WO2009124419A1 - 拓扑抽象方法、拓扑抽象装置以及路由控制器 - Google Patents

拓扑抽象方法、拓扑抽象装置以及路由控制器 Download PDF

Info

Publication number
WO2009124419A1
WO2009124419A1 PCT/CN2008/000741 CN2008000741W WO2009124419A1 WO 2009124419 A1 WO2009124419 A1 WO 2009124419A1 CN 2008000741 W CN2008000741 W CN 2008000741W WO 2009124419 A1 WO2009124419 A1 WO 2009124419A1
Authority
WO
WIPO (PCT)
Prior art keywords
topology
abstraction
way
fully connected
network
Prior art date
Application number
PCT/CN2008/000741
Other languages
English (en)
French (fr)
Inventor
万鹏
焦文华
Original Assignee
郎讯科技公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 郎讯科技公司 filed Critical 郎讯科技公司
Priority to US12/735,732 priority Critical patent/US8406154B2/en
Priority to JP2011503322A priority patent/JP2011517220A/ja
Priority to EP08733944A priority patent/EP2271028A1/en
Priority to PCT/CN2008/000741 priority patent/WO2009124419A1/zh
Priority to KR1020107025146A priority patent/KR20100133003A/ko
Publication of WO2009124419A1 publication Critical patent/WO2009124419A1/zh

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00Routing or path finding of packets in data switching networks
    • H04L45/02Topology update or discovery
    • H04L45/08Learning-based routing, e.g. using neural networks or artificial intelligence
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L45/00Routing or path finding of packets in data switching networks
    • H04L45/02Topology update or discovery

Definitions

  • Topological abstraction method topology abstraction device and routing controller
  • the present invention relates to network technologies, and more particularly to topology abstraction methods, topology abstraction devices, and routing controllers. Background technique
  • IP Internet Protocol
  • ATM Automatically Switched Optical Network
  • ASON Automatically Switched Optical Network
  • each routing domain For scalability and security reasons, the internal topology information of each routing domain is first abstracted by a specific topology abstraction method to be published to other routing domains in the network. In this way, each routing domain maintains only its own detailed topology information and abstract topology information of other routing domains, thereby greatly reducing the amount of information that needs to be stored and distributed in the network.
  • the topology abstraction process generally first constructs a fully connected topology consisting of boundary nodes based on the real topology.
  • the fully connected topology is then further compressed to be compressed into a more sparse topology such as a tree or star. Among them, further compression of the fully connected topology is optional.
  • Figure 1 shows an example of a topology abstraction process.
  • the topology 100 is a real topology
  • the topology 100 has 8 nodes 1-8 and 10 links, and two working wavelengths ⁇ and 1 2 .
  • nodes 1-4 are boundary nodes connected to the external peer domain
  • nodes 5-8 are internal nodes
  • the solid line/dashed line between the nodes respectively indicate that the wavelength channel of the link is idle, and a new connection can be established.
  • the internal nodes 5-8 of the topology 100 are hidden, leaving only the boundary nodes 1-4 and the resource availability between them.
  • the connectivity relationship between the four boundary nodes can be represented by a connectivity matrix as shown in equation (1).
  • Formula (1) Since the topology 101 shown in the figure has four boundary nodes 1-4, for each wavelength ⁇ and 1 2 , a 4x4 matrix can be used to represent the connectivity between the boundary nodes. When there are paths between two boundary nodes that can connect them (for example, node 1 and node 4 can communicate with each other through wavelength ⁇ path 1-5-6-3-4), the corresponding matrix element is 1, otherwise 0.
  • Topology 101 is a fully connected topology constructed from a connectivity matrix that includes only boundary nodes 1 -4. It can be seen that after the fully connected topology is constructed, the number of links is reduced to 5 (the worst case is 6, that is, all nodes are connected).
  • the topology 102 is a result of further compression based on the topology 101.
  • the redundant logical link in the original topology 101 is deleted.
  • the wavelength logical link between node 3 and node 4 is replaced by path 3-1-4
  • the 12- wavelength logic between node 2 and node 3 The link is replaced by path 2-1-3.
  • the total number of links is further reduced to three.
  • symmetric node method Most of the existing topology abstract technologies are designed for ATM networks. There are three common methods: symmetric node method, full connectivity method, and star method.
  • the basic idea of the symmetric node method is to combine all the boundary nodes in the real network topology into a single virtual node, and use a common value to represent the connection properties between the original boundary nodes.
  • the advantage of this method is that only a very small amount of link information needs to be exchanged. But the disadvantage is that the information provided is too coarse and inaccurate, which can result in the resources in the domain not being used properly.
  • the fully connected approach looks at the accuracy of abstract information. It assumes that all boundary nodes in the real network topology are connected by logical links, and each logical link is configured with one or more QoS parameters, such as delay or bandwidth.
  • This method accurately preserves the connectivity attribute between the original real topology boundary nodes, but since it must maintain the information of N ( N - 1 ) /2 (N is the number of boundary nodes) logical links, when the network size is large The scalability is very poor.
  • the star method assumes that there is a centrally located virtual node, and all boundary nodes in the real network topology are connected to it through logical links. And each logical link can have different attributes. Therefore, the star method can represent more detailed link information, which is much more accurate than the symmetric node method. At the same time, the star method only needs to maintain the information of N logical links, and has good scalability compared with the fully connected method, and is suitable for a large-scale network.
  • the present invention aims to provide a topology abstract scheme that is more suitable for an optical network.
