice is divided into one group, and server group number is S, every group of heterogeneous computing environment parameter is extracted successively, including server group number S, every group of meterThe CPU core number CPU of operator node_i, i=1,2 ..., S, memory size Memory_i, i=1,2 ..., S, throughput thresholdThroughput_i, the number Num of i=1,2 ..., S and every group of computing environment_i, i=1,2 ..., S, the total number of server

S6：Calculate the load balance parameter of remaining Ns-M server

Judge server whether redundancy, that is, meetStep S7 is performed if redundancy, if being unsatisfactory for performingStep S8；

S7：According to the load balance parameter LB of each server_i, with multilevel scheme partitioning algorithm by T ' (R, E) be divided into T '={T’₁,…,T’_i,…,T’_n, wherein n isJust over N_TiWhen, the number of server used is optimized using full mappingAlgorithm optimizes result, and result is emulated using fully virtualized；

S8：According to the load balance parameter LB of each server_i, with multilevel scheme partitioning algorithm by T ' (R, E) be divided into T '={T’₁,…,T’_i,…,T’_n, whereinResult is optimized using elasticity mapping optimization algorithm, and willAs a result it is emulated by the way of virtualization is merged with lightweight using fully virtualized；

I involved in above-mentioned steps, j do not have specific physical significance and the connection that is independent of each other, only the natural number belonging to representative in range.

2. the Multiscale Fusion network simulation duty mapping method under isomerous environment as claimed in claim 1, feature existIn the full mapping optimization algorithm described in step S7 includes the following steps：

S7.1, the server T ' that mapping tasks in the server for mapping current results exceed load balance is found_i(T’_i∈T’)；

S7.2, T ' is extracted successively_iThe node R of middle mapping_i, R_iSelection with T '_iIt is middle there are remote link and the minimum node of point powerFor elder generation；

S7.3, do not reach load balance server T '_j(T’_j∈ T ') in match and R_iThere are the node Rs of link_j；

S7.4, by node R_iIt migrates to the server T ' for not reaching load balance_jIn；

If S7.5, all T '_iChain is not present in all nodes that the node of middle mapping is not reached with remaining in the server of load balanceRoad, then by T '_iIt is middle there are remote link and the minimum node R of point power_iIt migrates to the server T ' for not reaching load balance_jIn.

3. the Multiscale Fusion network simulation duty mapping method under isomerous environment as claimed in claim 1, feature existIn the elasticity mapping optimization algorithm described in step S8 includes the following steps：

S8.1, the server T ' that mapping tasks in the server for mapping current results do not reach load balance is found_i(T’_i∈T’)；

S8.2, T ' is extracted successively_iThe node R of middle mapping_i；

S8.3, in the server T ' beyond load balance_j(T’_j∈ T ') in match and R_iThere are the node Rs of link_j, R_jSelectionWith T '_jThe middle node there are remote link and point power minimum is first；

S8.4, by node R_jIt migrates to the server T ' for not reaching load balance_iIn；

If S8.5, all T '_iChain is not present in all nodes that the node of middle mapping exceeds with remaining in the server of load balanceRoad, then will be beyond the server T ' of load balance_iIt is middle there are remote link and the minimum node R of point power_jIt migrates to server T '_iIn.