技术领域technical field
本发明属于虚拟网络技术领域，更为具体地讲，涉及一种业务需求感知的虚拟数据中心映射方法。The invention belongs to the field of virtual network technology, and more specifically, relates to a virtual data center mapping method with service demand perception.
背景技术Background technique
硬件虚拟化技术以及KVM(Kernel-basedVirtualMachine，基于内核的虚拟机)、OpenVZ(Operatingsystem-levelvirtualization，操作系统虚拟化)、QEMU等虚拟机技术的发展使资源共享能够以更加方便、安全、灵活的方式进行。在云数据中心，用户的计算任务通常由一组虚拟机通过协作完成。用户的需求可以描述为：请求一组拥有指定资源的虚拟机，并为虚拟机之间配备指定的网络带宽。这样的用户需求可以看作是对虚拟数据中心的请求，虚拟数据中心被定义为一组拥有虚拟链路需求和物理主机资源需求的虚拟机集合。Hardware virtualization technology and the development of virtual machine technologies such as KVM (Kernel-basedVirtualMachine, kernel-based virtual machine), OpenVZ (Operating system-level virtualization, operating system virtualization), QEMU, etc. enable resource sharing in a more convenient, safe and flexible way conduct. In a cloud data center, a user's computing task is usually completed by a group of virtual machines through cooperation. The user's demand can be described as: request a group of virtual machines with specified resources, and provide specified network bandwidth between the virtual machines. Such user requirements can be regarded as a request for a virtual data center, and a virtual data center is defined as a set of virtual machines with virtual link requirements and physical host resource requirements.
出于提高资源利用率、减少电力消耗、减少建设和运营成本等方面的考虑，云服务提供商希望使用尽可能少的物理主机提供能够满足所有用户需求的服务，因此会将用户的虚拟机放置到尽量少的物理主机上。在这样的数据中心中，位于同一物理主机上的多个虚拟机共享这个物理主机的CPU、内存、硬盘等物理主机资源，也共享这个物理主机的网卡、接入带宽等网络资源。另一方面，现代云计算数据中心规模巨大，出于减少建设成本的考虑，组建数据中心的设备一般采用通用商用硬件，因而物理主机和网络设备故障并不少见，加之数据中心规模庞大，设备故障宕机事件十分平常。为了减少设备故障对用户服务带来的影响，隶属同一用户的虚拟机往往部署在不同的物理主机上。In consideration of improving resource utilization, reducing power consumption, and reducing construction and operating costs, cloud service providers hope to use as few physical hosts as possible to provide services that can meet the needs of all users, so they will place users' virtual machines in to as few physical hosts as possible. In such a data center, multiple virtual machines on the same physical host share the physical host resources such as the CPU, memory, and hard disk of the physical host, as well as network resources such as the network card and access bandwidth of the physical host. On the other hand, the scale of modern cloud computing data centers is huge. In order to reduce construction costs, the equipment used to build data centers generally uses general-purpose commercial hardware. Therefore, failures of physical hosts and network equipment are not uncommon. Downtime events are common. In order to reduce the impact of device failures on user services, virtual machines belonging to the same user are often deployed on different physical hosts.
云服务提供商需要将所有用户的虚拟数据中心请求放置到实际的物理主机上，并分配对应的计算、存储、网络等资源实现服务的提供。对于任意一个虚拟数据中心请求而言，云服务提供商除了需要为所有的虚拟机请求指定对应的物理主机、分配资源外，还需要为所有拥有通信需求的虚拟机之间提供数据转发链路、分配网络带宽。由于隶属同一个用户的多个虚拟机不能简单的聚集在少量的物理服务器上，因此同一个物理主机上一般会同时托管多个用户的虚拟机。同一用户的虚拟机位于不同的物理主机上，它们之间的数据传输需要通过连接物理主机的数据中心网络完成。单纯只考虑物理主机资源限制的虚拟机放置是经典的装箱问题，而网络资源分配的引入使得虚拟机的放置问题变得十分困难。因为在进行虚拟数据中心放置时，除了需要考虑物理主机的CPU、内存、硬盘、接入带宽等资源容量限制外，还需要考虑如何为所有的虚拟链路需求分配实际的转发路径。由于任何一条路径所分配到的资源量由组成它的各段链路中的最少值决定，因此虚拟链路需求的满足需要同时考虑每一段链路上的资源分配，而每段链路上的资源分配除了需要考虑共享当前这条链路所有用户的资源需求外，还需要考虑这些用户在其他链路上的资源竞争与分配，这使得通信需求的满足十分困难。另一方面，由于网络设备的转发能力并不是无限的(事实上交换机的转发表、缓存空间等资源十分有限)，因而在进行虚拟数据中心虚拟链路的映射时，除了需要考虑链路的带宽容量限制外，还需要考虑网络设备转发能力的限制。Cloud service providers need to place all users' virtual data center requests on actual physical hosts, and allocate corresponding computing, storage, network and other resources to provide services. For any virtual data center request, in addition to specifying the corresponding physical host and allocating resources for all virtual machine requests, the cloud service provider also needs to provide data forwarding links, Allocate network bandwidth. Since multiple virtual machines belonging to the same user cannot simply be aggregated on a small number of physical servers, the same physical host generally hosts virtual machines of multiple users at the same time. The virtual machines of the same user are located on different physical hosts, and the data transmission between them needs to be completed through the data center network connecting the physical hosts. The placement of virtual machines that only considers the resource constraints of physical hosts is a classic box packing problem, but the introduction of network resource allocation makes the placement of virtual machines very difficult. Because when placing a virtual data center, in addition to considering resource capacity limitations such as the CPU, memory, hard disk, and access bandwidth of the physical host, it is also necessary to consider how to allocate actual forwarding paths for all virtual link requirements. Since the amount of resources allocated to any path is determined by the minimum value among the links that make up it, the satisfaction of the virtual link needs to consider the resource allocation on each link at the same time, and the resource allocation on each link Resource allocation not only needs to consider the resource requirements of all users sharing the current link, but also needs to consider the resource competition and allocation of these users on other links, which makes it very difficult to meet the communication needs. On the other hand, since the forwarding capabilities of network devices are not unlimited (in fact, resources such as forwarding tables and cache space of switches are very limited), when mapping virtual links in virtual data centers, in addition to considering link bandwidth In addition to the capacity limitation, the limitation of the forwarding capability of the network equipment also needs to be considered.