  • a topology abstraction method comprising: obtaining a link separation path number ⁇ , ) between each pair of boundary nodes in a real topology G R ;
  • the connectivity matrix C is obtained by using the link separation path number ⁇ , ⁇ ) and the corresponding fully connected topology G B is constructed .
  • a topology abstracting apparatus including: a link separation path number obtaining apparatus configured to obtain a real topology (the number of link separation paths between each pair of boundary nodes in the ⁇ ) ⁇ , ⁇ ) ; and
  • the topology constructing device is configured to obtain the connectivity matrix C by using the number of separated paths of the link and construct a corresponding fully connected topology G B .
  • a routing controller comprising: the topology abstracting device described above, and
  • the link state sending device is configured to broadcast LSA information based on the topology generated by the topology abstraction device, and send the LSA information.
  • FIG. 2 shows a flow chart of a topology abstraction method in accordance with one embodiment of the present invention
  • 3 shows a flow chart of a topology compression process in accordance with one embodiment of the present invention
  • FIG. 4 shows an example of a two-way shuffling network topology
  • Figure 5 is a flow chart showing the process of abstracting into a two-way shuffled network topology in accordance with one embodiment of the present invention
  • FIG. 6 shows a flow chart of a process for optimizing a two-way shuffle network topology in accordance with one embodiment of the present invention
  • Figure 7 shows how nodes in a fully connected topology are mapped into a two-way shuffling network;
  • Figure 8 shows an example of hybridization and mutation operations in accordance with one embodiment of the present invention;
  • Figure 9 shows A schematic block diagram of a topology abstraction device of an embodiment;
  • Figure 10 shows a schematic block diagram of a topology abstraction apparatus in accordance with another embodiment of the present invention.
  • Figure 11 is a schematic block diagram showing a silent shuffle network topology optimization apparatus in a topology abstraction apparatus according to another embodiment of the present invention.
  • Figure 12 shows a schematic block diagram of a topology abstraction apparatus in accordance with another embodiment of the present invention.
  • Figure 13 shows a schematic block diagram of a routing controller in accordance with one embodiment of the present invention.
  • Figure 14 shows a flow chart of the beginning phase of a routing controller in accordance with one embodiment of the present invention
  • Figure 15 shows a flow diagram of the operational phase of a routing controller in accordance with one embodiment of the present invention.
  • FIG. 2 shows a flow chart of a topology abstraction method in accordance with one embodiment of the present invention. Its style
  • step 210 the true topology of the domain is obtained:
  • link-diversity path means the path without any shared links between the two border nodes.
  • the calculation of the number of link separation paths between two boundary nodes is converted into the calculation of the maximum flow between the two boundary nodes, because the number of link separation paths between the two boundary nodes can be equivalent to The largest flow between source and sink in a pipe network. Therefore, in this embodiment, the number of link separation paths between two boundary nodes is calculated by calculating the maximum flow between two boundary nodes by using the maximum flow algorithm.
  • c(i, , k ) MAXFLOW (i, j, ) Equation (4)
  • the existing highest-label-preflow-push algorithm is used to calculate two Maximum flow between boundary nodes.
  • the Augmenting Path Algorithm is used to calculate the maximum flow between two boundary nodes. Then, in step 240, using the link separation path number c ', ), the connectivity matrix C is obtained and a corresponding fully connected topology is constructed (3 ⁇ 4.
  • the connectivity matrix C of the present invention is different from the conventional connectivity matrix shown in the equation (1).
  • the connectivity matrix of the present invention is not represented by 0 and 1, but by the number of link separation paths between two boundary nodes. Constructing a connectivity matrix by using the number of link separation paths (:, not only provides information on available wavelengths, but also provides resource information related to the wavelength. Therefore, this richer information can be utilized to help select the optimal path, so that performance is obtained. For example, when establishing a connection, it is preferable to use a wavelength with a small amount of resources and a small possibility of exhaustion.
  • the resulting fully connected topology is further compressed to obtain a more sparse topology, thereby further reducing the amount of information exchanged between domains.
  • FIG. 3 shows a flow chart of a topology compression process in accordance with one embodiment of the present invention.
  • a shuffle-net topology is constructed for the fully connected topology to obtain a two-way shuffle network topology with all nodes vacant.
  • Two-way shuffled network topology is an extension of the traditional shuffled network topology
  • the difference between the joyful shuffling network topology and the traditional shuffled network topology is that its links are bidirectional, that is, the nodes in each column not only have the corpse links to the nodes in the next column, but also A corpse link with a node to the previous column.
  • Each node has 2 links to the nodes of the next column and 2 links to the nodes of the previous column.
  • the number of logical links can be reduced to 7, which is much lower than the number of logical links of the fully connected topology, and thus reduces the link information that needs to be flooded.
  • the quantity reduces the load on the signaling network.
  • step 510 for the fully connected topology, parameters K and P of the corresponding bidirectional shuffle network topology are calculated, where K, P can be solved according to the following formula: Max ⁇ ( ⁇ -1 ) ⁇ ⁇ ' ⁇ , ⁇ ( ⁇ -1 ) K ⁇ N ⁇ ⁇ ⁇ ( ⁇ , ⁇ > 2 ) Equation (7), where ⁇ is the number of nodes in the fully connected topology, ie in the actual network The number of boundary nodes.
