CN102075429B - Virtual network mapping method based on principle of proximity - Google Patents

Abstract

The invention provides a virtual network mapping method based on the proximity principle. This method not only considers the remaining resources of the underlying physical network nodes in the node mapping, but also considers whether the principle of proximity is satisfied, that is, whether there is a link connection with the underlying physical network nodes that have been successfully mapped; in addition, in this method It also introduces a mechanism for queuing nodes according to their remaining resources during virtual network node mapping, and a mechanism for queuing links according to their bandwidth before virtual network link mapping. The method provided by the present invention is applicable to networks such as experimental networks and operator networks that have or will use network virtualization technology for network separation, resource management and scheduling, or to provide customized services. Features such as high utilization rate and high success rate of virtual network request mapping.

Description

A Virtual Network Mapping Method Based on Proximity Principle

technical field

Network virtualization technology is one of the important methods to promote the development of the Internet architecture. Its essence is to independently operate multiple virtual subnets on a public physical network through abstraction, distribution, and isolation mechanisms. Each virtual subnet can use independent The protocol system, and can reasonably configure the node and link resources in the entire network according to the dynamically changing needs of users, thereby enhancing the flexibility and diversity of the network, realizing the measurability and controllability of the network, and optimizing the allocation and allocation of network resources. Scheduling, improving security and service quality, reducing operation and maintenance costs, in order to fundamentally solve the existing rigidity of the Internet, mainly patch and update development status.

Network virtualization technology can be used to provide the basis for a shared physical experiment network for research on new network architectures, and it can also separate the underlying physical facility provider from the network service operator, allowing multiple operators' networks to share the same network The public underlying physical network infrastructure (links, switching nodes, etc.), each network has a share of network resources that is not affected by other networks and can be flexibly adjusted. Different network operators can use different network protocols to provide Innovative end-to-end services, so network virtualization is also expected to become a mainstream operating mode of future networks.

Background technique

The problem of virtual network mapping is an essential link in network virtualization technology. Its main function is to reasonably map the user's virtual network request (Virtual Request) to the underlying physical network facility (Substrate Network) provided by the operator. The process should not only realize the separation and non-interference between virtual networks, so as to ensure the quality of service (QoS) of each virtual network user, but also allocate the underlying physical network resources as reasonably as possible to improve resource utilization. As shown in Figure 1.

In Figure 1, two different virtual networks are mapped on the underlying physical network and provide services to corresponding users. Due to the diversity of virtual network request topology and the need to consider two sets of constraints of nodes and links at the same time, it becomes an NP-hard problem to map multiple different virtual networks to a common underlying physical network. In order to solve this problem, many foreign researchers have proposed some mapping methods to solve the suboptimal solution of mapping matching, but the existing algorithms generally have problems such as complex solution of matching equations, high computational cost, and lack of specific path selection methods.

The implementation process of virtual network mapping can be divided into two steps: node mapping and link mapping. The existing main method is to use greedy algorithm for node mapping and K shortest path algorithm for link mapping. The system uses a time window as a unit, and all virtual network requests within a time window will be sorted according to their income, starting from the largest request for mapping. If the mapping is successful, the underlying physical network status will be updated; if it fails, the request will be placed in the waiting queue; if the number of failures exceeds the preset parameter DELAY, the request will be rejected directly.

Among them, the mapping steps for each virtual network request are as follows:

First perform node mapping: for each virtual network node (V _node ) in the virtual network request, use the greedy algorithm to find the underlying physical network node (S _node ) with the largest remaining resources; if the S _node meets the CPU limit of the V _node , then the V _node is successfully mapped; if there is no S _node that meets the requirements for a certain V _node , the node mapping fails; if all V _nodes are successfully mapped, the node mapping is completed.

After the node mapping is completed, perform link mapping: for each virtual network link (V _hnk ) in the virtual network request, determine that its two ends V _node1 and V _node2 are mapped to S _node1 and S _node2 in the underlying physical network; use K The shortest path algorithm finds the 1-K shortest paths between S _node1 and S _node2 ; if one of the paths meets the bandwidth requirements of the V _hnk , the V _hnk mapping is successful; if all K paths do not meet the bandwidth requirements , the link mapping fails; if all V _links are successfully mapped, the link mapping is completed.