由此看来，虚拟数据中心的需求放置和资源分配问题集结了服务器资源、链路带宽资源、设备转发能力等因素的共同限制，该问题已经被证明是一个NP难的问题。虚拟数据中心内部各虚拟机的通信需求在放置过后被映射为数据中心网络内部流量，当前的数据中心网络一般采用多级结构，如何减少跨区域的内部流量十分的重要。一方面，减少对核心网络设备的使用可以让数据中心拥有更大的可能接收更多后来的请求，而减少跨区域的内部流量也为以节能为目标的流量工程等最佳路由创造了优化空间。如何设计一个虚拟数据中心映射算法，实现快速地虚拟机请求放置和虚拟链路映射，同时获得高资源利用率、低跨域网络流量十分重要。From this point of view, the demand placement and resource allocation problems of virtual data centers are combined with the common constraints of server resources, link bandwidth resources, and equipment forwarding capabilities. This problem has been proven to be an NP-hard problem. The communication requirements of each virtual machine in the virtual data center are mapped to the internal traffic of the data center network after placement. The current data center network generally adopts a multi-level structure. How to reduce the internal traffic across regions is very important. On the one hand, reducing the use of core network equipment can make the data center more likely to receive more subsequent requests, and reducing cross-regional internal traffic also creates optimization space for optimal routing such as traffic engineering with the goal of energy saving . How to design a virtual data center mapping algorithm to achieve fast virtual machine request placement and virtual link mapping, while obtaining high resource utilization and low cross-domain network traffic is very important.
现有的各类虚拟数据中心映射算法都没有考虑网络设备转发能力限制的问题，因此并不能很好的处理转发能力受限情形下的虚拟数据中心映射问题。All kinds of existing virtual data center mapping algorithms do not consider the problem of network equipment forwarding capacity limitation, so they cannot deal with the virtual data center mapping problem under the condition of limited forwarding capacity.
发明内容Contents of the invention
本发明的目的在于克服现有技术的不足，提供一种业务需求感知的虚拟数据中心映射方法，综合考虑了物理主机资源容量、物理链路资源、交换机转发能力限制，实现低成本的虚拟数据中心映射。The purpose of the present invention is to overcome the deficiencies of the prior art, provide a business demand-aware virtual data center mapping method, comprehensively consider the limitations of physical host resource capacity, physical link resources, and switch forwarding capabilities, and realize a low-cost virtual data center map.
为实现上述发明目的，本发明业务需求感知的虚拟数据中心映射方法，包括以下步骤：In order to achieve the purpose of the above invention, the business demand-aware virtual data center mapping method of the present invention includes the following steps:
S1：用户向数据中心发起虚拟数据中心请求，请求包括各虚拟机的物理主机资源需求、各虚拟机之间的虚拟链路、以及各虚拟链路的带宽需求，其中各虚拟机具有的虚拟链路数量即为该虚拟机所需建立的转发表项数量；数据中心网络采用胖树结构，物理主机与数据中心的边缘交换机连接，数据中心中各物理链路配置的初始带宽资源相同，各个交换机配置的转发表初始容量相同；S1: The user initiates a virtual data center request to the data center. The request includes the physical host resource requirements of each virtual machine, the virtual links between each virtual machine, and the bandwidth requirements of each virtual link. The number of paths is the number of forwarding entries required for the virtual machine; the data center network adopts a fat tree structure, and the physical host is connected to the edge switch of the data center. The initial bandwidth resources configured for each physical link in the data center are the same, and each switch The initial capacity of the configured forwarding table is the same;
S2：根据请求中各虚拟机的物理主机资源需求中的最大值，以及各物理主机的剩余资源容量，筛选出剩余资源容量大于等于虚拟机资源需求最大值的可用物理主机，统计出数据中心中每个机架的可用物理主机数Ri，其中i表示机架序号，并将每个交换机组中各机架的可用物理主机数相加得到每个交换机组的可用物理主机数Pj，其中j表示交换机组序号；S2: According to the maximum value of the physical host resource requirements of each virtual machine in the request, and the remaining resource capacity of each physical host, screen out the available physical hosts whose remaining resource capacity is greater than or equal to the maximum value of the virtual machine resource requirements, and calculate the data center The number of available physical hosts Ri of each rack, where i represents the rack number, and the number of available physical hosts of each rack in each switch group is added to obtain the number of available physical hosts Pj of each switch group, where j represents the switch group serial number;
S3：将交换机组按照可用物理主机数Pj进行降序排列，每个交换机组中的机架也按照可用物理主机数Ri进行降序排列；S3: arrange the switch group in descending order according to the number of available physical hosts Pj , and arrange the racks in each switch group in descending order according to the number of available physical hosts Ri ;
S4：选择带宽资源需求最大的虚拟机，根据步骤S3排序得到的物理主机顺序，依次尝试将该虚拟机映射到物理主机上，映射条件为：物理主机与其连接的边缘交换机间的物理链路带宽资源能满足该虚拟机带宽资源需求、以及该边缘交换机剩余转发表容量能满足该虚拟机转发表项的建立；如果存在可映射该虚拟机的物理主机，则将该虚拟机映射到找到的第一个物理主机上，更新与被映射的物理主机连接的边缘交换机的转发表容量以及对应物理链路的剩余带宽资源；如果不能找到，返回映射失败消息，虚拟数据中心映射结束；S4: Select the virtual machine with the largest demand for bandwidth resources, and try to map the virtual machine to the physical host in turn according to the order of the physical hosts sorted in step S3. The mapping condition is: the physical link bandwidth between the physical host and the edge switch connected to it The resources can meet the bandwidth resource requirements of the virtual machine, and the remaining forwarding table capacity of the edge switch can meet the establishment of the virtual machine forwarding table entry; if there is a physical host that can map the virtual machine, map the virtual machine to the found first On a physical host, update the forwarding table capacity of the edge switch connected to the mapped physical host and the remaining bandwidth resources of the corresponding physical link; if it cannot be found, return a mapping failure message, and the virtual data center mapping ends;