  • step 520 using the parameters and corpses obtained according to equation (7), the two-way shuffling network topology is initialized to obtain an initialized two-way shuffling network topology:
  • step 320 the nodes in the fully connected topology are mapped to the nodes in the shuffle-net through a certain mapping relationship, and the corresponding logical link attributes are assigned to the bidirectional links in the shuffle-net ( The following is described).
  • Figure 6 illustrates a flow chart for optimizing the joy to shuffle network topology in accordance with one embodiment of the present invention.
  • the two-way shuffling network topology is optimized using a genetic algorithm.
  • the first generation V of the two-way shuffled network topology is first obtained. It includes thousands of chromosomes p, and the chromosome p can represent the arrangement of the fully connected topology nodes in the two-way shuffled network topology. . In one embodiment, the chromosome is constructed using one-dimensional text permutation coding.
  • chromosome 7 2 3 1 5 6 4 8 which represents the order in which the nodes of the fully connected topology are arranged in the bidirectional shuffle-net (where node 8 is the virtual node filled in the number of complements), ie Node 7 in the connectivity topology corresponds to the (0, 0) node in the bidirectional shuffle-net, 2 corresponds to the (1, 0) node, ..., 4 corresponds to the (2, 1) node, and the virtual node 8 is located at (3, 1) (as shown in Figure 7).
  • the chromosomes in the present invention are different from the chromosomes represented by binary representations usually composed of 0, 1.
  • the present invention utilizes one-dimensional text permutation coding to construct chromosomes, each of which is a node number. Therefore, the mutation operation of the chromosome of the present invention is a position-based operation, which usually changes the position of a gene in a chromosome, thereby ensuring that the next generation can inherit most of the characteristics of the previous generation and retain compatible genes.
  • step 620 after mapping all the nodes ⁇ of the fully connected topology to the node V' of the bidirectional shuffled network topology G' for each chromosome p, logical link mapping can be performed according to the node position.
  • node 7 is connected to nodes 4, 5, and 6, and the bidirectional logical links (4, 7), (5, 7), (6, 7) in the fully connected topology. Reserved, the rest of the logical links connected to node 7 are deleted.
  • Node 8 is a virtual node and does not participate in logical link mapping.
  • the connectivity matrix C can be obtained as obtained by the connectivity matrix C described above.
  • the fitness value of each chromosome is obtained by taking the connectivity matrix between the real topology boundary nodes as an objective function:
  • Deviatiorijj ⁇ .
  • the fitness value represents the degree of closeness between the connectivity matrix of the fully connected topology and the connectivity matrix of the corresponding topology of the chromosome, and the value is the inverse of the sum of the deviation degrees.
  • the negative sign indicates that the smaller the deviation value is, the larger the fitness value is; the deviation value is the difference between the target connectivity matrix and the connectivity matrix of the chromosome corresponding topology and the target connectivity matrix is connected with the chromosome corresponding topology.
  • the degree of deviation is an absolute value of a difference between a target connectivity matrix and a chromosome corresponding topological connectivity matrix.
  • step 640 genetic manipulation is performed based on the obtained fitness value to obtain the next generation chromosome.
  • a selection operation is performed on the chromosome of this generation based on the obtained fitness value to obtain a chromosome for breeding the next generation. That is to say, based on the obtained fitness value, chromosomes with low fitness (for example, below a threshold or below a certain ratio) are eliminated, and only those with high fitness are selected to the next generation.
  • hybridization and mutation of the selected chromosome are performed to obtain the next generation chromosome.
  • Figure 8 shows an example of hybridization and mutation operations in accordance with one embodiment of the present invention. It should be understood that in the present invention, it is not limited to the hybridization and mutation operations shown in Fig. 8.
  • step 650 it is determined if the genetic algebra has reached a predetermined threshold. If the predetermined threshold has been reached, then the resulting best fit chromosome is used as an optimization result in step 660, that is, an optimized two-way shuffled network topology is obtained. If the predetermined threshold has not been reached, then jump to step 620 to continue the iteration.
  • the genetic operation is stopped, and The obtained chromosome arrangement and the connected two-way shuffled network topology are used as a result of the topology.
  • the genetic algebra can be determined as needed. The more algebraic the genetics, the better the results obtained, but more time is needed to calculate.
  • the topology abstract method of the present invention can obtain a more accurate topology and can provide more accurate and rich information for routing decisions, thereby increasing performance.
  • the topology abstraction method of the present invention provides detailed link information suitable for an optical network. More specifically, by optimizing the two-way shuffled network topology using the genetic algorithm, a more accurate topology is described with fewer logical link numbers, thereby improving performance.
  • the node pairs having more link separation paths in the fully connected topology are mapped to the directly connected nodes in the joyful shuffling network.
  • the heuristic method is used to compress the fully connected topology into a two-way shuffled network topology.
  • the fully connected topology is compressed into a symmetric star network using the resulting connected matrix of fully connected topologies.
  • the re-abstraction is performed based on a predetermined time interval.
  • a re-abstracting time interval is defined in advance, based on which the re-abstracting is periodically performed regardless of the change in the link bandwidth within the domain.
  • an event-based re-abstracting strategy is employed that performs re-abstracting only when a predetermined event occurs.
  • a dull emphasis abstraction strategy may be employed, ie only when a large topology change occurs or the connection rejection rate reaches a predetermined threshold When you perform heavy abstraction.
  • the re-abstract is performed when the predetermined connection rejection rate is reached. For example, during a predetermined re-abstraction detection interval, the re-abstract is performed immediately when the connection rejection rate r exceeds the predetermined threshold a, wherein the predetermined threshold ⁇ may be a constant based on the customer's QoS requirement, for example, may be 20%.