In the current virtual network mapping algorithm, because the design of the node mapping algorithm does not take its influence on the link mapping into consideration, a node mapping result optimized by using a greedy algorithm may cause the link mapping to be complicated or impossible to complete. For example, the nodes selected by the greedy algorithm may be far apart in the underlying physical network topology (multi-hop), then the links and nodes among them will be frequently used, thereby increasing the burden; at the same time, because a virtual link will occupy With multiple underlying physical network links, the resource utilization rate of the underlying physical network will become poor, that is, the revenue/expenditure (Revenue/Cost) of the system will decrease.

Contents of the invention

The present invention analyzes the relationship between the node mapping algorithm and the link mapping algorithm, and finds that during the node mapping process, if only the remaining resources of the underlying physical network nodes are targeted, the above-mentioned negative impact on the link mapping is almost unavoidable, so if A method can be proposed that considers the distance between nodes while considering the remaining resources of the underlying physical network nodes, and uses parameters to adjust the influence of these two factors, which can greatly simplify the complexity of link mapping and improve the underlying Resource utilization of the physical network.

According to the starting point, the present invention introduces a proximity principle for node mapping, and uses the correlation coefficient Corr to adjust the weight of the two factors of the remaining resources of the underlying physical network node and the node distance, forming a new and more comprehensive node mapping judgment standard: weighted Remaining resources (WeightedAvailable Resource, WAR). Taking the weighted remaining resources as the judgment standard of node mapping, the principle of node proximity is added as another influencing factor in the node mapping process in addition to the remaining resources of the underlying physical network nodes, so that link mapping is taken into account in the node mapping process, avoiding only Considering the adverse impact on link mapping caused by the remaining resources of the underlying physical network nodes, the complexity of link mapping is reduced, the utilization rate of the underlying physical network link resources is improved, and the income-expenditure ratio of the system is improved.

Definitions involved in the present invention:

1) Revenue

Revenue refers to the profit obtained by successful virtual network mapping, defined according to the successful virtual network bandwidth and CPU:

Revenue＝α _R ∑BW _R +β _R ∑CPU _R

Among them, BW _R is the successfully mapped virtual network bandwidth, CPU _R is the successfully mapped virtual network node CPU resources, α _R and β _R are weight coefficients used to adjust bandwidth and CPU, and can also be understood as providing virtual network services for operators The unit price of hourly bandwidth and CPU resources, and the subscript R refers to revenue.

2) Expenditure (Cost):

Cost refers to the cost of successful virtual network mapping, defined according to the underlying physical network bandwidth and CPU used for mapping:

Cost＝α _C ∑HOPS·BW _C +β _C ∑CPU _C

Among them, BW _C is the underlying physical network bandwidth used for mapping, HOPS is the actual number of links occupied by a virtual link on the underlying physical network, CPU _C is the underlying physical network node CPU used for mapping, α _C and β _C are used It can also be understood as the cost of bandwidth and CPU resources when operators provide virtual network services, and the subscript C refers to Cost.

3) Residual resources (Available Resource, AR):

AR refers to the remaining resources of an underlying physical network node, defined according to the remaining CPU resources and the remaining link resources connected to the underlying physical network node:

AR＝CPU _A ∑BW _A

Wherein, Corr is a correlation coefficient for a certain underlying physical network node (S _node ). As Corr increases, the effect of the proximity principle is enhanced, and Corr>1. The index n indicates the number of virtual network nodes that have been successfully mapped in the current virtual network request and the mapped image nodes are directly connected to the S _node . The size of n represents the distance between the S _node and the underlying physical network nodes that have been successfully mapped before. The following The mark A refers to AR.

The present invention, firstly, introduces the principle of proximity by defining weighted residual resources, and takes into account link mapping while node mapping. The second is to introduce a mechanism of sorting V _nodes according to their remaining resources in the virtual network node mapping process. The third is to introduce the mechanism of sorting V _hnk according to its bandwidth in the process of virtual network link mapping.