S5：统计所有已映射和未映射的虚拟机之间的虚拟链路带宽资源需求，选择其中带宽资源需求最大的虚拟链路，将该虚拟链路对应的已映射虚拟机标记为u，其映射的物理主机标记为a，未映射虚拟机标记为v，根据步骤S3排序得到的物理主机顺序，依次尝试将虚拟机v映射到与物理主机a位于同一个机架的其它可用物理主机上，如果该机架存在可映射虚拟机v的物理主机，进入步骤S8，否则进入步骤S6；S5: Count the virtual link bandwidth resource requirements between all mapped and unmapped virtual machines, select the virtual link with the largest bandwidth resource demand, mark the mapped virtual machine corresponding to the virtual link as u, and its mapping The physical host marked as a, the unmapped virtual machine marked as v, according to the order of the physical hosts sorted in step S3, try to map the virtual machine v to other available physical hosts in the same rack as the physical host a, if The rack has a physical host that can map the virtual machine v, go to step S8, otherwise go to step S6;
S6：根据步骤S3排序得到的物理主机顺序，依次尝试将虚拟机v映射到与物理主机a位于同一个交换机组的其它机架的可用物理主机上，如果其他机架存在可映射虚拟机v的物理主机，进入步骤S8；否则进入步骤S7；S6: According to the order of the physical hosts sorted in step S3, try to map the virtual machine v to the available physical hosts in other racks in the same switch group as the physical host a, if there are other racks that can map the virtual machine v Physical host, go to step S8; otherwise go to step S7;
S7：根据步骤S3排序得到的物理主机顺序，依次尝试将虚拟机v映射到物理主机a的交换机组外的其它交换机组的可用物理主机上，如果其它交换机组存在可映射虚拟机v的物理主机，进入步骤S8；否则返回映射失败消息，虚拟数据中心映射结束；S7: According to the order of the physical hosts sorted in step S3, try to map the virtual machine v to the available physical hosts of other switch groups other than the switch group of physical host a in turn, if there are physical hosts that can map the virtual machine v in other switch groups , enter step S8; otherwise, a mapping failure message is returned, and the virtual data center mapping ends;
S8：将虚拟机v映射到找到的第一个物理主机上，更新与被映射的物理主机连接的边缘交换机的转发表容量以及对应物理链路的剩余带宽资源；判断是否所有虚拟机都已实现映射，如果不是，返回步骤S5，如果是，进入步骤S9；S8: Map the virtual machine v to the first physical host found, update the forwarding table capacity of the edge switch connected to the mapped physical host and the remaining bandwidth resources of the corresponding physical link; determine whether all virtual machines have been implemented Mapping, if not, return to step S5, if yes, enter step S9;
步骤S9：根据所有虚拟机的映射结果以及虚拟机之间的带宽需求，进行虚拟链路映射，返回虚拟数据中心映射成功消息，虚拟数据中心映射结束。Step S9: Perform virtual link mapping according to the mapping results of all virtual machines and the bandwidth requirements between the virtual machines, return a virtual data center mapping success message, and complete the virtual data center mapping.
其中，转发表为路由表。Wherein, the forwarding table is a routing table.
其中，转发表为流表。Among them, the forwarding table is a flow table.
进一步地，步骤S3还包括将各机架中可用物理主机根据剩余资源容量进行降序排列。Further, step S3 also includes arranging the available physical hosts in each rack in descending order according to the remaining resource capacity.
其中，步骤S9中虚拟链路映射时每一跳均选择剩余带宽最大的链路。Wherein, in the virtual link mapping in step S9, the link with the largest remaining bandwidth is selected for each hop.
本发明是具有业务需求感知的虚拟数据中心映射方法，数据中心网络采用胖树结构，数据中心中各物理链路配置的带宽资源相同、各个交换机配置的转发表容量相同。当有用户发起虚拟数据中心映射请求时，首先从所有物理主机中筛选出剩余资源容量大于等于虚拟机中最大物理主机资源需求量的作为可用物理主机，将交换机组和机架按照可用物理主机数进行降序排列，选择链路资源需求最大的虚拟机，在所有物理主机中依次尝试进行映射。映射成功后，再选择已映射虚拟机与未映射虚拟机之间链路资源需求最大的虚拟链路，选择其对应的未映射虚拟机，依次在与对应已映射虚拟机所在物理主机相同机架、相同交换机组、整个数据中心网络中的物理主机中尝试进行映射，所有虚拟机映射完毕后，再进行虚拟链路映射，从而完成虚拟数据中心映射。The present invention is a virtual data center mapping method with service demand perception. The data center network adopts a fat tree structure, and the bandwidth resources configured by each physical link in the data center are the same, and the capacity of the forwarding table configured by each switch is the same. When a user initiates a virtual data center mapping request, first screen out all physical hosts with a remaining resource capacity greater than or equal to the maximum physical host resource demand in the virtual machine as an available physical host, and assign switch groups and racks according to the number of available physical hosts Sort in descending order, select the virtual machine with the greatest demand for link resources, and try to perform mapping among all physical hosts in turn. After the mapping is successful, select the virtual link with the largest link resource requirements between the mapped virtual machine and the unmapped virtual machine, select its corresponding unmapped virtual machine, and then select the virtual link in the same rack as the physical host where the corresponding mapped virtual machine is located. , the same switch group, and physical hosts in the entire data center network try to map. After all virtual machines are mapped, virtual link mapping is performed to complete the virtual data center mapping.