  • a heavy abstraction is performed when there is a large change in the topology.
  • the capacity of the link is one channel, and the number of available channels at the time is c t .
  • the number of available channels at the time is equal to the value of the element in the predetermined sequence of values, a re-abstract is performed.
  • the predetermined value sequence ⁇ has more elements such that when the number of available channels is larger, the topology re-abstract frequency is lower, and when the number of available channels is small, the topology re-abstract frequency is higher.
  • a predetermined sequence of values can be obtained by:
  • re-abstracting strategy of the present invention re-abstraction is performed only when a large topology change occurs, and no re-abstract is performed when the topology change is small. Therefore, under the premise of ensuring that the abstract topology is updated in time, the number of times of abstraction is reduced, thereby reducing resources consumed by performing heavy abstraction and improving performance.
  • the method of the present invention utilizes the "link separation path number" to construct a connectivity matrix, which not only provides usability information of available wavelengths, but also provides resource-rich resource richness information, and thus is a routing decision. More detailed link information is provided to optimize resource allocation and improve network performance.
  • the obtained fully connected topology is compressed, thereby reducing the amount of information exchanged between domains, and further improving performance.
  • compressing the fully connected topology into a two-way shuffled network topology greatly reduces the number of links, thereby reducing the amount of routing information that needs to be exchanged between domains and improving performance.
  • the event-based blunt re-abstracting strategy is used to perform re-abstracting, thereby reducing resources consumed by performing heavy abstraction, further improving performance.
  • FIG. 9 illustrates a topology abstraction apparatus 900 in accordance with one embodiment of the present invention.
  • the topology abstraction apparatus 900 includes: a link separation path number obtaining means 910 configured to obtain the number of link separation paths c(i, j, between each pair of boundary nodes in the real topology. k );
  • the topology constructing means 920 is configured to obtain the connectivity matrix C by using the link separation path number c ( , , A) and construct a corresponding fully connected topology G B .
  • the link separation path number obtaining means 910 can perform operations according to the description of the steps 210-230 described above: First, the adjacency matrix is calculated according to the real topology of the domain, and then the boundary node is obtained by calculating the maximum flow between the boundary nodes. The number of paths separated between the links. In calculating the number of link separation paths, the highest label pre-flow advancement algorithm, the augmentation path algorithm, or other maximum flow algorithms known in the art may be utilized.
  • the topology constructing means 920 can perform operations according to the description of the step 240 described above: constructing a connectivity matrix by using the number of link separation paths obtained by the link separation path number obtaining means 910.
  • the topology abstraction apparatus 900 of the present invention constructs a connectivity matrix by using the number of link separation paths, which not only provides usable wavelength information, but also provides resource information related to the wavelength richness. This can help with routing and improve performance.
  • the topology abstraction apparatus 900 further includes: a topology compression apparatus 930 configured to compress the fully connected topology to further reduce the amount of information exchanged between domains.
  • the topology compression device 930 includes: The two-way shuffling network topology construction device 932 is configured to initialize the two-way shuffle network topology according to the determined structural parameters of the two-way shuffle network topology; and the two-way shuffling network topology optimization device 934, configured The two-way shuffling network topology is optimized using a connectivity matrix of the fully connected topology.
  • the two-way shuffling network topology optimization device 934 can further optimize the two-way shuffling network topology based on a genetic algorithm.
  • Figure 11 illustrates an embodiment of a genetic algorithm based two-way shuffling network topology optimization device 934.
  • the two-way shuffling network topology optimization device 934 includes: a first generation generating device 9342 configured to obtain a first generation of a two-way shuffled network topology in a random manner; a connectivity matrix obtaining device 9344 configured to obtain a current generation a connectivity matrix of chromosomes; fitness value calculation means 9346 configured to calculate a fitness value of the current generation of chromosomes with a connectivity matrix between real topographic boundary nodes; and genetic manipulation device 9348, configured Performing a genetic operation on the chromosome according to fitness values to obtain a next generation chromosome, wherein the genetic operation includes a selection operation, a mutation operation, a hybridization operation, etc.; mapping means 9350, configured to utilize when conditions
  • the fitness value calculated by the fitness value calculation device 9346 is the absolute value of the difference between the connectivity matrix of the current generation chromosome and the connectivity matrix between the real topology boundary nodes; In another embodiment, the fitness value is a difference between a connectivity matrix of the current generation chromosome and a connectivity matrix between the real topology boundary nodes, and a connectivity matrix of the current generation chromosome The absolute value of the ratio of the sum of the connectivity matrices between the real topological boundary nodes.
  • the condition for terminating the genetic manipulation is that the genetic algebra exceeds a predetermined value or the fitness value reaches a predetermined value.
  • the chromosomes are encoded based on a one-dimensional text arrangement.
  • the topology abstraction apparatus 900 further includes a re-abstract triggering device 940 configured to trigger a red abstraction when a network state changes or to trigger a heavy abstraction based on a predetermined time interval.
  • the re-abstract triggering device 940 triggers a re-abstract based on a dull re-abstracting strategy when a predetermined topology change event occurs or the connection rejection rate reaches a predetermined threshold.
  • the predetermined topology change event may be an event in which the available resources are increased or decreased to a predetermined value, and the heavy abstraction occurs less frequently when there are more available resources, and occurs more frequently when the available resources are less.