(1) Proximity principle

As mentioned above, in the existing virtual network node mapping process, when looking for the underlying physical network nodes, the measurement standard is the remaining resources of the nodes. This measurement standard only considers the remaining resource size of the underlying physical network nodes, and does not consider link mapping. The distance problem, so the present invention proposes to use the weighted residual resources as a measure, defined as follows:

WAR＝Corr ⁿ CPU _A ∑BW _A

Among them, Corr is the correlation coefficient for an underlying physical network node S _node . As Corr increases, the effect of the proximity principle is enhanced, and it should be set as Corr>1. The index n indicates the number of virtual network nodes that have been successfully mapped in the current virtual network request and the mapped image node is directly connected to the S _node . The size of n represents the distance between the S _node and the underlying physical network node that has been successfully mapped before. n The larger the value, the closer the distance, and the easier the subsequent link mapping. After using the weighted remaining resources as the measurement standard, the remaining resources of the underlying physical network nodes and the distance of the link mapping are considered at the same time, which significantly reduces the complexity of the link mapping and improves the utilization of the underlying physical network link resources. , which increases the system's revenue-to-expenditure ratio.

(2) V _node sorting mechanism

In the process of mapping a single virtual network request, the existing virtual network node mapping method does not sort the V _nodes , so the unimportant virtual network nodes in the virtual network request are likely to be preferentially mapped to the underlying physical network nodes with the most abundant resources , then when the more important virtual network nodes are mapped, the underlying physical network nodes that meet the resource requirements may not be found, resulting in node mapping failures. Therefore, the present invention introduces a mechanism for sorting V _nodes according to their remaining resources, which ensures the priority mapping of important V _nodes to a certain extent, thereby improving the overall success rate of node mapping.

(3) V _hnk sorting mechanism

In the process of mapping a single virtual network request, the existing virtual network link mapping method does not perform V _hnk sorting, so the virtual network link with the smaller bandwidth in the virtual network request is likely to be preferentially mapped to the underlying physical network with the most abundant resources If there is a network link, then when mapping a virtual network link with a large bandwidth later, it may not be able to find an underlying physical network link that meets the bandwidth requirements, resulting in link mapping failure. Therefore, the present invention introduces a mechanism for V _hnks to be sorted according to their bandwidths, which guarantees the priority mapping of V _hnks with larger bandwidths to a certain extent, thereby improving the overall success rate of link mapping.

Description of drawings

Figure 1 Virtual network mapping problem

Figure 2 Virtual network mapping process in time window mode

Figure 3 The specific process of virtual network mapping

Implementation

The core of the specific operation process of the present invention is the time window, a virtual network mapping is performed within a time window, and the process of virtual network mapping in the time window mode is shown in Figure 2:

A. Release the underlying physical network resources occupied by the virtual network requests that left within the previous time window. The above virtual network requests include requests to complete services and requests that are actively rejected; virtual network requests include virtual network node requests and virtual network link requests two parts;

B. Statistics of virtual network requests arriving within this time window, virtual network requests include newly arrived requests and requeued requests;

C. Sort the virtual network requests counted in step B according to their revenue (Revenue) from large to small, and then map to the underlying physical network in order. If any one of the virtual network requests is successfully mapped, that is, the virtual network node and the virtual network If the link mapping is successful at the same time, the status of the underlying physical network will be updated; if the mapping fails, the virtual network request will be sent to the waiting queue and wait for the next time window; the number of re-queuing virtual network requests, such as the number of mapping failures, will be set in advance If the preset number of times is exceeded, the request will not be sent to the waiting queue, but will be rejected directly.

The virtual network mapping process ensures that virtual network requests can be processed in real time, and the admission mechanism of virtual network requests can be controlled by adjusting parameters (such as delaying and rejecting virtual network requests).