本发明具有以下有益效果：The present invention has the following beneficial effects:
(1)、所有虚拟机映射的物理主机各不相同，可以减少物理主机故障带来的影响；(1) The physical hosts mapped by all virtual machines are different, which can reduce the impact of physical host failures;
(2)、将交换机组和机架按照可用物理主机数进行降序排列，可以使同属一个虚拟数据中心的虚拟机尽可能多地映射在同一个机架或者同一个交换机组中，减少跨区域的数据流量；(2) Arrange the switch groups and racks in descending order according to the number of available physical hosts, so that as many virtual machines belonging to the same virtual data center can be mapped to the same rack or the same switch group as much as possible, reducing cross-regional conflicts Data Flow;
(3)、综合考虑了物理主机资源容量、物理链路资源、交换机转发能力限制，使虚拟机的映射更为合理；(3) The resource capacity of the physical host, the physical link resource, and the forwarding capability of the switch are comprehensively considered to make the mapping of the virtual machine more reasonable;
(4)、采用本发明进行虚拟数据中心映射的网络成本较低。(4) The network cost of virtual data center mapping by using the present invention is relatively low.
附图说明Description of drawings
图1是本发明业务需求感知的虚拟数据中心映射方法的一种具体实施方式流程图；Fig. 1 is a flow chart of a specific embodiment of the business demand-aware virtual data center mapping method of the present invention;
图2是Fat-Tree网络结构的示意图；Fig. 2 is the schematic diagram of Fat-Tree network structure;
图3是虚拟数据中心的示例图。Fig. 3 is an example diagram of a virtual data center.
具体实施方式detailed description
下面结合附图对本发明的具体实施方式进行描述，以便本领域的技术人员更好地理解本发明。需要特别提醒注意的是，在以下的描述中，当已知功能和设计的详细描述也许会淡化本发明的主要内容时，这些描述在这里将被忽略。Specific embodiments of the present invention will be described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when detailed descriptions of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.
实施例Example
图1是本发明业务需求感知的虚拟数据中心映射方法的一种具体实施方式流程图。如图1所示，本发明业务需求感知的虚拟数据中心映射方法，具体包括以下步骤：FIG. 1 is a flow chart of a specific embodiment of the service demand-aware virtual data center mapping method of the present invention. As shown in Figure 1, the method for mapping a virtual data center based on business demand perception in the present invention specifically includes the following steps:
S101：用户向数据中心发起虚拟数据中心请求，请求包括各虚拟机的物理主机资源需求、各虚拟机之间的虚拟链路、各虚拟链路的带宽需求，其中各虚拟机具有的虚拟链路数量即为该虚拟机所需建立的转发表项数量；数据中心网络采用胖树结构，物理主机与数据中心的边缘交换机连接，数据中心中各物理链路配置的带宽初始资源相同，各个交换机配置的转发表初始容量相同。S101: The user initiates a virtual data center request to the data center, and the request includes the physical host resource requirements of each virtual machine, the virtual links between each virtual machine, and the bandwidth requirements of each virtual link, among which the virtual links of each virtual machine The number is the number of forwarding entries that need to be established for the virtual machine; the data center network adopts a fat tree structure, and the physical host is connected to the edge switch of the data center. The initial capacity of the forwarding table is the same.
图2是Fat-Tree网络结构的示意图。如图2所示，Fat-Tree网络是一种类clos网络结构，其规模大小可以用组建Fat-Tree网络所使用交换机的端口数量k来描述。图2中为k＝4的Fat-Tree网络。Fat-Tree网络中一般可被看作由边缘(edge)、汇聚(aggregation)、核心(core)三级结构组成。一个参数为k的Fat-Tree包含k个交换机组(pod)202，每个交换机组由k个交换机构成，而每个交换机具有k个端口。每个交换机内的k个交换机被分为两组，分别是边缘交换机204和汇聚交换机203，k/2个边缘交换机204中，每个交换机的k/2个端口与物理主机205相连，剩下k/2个端口分别与本交换机组内的k/2个汇聚交换机203相连。这k/2个汇聚交换机203剩下的k2/4个端口，分别与k2/4个核心交换机201相连。每个核心交换机201的k个端口分别和k个交换机组202内的对应汇聚交换机203相连。在数据中心中，每个边缘交换机204负责将物理主机205接入到网络中，每个边缘交换机204和对应的物理主机205位于同一个机架(rack)206上。一个参数为k的Fat-Tree数据中心的构建需要5k2/4个k端口的交换机，能够支持k3/4个物理主机的互联。Fig. 2 is a schematic diagram of the Fat-Tree network structure. As shown in Figure 2, the Fat-Tree network is a clos-like network structure, and its size can be described by the number of ports k of switches used to form the Fat-Tree network. Figure 2 is a Fat-Tree network with k=4. Generally, the Fat-Tree network can be regarded as composed of three levels of structure: edge, aggregation, and core. A Fat-Tree with a parameter of k includes k switch groups (pod) 202, each switch group is composed of k switches, and each switch has k ports. The k switches in each switch are divided into two groups, which are edge switches 204 and aggregation switches 203 respectively. Among the k/2 edge switches 204, k/2 ports of each switch are connected to the physical host 205, and the remaining The k/2 ports are respectively connected to the k/2 convergence switches 203 in the switch group. The remaining k2 /4 ports of the k/2 aggregation switches 203 are respectively connected to the k2 /4 core switches 201 . The k ports of each core switch 201 are respectively connected to the corresponding convergence switches 203 in the k switch groups 202 . In the data center, each edge switch 204 is responsible for connecting the physical host 205 to the network, and each edge switch 204 and the corresponding physical host 205 are located on the same rack (rack) 206 . The construction of a Fat-Tree data center with a parameter of k requires 5k2 /4 switches with k ports, which can support the interconnection of k3 /4 physical hosts.