  • the present invention also provides a routing controller.
  • Figure 13 shows a schematic block diagram of a routing controller in accordance with one embodiment of the present invention.
  • the routing controller 1300 includes: a real topology obtaining device 1310 for obtaining a real topology of a domain; a topology abstracting device 900 according to the present invention, for authenticating a real topology of a domain
  • the link state sending device 1320 is configured to generate Link-State Advertisement (LSA) information based on the topology obtained by the topology abstracting device 900 and send the LSA information.
  • LSA Link-State Advertisement
  • the topology abstraction device 900 constructs a connectivity matrix by using the number of link separation paths, which not only provides available wavelength information, but also provides resource information related to the wavelength. This can help with routing and improve performance.
  • Figure 14 is a flow chart showing the processing of the routing controller of the present invention at the beginning.
  • the routing controller starts operating, first in step 1410, the real topology obtaining device obtains a real topology of a domain. Then in step 1420, the topology abstraction device performs a topological abstraction of the real topology.
  • the link state transmitting device generates an LSA based on the abstract topology and sends it out.
  • the topology device triggers a heavy abstraction when it satisfies the heavy abstraction condition.
  • a red abstraction can be triggered when the network state changes or a heavy abstraction is triggered based on a predetermined time interval.
  • a red abstraction is triggered when a predetermined topology change event occurs or the connection rejection rate reaches a predetermined threshold.
  • the predetermined topology change event may be an event that increases or decreases the available resources to a predetermined value, and the predetermined topology change event occurs at a lower frequency when the available resources are more, and the available resources are available. When there are fewer, the frequency is higher.
  • Figure 15 is a flow chart showing the processing of the routing controller during the working phase, in accordance with one embodiment of the present invention.
  • step 1501 when the request message is received, it is first determined in step 1501 whether the request is a connection establishment request or a connection removal request. If it is a connection establishment request, the process proceeds to step 1502.
  • step 1502 the resource between the specified ingress node and the egress node is verified. If the check indicates that there is not enough resources between the designated ingress node and the egress node to support the connection request, the flow proceeds to step 1504. At step 1504, a message is sent to the requester to inform the requester that the connection failed. Then, in step 1505, the number of connection failures is increased, and the connection rejection rate is calculated, that is, the number of connection failures divided by the sum of the number of connection successes and the number of connection failures.