The core step of virtual network mapping is to map a group of virtual network requests within the time window to the underlying physical network (step C). This mapping process can be described by the following process, as shown in Figure 3

C1) First perform node mapping, and sort virtual network requests according to their income from large to small;

C2) Judging whether there is a virtual network request without node mapping after step C1 is sorted, if there is a virtual network request without node mapping, perform step C3, if there is no virtual network request without node mapping, perform step C9;

C3) Select the virtual network request with the largest income after sorting in step C1, and sort the virtual network nodes (V _node ) in the virtual network request according to their remaining resources (Available Resource, AR) from large to small;

C4) judging whether there is a virtual network node unmapped in the virtual network request selected in step C3, if there is an unmapped virtual network node, perform step C5, if there is no unmapped virtual network node, perform step C2;

C5) Select the virtual network node with the largest remaining resources after sorting in step C3, and select such a physical network node (S _node ) in the underlying physical network: its CPU resource is greater than the CPU resource of the above virtual network node, and the selected physical network node Nodes are sorted from largest to smallest according to their Weighted Available Resource (WAR);

C6) Determine whether there are physical network nodes that have not been mapped after the steps of C5 are sorted, if there are unmapped physical network nodes, perform step C7, and if there are no unmapped physical network nodes, perform step C8;

C7) assign the physical network node with the largest weighted remaining resources after sorting in the C5 step to the virtual network node with the largest remaining resources selected in the C5 step, and perform step C4;

C8) Send the virtual network request with the largest income selected in step C3 to the waiting queue or reject the request, and execute step C2;

C9) start the link mapping, and sort the virtual network requests with successful node mapping according to income from large to small;

C10) judging whether there is a virtual network request without link mapping after step C9 is sorted, if there is a virtual network request without link mapping, perform step C11, if there is no virtual network request without link mapping, the mapping algorithm ends;

C11) Select the virtual network request with the largest income after sorting in step C9, and sort the virtual network links (V _hnk ) in the virtual network request according to their bandwidths from large to small;

C12) judging whether there is a virtual network link unmapped in the virtual network request selected in the step C11, if there is an unmapped virtual network link, perform step C13, if there is no unmapped virtual network link, perform step C10;

C13) Select the virtual network link with the largest bandwidth after sorting in step C11, and use the K-shortest path (K-Shortest) algorithm to sequentially find the 1st to K shortest paths, which are composed of one or more underlying physical network links (S _link ), K is an integer greater than 1, and only those K paths that meet the bandwidth of the virtual network link are reserved;

C14) judging whether the physical network path reserved in the C13 step has not been mapped, if there is an unmapped physical network path, perform step C15; if there is no unmapped physical network path, perform step C16;

C15) assign the shortest path in the physical network path reserved in the C13 step to the virtual network link with the largest bandwidth selected in the C13 step, and perform step C12;

C16) Send the virtual network request with the highest income selected in step C11 to the waiting queue or reject the request, and execute step C10.

Claims (5)