交换机的转发能力通常包括出入带宽资源以及转发表容量，本发明中数据中心中各物理链路配置的带宽初始资源相同，各个交换机配置的转发表初始容量相同。由于边缘交换机直接与物理主机相连，那么任何物理主机上映射虚拟机时，与该物理连接的边缘交换机的转发表容量和对应的物理链路的带宽资源都会消耗，而汇聚交换机和核心交换机及其物理链路不一定会被消耗，而且由于Fat-Tree的网络拓扑结构特点，只要边缘交换机的转发表容量和其与物理主机之间的物理链路带宽资源能够满足虚拟机的需求，那么就一定可以在数据中心网络中找到满足需求的物理链路和中间交换机以进行虚拟链路映射。The forwarding capability of the switch usually includes inbound and outbound bandwidth resources and forwarding table capacity. In the present invention, the bandwidth initial resources configured for each physical link in the data center are the same, and the forwarding table configured for each switch has the same initial capacity. Since the edge switch is directly connected to the physical host, when a virtual machine is mapped on any physical host, the forwarding table capacity of the edge switch connected to the physical host and the bandwidth resource of the corresponding physical link will be consumed, while the aggregation switch and the core switch and their Physical links may not be consumed, and due to the characteristics of the Fat-Tree network topology, as long as the forwarding table capacity of the edge switch and the bandwidth resources of the physical link between it and the physical host can meet the requirements of the virtual machine, then it must be Physical links and intermediate switches that meet the requirements can be found in the data center network for virtual link mapping.
转发表根据实际组网设备的不同而不同，例如如果是采用传统的路由器组网，转发表即为路由表，如果是采用OpenFlow设备组网，那么转发表即为流表。The forwarding table varies according to the actual networking equipment. For example, if a traditional router is used for networking, the forwarding table is a routing table; if an OpenFlow device is used for networking, the forwarding table is a flow table.
S102：筛选可用物理主机：S102: Screening available physical hosts:
根据请求中各虚拟机的物理主机资源需求中的最大值，以及各物理主机的剩余资源容量，筛选出剩余资源容量大于等于虚拟机资源需求最大值的可用物理主机，统计出数据中心中每个机架的可用物理主机数Ri，其中i表示机架序号，并将每个交换机组中各机架的可用物理主机数相加得到每个交换机组的可用物理主机数Pj，其中j表示交换机组序号。According to the maximum value of the physical host resource requirements of each virtual machine in the request, and the remaining resource capacity of each physical host, the available physical hosts whose remaining resource capacity is greater than or equal to the maximum value of the virtual machine resource requirements are screened out, and each The number of available physical hosts Ri of the rack, where i represents the serial number of the rack, and the number of available physical hosts of each rack in each switch group is added to obtain the number of available physical hosts Pj of each switch group, where j represents Switch group serial number.
物理主机的资源通常会包括CPU、内存、硬盘、接入带宽等，那么可用物理主机应该每一项资源都能够满足虚拟机的需求。The resources of the physical host usually include CPU, memory, hard disk, access bandwidth, etc., so each resource of the available physical host should be able to meet the needs of the virtual machine.
S103：将交换机机组和机架按可用物理主机数量排序：S103: Sort the switch groups and racks according to the number of available physical hosts:
将交换机组按照可用物理主机数Pj进行降序排列，每个交换机组中的机架也按照可用物理主机数Ri进行降序排列。进行排序的主要目的是在进行虚拟机映射时，能尽量映射在同一个机架或者同一个交换机组内，从而减少数据中心中跨区域的内部流量。The switch groups are arranged in descending order according to the number of available physical hosts Pj , and the racks in each switch group are also arranged in descending order according to the number of available physical hosts Ri . The main purpose of sorting is to try to map the virtual machines in the same rack or the same switch group when performing virtual machine mapping, thereby reducing cross-regional internal traffic in the data center.
在进行排序时，还可以对机架中的物理主机根据剩余资源容量进行降序排列，虚拟机会优先映射在剩余资源容量较大的物理主机上，这样可以使物理主机的负载更加均衡。When sorting, the physical hosts in the rack can also be arranged in descending order according to the remaining resource capacity, and the virtual machines are preferentially mapped to the physical hosts with larger remaining resource capacity, which can make the load of the physical hosts more balanced.
S104：选择带宽资源需求最大的虚拟机进行映射：S104: Select the virtual machine with the largest demand for bandwidth resources for mapping:
选择带宽资源需求最大的虚拟机，根据步骤S103得到的物理主机顺序，依次尝试将该虚拟机映射到物理主机上，映射条件为：物理主机与其连接的边缘交换机间的物理链路的剩余带宽资源能满足该虚拟机带宽资源需求、以及该边缘交换机剩余转发表容量能满足该虚拟机转发表项的建立；如果存在可映射该虚拟机的物理主机，则将该虚拟机映射到找到的第一个物理主机上。Select the virtual machine with the largest demand for bandwidth resources, and try to map the virtual machine to the physical host in turn according to the order of the physical hosts obtained in step S103. The mapping condition is: the remaining bandwidth resources of the physical link between the physical host and the edge switch connected to it Can meet the bandwidth resource requirements of the virtual machine, and the remaining forwarding table capacity of the edge switch can meet the establishment of the virtual machine forwarding table; if there is a physical host that can map the virtual machine, map the virtual machine to the first found on a physical host.