  • step 1506 it is determined whether the rejection rate has reached a predetermined rejection rate threshold. If the predetermined threshold is reached, then a re-abstract is triggered at step 1513 for topology abstraction. The connection is then successfully completed and the failure count is cleared in step 1514 to recount. Next, in step 1515, an LSA is generated based on the re-abstract topology, and the LSA is sent out. At this point, the processing for one request ends.
  • step 1507 the request is sent to the next domain and a connection establishment indication is awaited. After the indication, at step 1508, wavelengths are assigned along selected links within the domain to establish a connection. Then, in step 1509, the connection success count is incremented, and the number of connection successes is increased by one. Then in step 1510, a flooding is triggered again to send the LSA information. Next, in step 1511, the remaining resources between the ingress node and the egress node are checked, and in step 1512, it is determined whether the remaining resources reach the threshold of re-abstract. .
  • step 1513 where heavy abstraction is performed.
  • the connection is then successfully completed and the failure count is cleared at step 1514.
  • the flow then proceeds to step S1515 to end the processing of the request. If the threshold L a has not been reached, the processing of the request is directly ended.
  • step 1501 if the request is a connection removal request, the flow proceeds to the step 1503. At this step, the removal of the connection is performed. The flow then proceeds to step 1510 where the subsequent steps as previously described are performed until the end.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Description

拓朴抽象方法、 拓朴抽象装置以及路由控制器 技术领域
本发明涉及网络技术, 更具体地涉及拓朴抽象方法、 拓朴抽象 装置以及路由控制器。 背景技术
无论 IP 网络、 ATM 网络还是近年来出现的自动交换光网络 ( ASON, Automatic Switched Optical Network )内部都包括多个路由 域。
出于可扩展性和安全性考虑, 每一个路由域的内部拓朴信息首 先通过特定的拓朴抽象方法加以抽象才能发布到网络中的其他路由 域。 这样, 每个路由域只维护自身的详细拓朴信息以及其他路由域 的抽象拓朴信息, 从而大大减少了网络中需要存储和发布的信息量。
拓朴抽象过程一般首先基于真实拓朴构造一个由边界节点构成 的全连通拓朴。 然后, 对该全连通拓朴进行进一步压缩, 以压缩为 树型或星型等更加稀疏的拓朴。 其中, 对全连通拓朴进行进一步压 缩是可选的。
图 1 示出了拓朴抽象过程的一个实例。 如图 1 所示, 拓朴 100 是真实拓朴, 该拓朴 100具有 8个节点 1-8和 10条链路, 以及两个 工作波长 ^和 12。 其中, 节点 1-4是连接到外部对等域的边界节点, 节点 5-8是内部节点, 节点间的实线 /虚线分别表示链路的 波长 通道空闲, 可以建立新的连接。
拓朴抽象的过程中, 拓朴 100的内部节点 5-8被隐藏, 只保留边 界节点 1-4及它们之间的资源可用情况。 4个边界节点间的连通关系 可以由如式子 ( 1 ) 所示的连通性矩阵来表示。 式子 ( 1 )
Figure imgf000003_0001
由于图中示出的拓朴 101具有四个边界节点 1-4, 因此对于每个 波长 ^和 12 , 可以用一个 4x4矩阵来表示边界节点之间的连通性。 当 两个边界节点之间存在可以使它们连通的路径时 (例如节点 1 和节 点 4之间可以通过波长 ^路径 1-5-6-3-4相通) , 相应的矩阵元素为 1, 否则为 0。
拓朴 101 是根据连通性矩阵构造出来的全连通拓朴, 其仅包括 边界节点 1 -4。 可以看出, 经过全连通拓朴构造后, 链路数减少为 5 (最坏情况为 6, 即所有节点间都是连通的) 。
拓朴 102是在拓朴 101的基础上进一步压缩得到的结果。 其中, 原拓朴 101 中的冗余逻辑链路被删除,例如节点 3和节点 4之间的 波长逻辑链路被路径 3-1-4替代, 节点 2和节点 3之间的 12波长逻辑 链路被路径 2-1 -3替代。 经过压缩后, 链路总数进一步减少为 3。