1. A virtual network mapping method based on the principle of proximity, the steps of performing a virtual network mapping within a time window include: A. Release the underlying physical network resources occupied by the virtual network requests that left within the previous time window. The above virtual network requests include requests to complete services and requests that are actively rejected; virtual network requests include virtual network node requests and virtual network link requests two parts; B. Statistics of virtual network requests arriving within this time window, virtual network requests include newly arrived requests and requeued requests; C. Sort the virtual network requests counted in step B according to their revenue (Revenue) from large to small, and then map to the underlying physical network in order. If any one of the virtual network requests is successfully mapped, that is, the virtual network node and the virtual network If the link mapping is successful at the same time, the status of the underlying physical network will be updated; if the mapping fails, the virtual network request will be sent to the waiting queue and wait for the next time window; the number of re-queuing virtual network requests, such as the number of mapping failures, will be set in advance If the preset number of times is exceeded, the request will not be sent to the waiting queue, but will be rejected directly. 2. method as claimed in claim 1, above-mentioned step C comprises: C1) First perform node mapping, and sort virtual network requests according to their income from large to small; C2) Judging whether there is a virtual network request without node mapping after step C1 is sorted, if there is a virtual network request without node mapping, perform step C3, if there is no virtual network request without node mapping, perform step C9; C3) Select the virtual network request with the largest income after sorting in step C1, and sort the virtual network nodes (V _node ) in the virtual network request according to their remaining resources (Available Resource, AR) from large to small; C4) judging whether there is a virtual network node unmapped in the virtual network request selected in step C3, if there is an unmapped virtual network node, perform step C5, if there is no unmapped virtual network node, perform step C2; C5) Select the virtual network node with the largest remaining resource after sorting in step C3, and select the physical network node (S _node ) with the largest remaining resource in the underlying physical network: its CPU resource is greater than the CPU resource of the above virtual network node, and the selected Physical network nodes are sorted from large to small according to their Weighted Available Resource (WAR); C6) Determine whether there are physical network nodes that have not been mapped after the steps of C5 are sorted, if there are unmapped physical network nodes, perform step C7, and if there are no unmapped physical network nodes, perform step C8; C7) assign the physical network node with the largest weighted remaining resources after sorting in the C5 step to the virtual network node with the largest remaining resources selected in the C5 step, and perform step C4; C8) Send the virtual network request with the largest income selected in step C3 to the waiting queue or reject the request, and set the number of virtual network request requeues in advance. If the number of mapping failures exceeds the preset number, the request will no longer be sent to the waiting queue Queue, but directly reject; execute step C2; C9) start the link mapping, and sort the virtual network requests with successful node mapping according to income from large to small; C10) judging whether there is a virtual network request without link mapping after step C9 is sorted, if there is a virtual network request without link mapping, perform step C11, if there is no virtual network request without link mapping, the mapping algorithm ends; C11) Select the virtual network request with the largest income after sorting in step C9, and sort the virtual network links (V _link ) in the virtual network request according to their bandwidths from large to small; C12) judging whether there is a virtual network link unmapped in the virtual network request selected in the step C11, if there is an unmapped virtual network link, perform step C13, if there is no unmapped virtual network link, perform step C10; C13) Select the virtual network link with the largest bandwidth after sorting in step C11, and use the K-shortest path (K-Shortest) algorithm to sequentially find the 1st to K shortest paths, which are composed of one or more underlying physical network links (S _link ), K is an integer greater than 1, and only those K paths that meet the bandwidth of the virtual network link are reserved; C14) judging whether the physical network path reserved in the C13 step has not been mapped, if there is an unmapped physical network path, perform step C15; if there is no unmapped physical network path, perform step C16; C15) assign the shortest path in the physical network path reserved in the C13 step to the virtual network link with the largest bandwidth selected in the C13 step, and perform step C12; C16) Send the virtual network request with the largest income selected in step C11 to the waiting queue or reject the request, and set the number of virtual network requests to be re-queued in advance. If the number of mapping failures exceeds the preset number of times, the request will no longer be sent to the waiting queue Queue, but reject directly; go to step C10. 3. The method of claim 2, wherein Income refers to the profit obtained by successful virtual network mapping, defined according to the successful virtual network bandwidth and CPU: $\mathrm{<span\; class="notranslate">\; Revenue</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; BW</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&beta;</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; CPU</span>}}_{\mathrm{<span\; class="notranslate">\; R</span>}}$ Among them, BW _R is the successfully mapped virtual network bandwidth, CPU _R is the successfully mapped virtual network node CPU resources, α _R and β _R are weight coefficients used to adjust bandwidth and CPU, and the subscript R refers to Revenue. 4. The method of claim 2, wherein The remaining resources refer to the remaining resources of an underlying physical network node, defined according to the remaining CPU resources and the remaining link bandwidth resources connected to the physical network node: $\mathrm{<span\; class="notranslate">\; AR</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; CPU</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; BW</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}$ Wherein, BW _A is the remaining CPU resource of the underlying physical network node, CPU _A is the remaining bandwidth resource of the underlying physical network link connected to the underlying physical network node, and the subscript A refers to the remaining resource (Available Resource, AR). 5. The method of claim 2, wherein, Weighted Available Resource (WAR) is defined as: $\mathrm{<span\; class="notranslate">\; WAR</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; Corr</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; CPU</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}{\mathrm{<span\; class="notranslate">\; BW</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}$ Among them, Corr is the correlation coefficient for an underlying physical network node S _node . As Corr increases, the effect of the proximity principle is enhanced, and Corr is a real number greater than 1; the index n indicates that the current virtual network request has been successfully mapped and mapped. The number of virtual network nodes directly connected to the node and the S _node , and the size of n represents the distance between the S _node and the underlying physical network nodes that have been successfully mapped before.