图3是虚拟数据中心的示例图。如图3所示，该虚拟数据中心包含5个虚拟机(VM，VirtualMachine)，各虚拟机之间连线上的数字代表该虚拟链路的带宽资源需求，例如虚拟机VM1和虚拟机VM2之间虚拟链路的带宽资源需求为10。带宽资源需求最大的虚拟机是指其所有与其他虚拟机连接的虚拟链路的带宽资源需求总和最大的虚拟机，可见图3中，虚拟机VM2的带宽资源需求最大，第一个映射的虚拟机是VM2。Fig. 3 is an example diagram of a virtual data center. As shown in Figure 3, the virtual data center includes five virtual machines (VM, Virtual Machine), and the numbers on the lines between the virtual machines represent the bandwidth resource requirements of the virtual links, such as the virtual machine VM1 and the virtual machine VM2. The bandwidth resource requirement of virtual links between virtual links is 10. The virtual machine with the largest bandwidth resource demand refers to the virtual machine with the largest bandwidth resource demand of all virtual links connected to other virtual machines. As shown in Figure 3, the virtual machine VM2 has the largest bandwidth resource demand, and the first mapped virtual machine The machine is VM2.
S105：判断步骤S104是否映射成功，如果成功，更新与被映射的物理主机连接的边缘交换机的转发表容量以及对应物理链路的剩余带宽资源，进入步骤S106，如果不成功，返回映射失败消息，虚拟数据中心映射结束。S105: Determine whether the mapping in step S104 is successful, if successful, update the forwarding table capacity of the edge switch connected to the mapped physical host and the remaining bandwidth resources of the corresponding physical link, enter step S106, if not successful, return a mapping failure message, The virtual datacenter mapping is complete.
如果一个虚拟机无法映射，那么整个虚拟数据中心都无法成功映射，因此结束虚拟数据中心映射。If one virtual machine cannot be mapped, then the entire virtual data center cannot be successfully mapped, thus ending the virtual data center mapping.
S106：根据虚拟链路带宽资源需求选择下一个未映射虚拟机进行映射：S106: Select the next unmapped virtual machine for mapping according to the bandwidth resource requirements of the virtual link:
统计所有已映射和未映射的虚拟机之间的虚拟链路带宽资源需求，选择其中带宽资源需求最大的虚拟链路，将该虚拟链路对应的已映射虚拟机标记为u，其映射的物理主机标记为a，未映射虚拟机标记为v。Count the virtual link bandwidth resource requirements between all mapped and unmapped virtual machines, select the virtual link with the largest bandwidth resource demand, mark the mapped virtual machine corresponding to the virtual link as u, and the mapped physical The host is marked a and the unmapped virtual machine is marked v.
如图3所示，当成功映射了虚拟机VM2后，与其虚拟链路带宽资源需求最大的虚拟机为VM3，第二个应该映射的虚拟机为VM3。依次在虚拟机VM2映射的物理主机的相同机架、相同交换机组、整个数据中心查找可映射虚拟机VM3的物理主机。As shown in FIG. 3 , after the virtual machine VM2 is successfully mapped, the virtual machine with the greatest demand for virtual link bandwidth resources is VM3, and the second virtual machine that should be mapped is VM3. Find the physical host that can map the virtual machine VM3 in the same rack, the same switch group, and the entire data center in turn.
根据步骤S103得到的物理主机顺序，依次尝试将虚拟机v映射到与物理主机a位于同一个机架的其它可用物理主机上，如果该机架存在可映射虚拟机v的物理主机，则将虚拟机v映射在找到的第一个物理主机上，否则再依次尝试将虚拟机v映射到与物理主机a位于同一个交换机组的其它机架的可用物理主机上，如果其他机架存在可映射虚拟机v的物理主机，则将虚拟机v映射在找到的第一个物理主机上，否则再依次尝试将虚拟机v映射到物理主机a的交换机组外的其它交换机组的物理主机上。According to the sequence of physical hosts obtained in step S103, try to map virtual machine v to other available physical hosts in the same rack as physical host a. If there is a physical host in this rack that can map virtual machine v, virtual Machine v is mapped to the first physical host found, otherwise try to map virtual machine v to an available physical host in another rack in the same switch group as physical host a, if there are mappable virtual machines in other racks If there is no physical host of machine v, map virtual machine v to the first physical host found, otherwise try to map virtual machine v to physical hosts of other switch groups other than the switch group of physical host a in turn.
S107：判断步骤S106是否映射成功，如果成功，更新与被映射的物理主机连接的边缘交换机的转发表容量以及对应物理链路的剩余带宽资源，进入步骤S108，如果不成功，返回映射失败消息，虚拟数据中心映射结束。S107: Determine whether the mapping in step S106 is successful, if successful, update the forwarding table capacity of the edge switch connected to the mapped physical host and the remaining bandwidth resources of the corresponding physical link, enter step S108, if not successful, return a mapping failure message, The virtual datacenter mapping is complete.
S108：判断是否所有虚拟机都已成功映射，如果不是，返回步骤S106，如果是，进入步骤S109。S108: Determine whether all virtual machines have been successfully mapped, if not, return to step S106, and if yes, enter step S109.