现有拓朴抽象技术大多是针对 ATM网络而设计, 常见的有三种 方法: 对称节点方法、 全连通方法以及星型方法。 对称节点方法的 基本思想是: 将真实网络拓朴中的所有边界节点都合并成单个虚拟 节点, 并利用某个公共值来表征原边界节点之间的连接属性。 该方 法的优点在于, 仅需要交换极少量的链路信息。 但是缺点在于提供 的信息太粗略并且不准确, 这会导致域内资源得不到合理利用。 全 连通方法着眼于抽象信息的准确性。 它假设真实网络拓朴中的所有 边界节点都通过逻辑链路连接, 并且每个逻辑链路配置有一个或多 个 QoS参数, 诸如延迟或者带宽。 该方法精确地保留了原真实拓朴 边界节点间的连通属性, 但是由于它必须保持 N ( N - 1 ) /2(N为边 界节点数)条逻辑链路的信息,因此当网络规模较大时可扩展性很差。 星型方法假设有个位于中央的虛拟节点, 真实网络拓朴中的所有边 界节点都通过逻辑链路与其连接。 并且每条逻辑链路可以具有不同 的属性。 因此, 星型方法可以表示比较详细的链路信息, 比对称节 点方法准确得多。 同时星型方法仅需维持 N条逻辑链路的信息, 相 比全连通方法具有良好的可扩展性, 适用于较大规模的网络。 然而, 当上述针对 ATM网络而设计的拓朴抽象方法应用到光网 络时, 因需满足波长连续性约束条件, 上述方法就太过粗略和不准 确, 因此无法得到良好的性能。 因此对于例如 ASON的光网络而言, 就需要一种更加适合的拓朴抽象方案。 发明内容
为此, 本发明旨在提供更加适合于光网络的拓朴抽象方案。
根据本发明的一个方面, 提供了一种拓朴抽象方法, 包括: 得到真实拓朴 GR 中每对边界节点之间的链路分离路径数 Φ·, ) ; 和
利用所述链路分离路径数 Φ·, Α)获得连通性矩阵 C并构造相应 全连通拓朴 GB
根据本发明的又一方面, 提供了一种拓朴抽象装置, 包括: 链路分离路径数获取装置, 配置用于得到真实拓朴(^中每对边 界节点之间的链路分离路径数 φ·, Α) ; 和
拓朴构造装置,配置用于利用所述链路分离路径数 获得连 通性矩阵 C并构造相应全连通拓朴 GB
根据本发明的又一方面, 提供了一种路由控制器, 包括: 前面所述的拓朴抽象装置, 以及
链路状态发送装置,配置成用于基于所述拓朴抽象装置得到的拓 朴生成链路状态广播 LSA信息, 并将所述 LSA信息发送出去。
通过本发明, 可以为路由决策提供更多的链路信息, 并且可以提 供更加准确的拓朴。 附图说明
通过以下结合附图的详细说明, 并且随着对本发明的更全面了 解, 本发明的其他目的和效果将变得更加清楚和易于理解, 其中: 图 1示出了根据现有技术的拓朴抽象过程的一个实例;
图 2示出了根据本发明的一个实施例的拓朴抽象方法的流程图; 图 3示出了根据本发明的一个实施例的拓朴压缩过程的流程图; 图 4示出了双向洗牌网拓朴的一个实例;
图 5 示出了根据本发明的一个实施例的抽象成双向洗牌网拓朴 的过程的流程图;
图 6 示出了根据本发明的一个实施例的对双向洗牌网拓朴进行 优化的过程的流程图;
图 7示出了如何将全连通拓朴中的节点映射到双向洗牌网中; 图 8示出了根据本发明的一个实施例的杂交和变异操作的实例; 图 9 示出了根据本发明的一个实施例的拓朴抽象装置的示意方 框图;
图 10示出了根据本发明的另一实施例的拓朴抽象装置的示意方 框图;
图 1 1示出了根据本发明的另一实施例的拓朴抽象装置中的默向 洗牌网拓朴优化装置的示意方框图;
图 12示出了根据本发明的另一实施例的拓朴抽象装置的示意方 框图;
图 13示出了根据本发明的一个实施例的路由控制器的示意方框 图;
图 14示出了根据本发明的一个实施例的路由控制器的开始阶段 的流程图;
图 15示出了根据本发明的一个实施例的路由控制器的工作阶段 的流程图。
在所有的上述附图中, 相同的标号表示相同、 相似或相应的特 征或功能。 具体实施方式
在下文中, 将参考附图通过具体实施例对本发明进行详细地描 述。
图 2 示出了根据本发明一个实施例的拓朴抽象方法的流程图。 其具式
在有子本中发明的实施例的以下描述中, 针对的是对 ASON中的一个域的 拓朴进行抽象。
如图 2所示, 首先在步骤 210, 得到该域的真实拓朴:
GR = {V, E) 式子 (2) 其中, 表示拓朴中的节点的集合, 即
Figure imgf000007_0001
v2, ..., vn}, E表示 节点之间的链路的集合, 即^; ={e/, e2, ..., em}, 其中每条链路可以 支持波长 W、 λ2 中的一个波长或多个波长。 其中 "为节 点总数, 为链路的总数, 为波长总数。 应指出的是, 可以是 一个多重图, 即两个节点间存在多条光纤链路。
接着, 在步骤 220, 针对波长 、 λ2 中的每一个, 得到一个包含其所有节点的邻接矩阵:
( 3 )
( , J' = 1, ...,", k = \,...,W) 。 对于波长 , 若节点 i 和节点 _/ 一条或者以上的链路, 则 ·,Α)=链路数, 否则 ^/Α)=ο。 接着,在步骤 230,根据所述邻接矩阵 得到每对边界节点之间 的链路分离路径数 c(i,j',Ak) 。 此处使用的术语 "链路分离路径
( link-diversity path ) " 是指在两个边界节点之间没有任何共享链路 的路径。
在该实施例中, 将两个边界节点之间链路分离路径数的计算转 换为两边界节点之间的最大流的计算, 这是因为两个边界节点之间 链路分离路径数可以等同于管道网络中源到宿之间的最大流。 因此, 在该实施例中, 通过利用最大流算法计算两个边界节点之间的最大 流来计算两个边界节点之间的链路分离路径数,
c(i, , k ) = MAXFLOW (i, j, ) 式子 ( 4 ) 在一个实施例中 , 使用现有的最高标号预流推进算法 ( highest-label-preflow-push algorithm ) 来计算两个边界节点之间最 大流。 在一个替代实施例中, 使用了增广路算法 (Augmenting Path Algorithm ) 来计算两个边界节点之间的最大流。 然后, 在步骤 240, 利用所述链路分离路径数 c ' , ), 获得连 通性矩阵 C并构造相应全连通拓朴(¾。 其中
C = [c(i,j, k)]NxNxl¥ 式子 (5 )
GB= {VB, EB) 式子 (6) 其 中 , M = N(N— 1)/2 , VB ={v{,V2,...,VN} EB ={e e2,...,eM} j, ) = c{i,j, k) o
从上述实施例可以看出, 本发明的连通性矩阵 C与式子( 1 ) 中 所示的传统连通性矩阵不同。 本发明的连通性矩阵不是由 0和 1 来 表示, 而是由两个边界节点之间的链路分离路径数来表示。 利用链 路分离路径数来构造连通性矩阵 (:, 不但提供了可用波长的信息, 而且提供了该波长相关的资源信息。 因此可以利用该更加丰富的信 息来帮助选择最优路径, 使得性能得到提高。 例如, 在建立连接时, 可以优选使用资源多且用尽的可能性较小的波长。