S109：根据所有虚拟机的映射结果以及虚拟机之间的带宽需求，进行虚拟链路映射，返回虚拟数据中心映射成功消息，虚拟数据中心映射结束。进行虚拟链路映射时，用现在许多技术都能实现，可以根据实际需要，选择满足负载均衡或者网络节能需求的最佳路由实现，例如需负载均衡时，可以在虚拟链路映射时每一跳均选择剩余带宽最大的链路。S109: Perform virtual link mapping according to the mapping results of all virtual machines and the bandwidth requirements between the virtual machines, and return a virtual data center mapping success message, and the virtual data center mapping ends. When performing virtual link mapping, many current technologies can be used to achieve it. According to actual needs, the best route to meet the load balancing or network energy saving requirements can be selected. For example, when load balancing is required, each hop can be used during virtual link mapping Both select the link with the largest remaining bandwidth.
本发明业务需求感知的虚拟数据中心映射方法，是综合考虑了物理主机资源容量、物理链路资源容量、交换机转发能力限制的情况下，占用最少的网络资源将用户的虚拟数据中心请求映射到物理数据中心中。以GV表示需要映射的虚拟数据中心，对于一个给定的映射方案π，πN和πL分别表示虚拟机和虚拟链路的映射方案，可以用相对映射成本R(GV,π)估计该映射方案的效果。计算公式为：The virtual data center mapping method of the service demand perception of the present invention is to map the user's virtual data center request to the physical data center by occupying the least network resources under the condition that the resource capacity of the physical host, the resource capacity of the physical link, and the forwarding capability of the switch are considered comprehensively. in the data center. Let GV represent the virtual data center that needs to be mapped. For a given mapping scheme π, πN and πL represent the mapping schemes of virtual machines and virtual links respectively, which can be estimated by the relative mapping cost R(GV , π) The effect of the mapping scheme. The calculation formula is:
是一个估计值，E(GV)表示虚拟链路集合，其中πL(e)表示虚拟链路e映射到的物理路径，hops(πL(e))表示该物理路径经过的跳数，γ表示带宽资源与交换机转发能力与资源成本之间的权重比值。对于任意一个虚拟数据中心请求，相对映射成本越低的映射方案拥有更好的效果。is an estimated value, E(GV ) represents the set of virtual links, where πL (e) represents the physical path to which the virtual link e is mapped, and hops(πL (e)) represents the number of hops passed by the physical path, γ represents the weight ratio between bandwidth resources, switch forwarding capability and resource cost. For any virtual data center request, a mapping scheme with a lower relative mapping cost has a better effect.
对于虚拟机数为n的虚拟数据中心和物理主机数位m的物理数据中心的映射计算，本发明算法的时间复杂度不超过O(n2,m)。并且本发明的方法与没有考虑虚拟机之间的业务需求和数据中心中不同物理主机之间的通信成本的普通算法相比，能够获得更高的资源利用率，在对比仿真实验中，本发明可以在将虚拟数据中心映射的成功率提升0.027的同时能减少约41％的相对网络成本。For the mapping calculation between a virtual data center with n number of virtual machines and a physical data center with m number of physical hosts, the time complexity of the algorithm of the present invention does not exceed O(n2 ,m). And the method of the present invention can obtain higher resource utilization rate compared with the common algorithm that does not consider the business demand between the virtual machine and the communication cost between different physical hosts in the data center, in the comparative simulation experiment, the present invention It can reduce the relative network cost by about 41% while increasing the success rate of virtual data center mapping by 0.027.
尽管上面对本发明说明性的具体实施方式进行了描述，以便于本技术领域的技术人员理解本发明，但应该清楚，本发明不限于具体实施方式的范围，对本技术领域的普通技术人员来讲，只要各种变化在所附的权利要求限定和确定的本发明的精神和范围内，这些变化是显而易见的，一切利用本发明构思的发明创造均在保护之列。Although the illustrative specific embodiments of the present invention have been described above, so that those skilled in the art can understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, As long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310488026.0ACN103560967B (en) | 2013-10-17 | 2013-10-17 | The virtual data center mapping method of a kind of business demand perception |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310488026.0ACN103560967B (en) | 2013-10-17 | 2013-10-17 | The virtual data center mapping method of a kind of business demand perception |
| Publication Number | Publication Date |
|---|---|
| CN103560967A CN103560967A (en) | 2014-02-05 |
| CN103560967Btrue CN103560967B (en) | 2016-06-01 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310488026.0AExpired - Fee RelatedCN103560967B (en) | 2013-10-17 | 2013-10-17 | The virtual data center mapping method of a kind of business demand perception |
| Country | Link |
|---|---|
| CN (1) | CN103560967B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106933654A (en)* | 2017-03-17 | 2017-07-07 | 中山大学 | A kind of virtual machine based on caching starts method |
| CN109314804A (en)* | 2016-07-22 | 2019-02-05 | 英特尔公司 | Technologies for Optical Communication in Rack Clusters |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9876737B2 (en) | 2014-07-11 | 2018-01-23 | Oracle International Corporation | System and method for supporting efficient load-balancing in a high performance computing (HPC) environment |
| CN105718299A (en)* | 2014-12-04 | 2016-06-29 | 中国移动通信集团广东有限公司 | Virtual machine configuration method, device and system |
| CN104468803B (en) | 2014-12-12 | 2018-02-09 | 华为技术有限公司 | A kind of virtual data center method for mapping resource and equipment |
| CN104901999B (en)* | 2015-03-31 | 2018-04-06 | 电子科技大学 | A kind of virtual data center cross-domain mapping method that reliability perceives |
| CN104796673B (en)* | 2015-05-19 | 2017-08-25 | 中国地质大学（武汉） | A kind of cloud video monitoring system task cut-in method towards energy optimization |
| CN105120514B (en)* | 2015-07-03 | 2018-05-29 | 深圳市纳斯达工贸有限公司 | A kind of wireless network Poewr control method based on cloud computing |
| CN105025564B (en)* | 2015-07-03 | 2018-05-22 | 广州市奥威亚电子科技有限公司 | A kind of wireless network power control system based on cloud computing |
| CN106851440A (en)* | 2016-12-09 | 2017-06-13 | 国家计算机网络与信息安全管理中心 | A kind of resource tasks dispatching method towards light matrix network |
| CN106681795B (en)* | 2016-12-09 | 2020-04-28 | 北京航空航天大学 | A Virtual Network Mapping Method for Node Local Topology and Available Resource Capacity |
| CN106789421B (en)* | 2016-12-17 | 2020-06-05 | 深圳中广核工程设计有限公司 | Method and apparatus for collaborative design |
| CN107749805A (en)* | 2017-10-30 | 2018-03-02 | 山东浪潮云服务信息科技有限公司 | A kind of virtual machine deployment method and device |
| CN111722907B (en)* | 2020-05-20 | 2024-01-19 | 中天通信技术有限公司 | DVFS-based data center mapping method, device and storage medium |
| CN113037819B (en)* | 2021-02-26 | 2022-09-23 | 杭州雾联科技有限公司 | Edge computing resource sharing method, device and equipment |
| CN113630328B (en)* | 2021-07-09 | 2022-09-16 | 西安电子科技大学 | A data center virtual network mapping method and system |
| CN114244715B (en)* | 2021-12-17 | 2023-05-23 | 浪潮商用机器有限公司 | vNIC main and standby port selection method, device and medium thereof |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101085535B1 (en)* | 2010-01-28 | 2011-11-23 | 한국과학기술원 | Resource Embedding System for Virtual Network Request in Wireless Environment and Its Method |
| US8423646B2 (en)* | 2010-07-09 | 2013-04-16 | International Business Machines Corporation | Network-aware virtual machine migration in datacenters |
| CN102904794A (en)* | 2012-09-27 | 2013-01-30 | 北京邮电大学 | A virtual network mapping method and device |
| CN102932479B (en)* | 2012-11-16 | 2015-01-07 | 北京邮电大学 | Virtual network mapping method for realizing topology awareness based on historical data |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109314804A (en)* | 2016-07-22 | 2019-02-05 | 英特尔公司 | Technologies for Optical Communication in Rack Clusters |
| CN109314804B (en)* | 2016-07-22 | 2022-08-09 | 英特尔公司 | Techniques for optical communications in rack clusters |
| CN106933654A (en)* | 2017-03-17 | 2017-07-07 | 中山大学 | A kind of virtual machine based on caching starts method |
| Publication number | Publication date |
|---|---|
| CN103560967A (en) | 2014-02-05 |
| Publication | Publication Date | Title |
|---|---|---|
| CN103560967B (en) | The virtual data center mapping method of a kind of business demand perception | |
| US10965586B2 (en) | Resilient network communication using selective multipath packet flow spraying | |
| US11057387B2 (en) | Infrastructure aware adaptive resource allocation | |
| US10284497B2 (en) | Networking method for data center network and data center network | |
| Gandhi et al. | Duet: Cloud scale load balancing with hardware and software | |
| Guo et al. | Secondnet: a data center network virtualization architecture with bandwidth guarantees | |
| JP2024096845A (en) | SYSTEM AND METHOD FOR EFFICIENT VIRTUATION IN A LOSSLESS NETWORK - Patent application | |
| CN103023827B (en) | Data forwarding method for virtualized data centre and realization equipment of data forwarding method | |
| CN103559084B (en) | A kind of virtual machine migration method at Energy-saving Data center | |
| Dong et al. | Energy-performance tradeoffs in IaaS cloud with virtual machine scheduling | |
| US20190140937A1 (en) | Weighted multipath routing configuration in software-defined network (sdn) environments | |
| Dong et al. | Virtual machine placement for improving energy efficiency and network performance in iaas cloud | |
| US20150180715A1 (en) | Method of constructing logical network and network system | |
| Roh et al. | Joint flow and virtual machine placement in hybrid cloud data centers | |
| CN105704180A (en) | Configuration method and configuration system for data center | |
| Luo et al. | Traffic-aware VDC embedding in data center: A case study of fattree | |
| CN104583954B (en) | Method and apparatus for providing flow redundancy eliminating perceived slot placement | |
| US11575620B2 (en) | Queue-to-port allocation | |
| Song et al. | Ashman: A bandwidth fragmentation-based dynamic flow scheduling for data center networks | |
| US20250094350A1 (en) | System and method for cache pooling and efficient usage and i/o transfer in disaggregated and multi-processor architectures via processor interconnect | |
| CN118368230A (en) | Method for forwarding computing messages in a network, forwarding node and computer storage medium | |
| CN209692803U (en) | SDN switching network based on fat tree structure | |
| US12289240B2 (en) | Dynamic service rebalancing in network interface cards having processing units | |
| Ilyadis | The evolution of next-generation data center networks for high capacity computing | |
| Shuo et al. | Nxt-Max: for supporting VDC-based maximum redundant bandwidth allocation in cloud datacenter |
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C56 | Change in the name or address of the patentee | ||
| CP01 | Change in the name or title of a patent holder | Address after:611731 Chengdu province high tech Zone (West) West source Avenue, No. 2006 Patentee after:Electronic Science and Technology Univ. Patentee after:Guangdong Institute of electronic and information engineering, University of Electronic Science and technology of China Address before:611731 Chengdu province high tech Zone (West) West source Avenue, No. 2006 Patentee before:Electronic Science and Technology Univ. Patentee before:Institute of Electronic and Information Engineering in Dongguan, UESTC | |
| CF01 | Termination of patent right due to non-payment of annual fee | Granted publication date:20160601 Termination date:20171017 | |
| CF01 | Termination of patent right due to non-payment of annual fee |