在另一优选的实施例中, 还进一步对得到的全连通拓朴进行压 缩, 以得到更为稀疏的拓朴, 从而进一步减少域间交换的信息量。
图 3示出了根据本发明的一个实施例的拓朴压缩过程的流程图。 如图 3 所示, 在步骤 310, 针对全连通拓朴进行双向洗牌网 ( shuffle-net )拓朴构造, 以得到一个所有节点空置的双向洗牌网拓 朴。
传统洗牌网 (尸, 拓朴是具有 , = (Λ K=\, 2, …)个 节点, 并且这 N,个节点排列成 列、 尸 行的网络。 其中每列中的每 个节点具有到下一列的节点的尸条链路, 并且最后一列还与第一列 连接, 整个构成了一个圆柱体型拓朴。 即, 节点(/, n ) ( 1=0, 1, …,
PK-\ , n=0, 1,…, -1 )具有到下一列的节点(/,,",)、 (/,,",+1 )
(/,, ",+尸-1 )的尸条直接链路, 其中 / modi^1, «'= ( n+l ) mod K。 双向洗牌网拓朴是传统洗牌网拓朴的扩展。 欢向洗牌网拓朴与传 统洗牌网拓朴的不同之处在于它的链路都是双向的, 即每列中的节 点不仅具有到下一列的节点的尸条链路, 而且还具有到上一列的节 点的尸条链路。 图 4 示出了一个双向洗牌网拓朴的实例, 其中 Κ=2、 Ρ=2, Ν' =8。 如图所示, 8个节点排成 2列、 4行。 每个节点具有至下一列 的节点的 2条链路和至上一列节点的 2条链路。 利用该双向洗牌网 拓朴, 可以将逻辑链路数降低至 7 ,, 这与全连通拓朴 ^数量级的 逻辑链路数相比, 降低了很多, 从而减少了需要泛洪的链路信息数 量, 减轻了信令网的负荷。
下面将参考图 5 来详细描述本发明的双向洗牌网拓朴的构造的 过程。
如图 5所示, 首先, 在步骤 510, 针对所述全连通拓朴, 计算相 应的双向洗牌网拓朴的参数 K和 P,其中 K,P可以根据下式来求解: Max{ (Κ -1 )ΡΚ'Ι , Κ (Ρ -1 )K <N< ΚΡΚ ( Κ,Ρ> 2 ) 式子(7) , 其中 Ν为全连通拓朴中节点的个数,即实际网络中边界节点的个数。
接着, 在步骤 520, 利用根据式 (7) 得到的参数 和尸, 初始 化双向洗牌网拓朴, 以得到初始化的双向洗牌网拓朴:
G' =(V, Ε') 式子 (8) 其中, 是双向洗牌网拓朴中的节点集合,并且 ={ν ,ν'2 ,..., }; E,是双向洗牌网拓朴中的链路的集合, 并且 £'=«, ,..., ,},
Figure imgf000009_0001
继续参考图 3, 在步骤 320, 通过某种映射关系, 将全连通拓朴 中的节点映射到 shuffle-net中的节点,并对 shuffle-net中的双向链路 赋予相应的逻辑链路属性 (以下有描述) 。
图 6 示出了根据本发明的一个实施例的对欢向洗牌网拓朴进行 优化的流程图。 在该实施例中, 利用遗传算法对双向洗牌网拓朴进 行优化。
如图所示, 在步骤 610, 首先得到双向洗牌网拓朴的第一代 V 它包括若千条染色体 p,染色体 p可以表示全连通拓朴节点在双向洗 牌网拓朴中的排列方式。 在一个实施例中, 利用一维文字排列编码 来构造所述染色体。 例如, 可以通过随机排列 N'个节点来得到第一 代 ^的染色体 p, 即 /?=random permutation (1, 2, .·., Λ^') p ^ Po 式子 ( 9) 以一节点数 =7的全连通拓朴为例, 通过前面的初始化操作, 得到 一个 P=2, N, =8 的双向 shuffle-net拓朴。 对于一个随机生成 的染色体 " 7 2 3 1 5 6 4 8" , 其表示全连通拓朴各个节点在双向 shuffle-net 中的排列顺序 (其中节点 8 为补足数目而填充的虚拟节 点) , 即全连通拓朴中的节点 7对应双向 shuffle-net中的(0, 0)节 点, 2对应(1, 0)节点, ..., 4对应(2, 1)节点, 虛拟节点 8位于(3, 1) (如图 7所示) 。
可以看出本发明中的染色体与通常由 0、 1 构成的二进制表示 的染色体不同。 本发明利用一维文字排列编码来构造染色体, 其中 的每个基因是节点编号。 因此对本发明的染色体进行变异操作是基 于位置的操作, 通常使染色体中基因的位置发生变化, 因此可以保 证下一代能够继承上一代的多数特征, 并保留相容的基因。
接着在步骤 620, 根据每条染色体 p将全连通拓朴的所有节点 β映射为双向洗牌网拓朴 G'的节点 V'后,就可以根据节点位置进行 逻辑链路映射。 如在上述双向 shuffle-net拓朴中, 节点 7与节点 4、 5、 6相连, 则将全连通拓朴中的双向逻辑链路 (4, 7), (5, 7), (6, 7)保留,其余与节点 7相连的逻辑链路全部删除。节点 8为虛拟节点, 不参与逻辑链路映射。 为了评价每条染色体的优劣, 我们获得其对 应的双向洗牌网拓朴 G'的连通性矩阵 C':
C=[c'(i, , k)]NxNxW 式子 ( 10) 可以像上面描述的获得连通性矩阵 C那样来获得连通性矩阵 C,。 接着在步骤 630,以真实拓朴边界节点间的连通性矩阵为目标函 数, 得到每个染色体的适应度值:
W N
Fitness=- ^∑ deviation (ζ·, j k) 式子
k=\ ij=]
( 11 )
其中: deviatiorijj ,λ. ) 时
式子 ( 12 )
Figure imgf000011_0001
在式子 ( 1 1 ) 中, 适应度值表示全连通拓朴的连通性矩阵与所 述染色体对应拓朴的连通性矩阵之间的相近程度, 其值为偏差度值 总和的相反数。 负号表示偏差度值越小, 适应度值越大; 所述偏差 度值为目标连通性矩阵与染色体对应拓朴的连通性矩阵之间的差值 和目标连通性矩阵与染色体对应拓朴连通性矩阵之和的比值的绝对 值。 在替代的方案中, 所述偏差度值为目标连通性矩阵与染色体对 应拓朴连通性矩阵之间的差值的绝对值。
接着在步骤 640,根据得到的适应度值进行遗传操作以得到下一 代染色体。 在一个实施例中, 首先, 根据得到的适应度值, 对这一 代的染色体进行选择操作, 以得到用于繁殖下一代的染色体。 也就 是说, 根据得到的适应度值, 将适应度低 (例如低于一个阈值或排 序低于某比例) 的染色体淘汰掉, 只选择适应度高的那些染色体到 下一代中。 接着对所选择的染色体进行杂交和变异等操作, 从而得 到下一代的染色体。 图 8 示出了根据本发明的一个实施例的杂交和 变异操作的实例。 应当理解, 在本发明中, 并不局限于图 8 所示的 杂交和变异操作。
然后在步骤 650, 判断遗传的代数是否已经达到预定阈值。 如果 已经到达预定阈值, 则在步骤 660将得到的适应度最好的染色体作 为优化结果, 也就是说, 得到优化的双向洗牌网拓朴。 如果还没有 到达预定阈值, 则跳转到步骤 620继续进行迭代。
在另一实施例中, 在达到预定遗传代数的阈值之前, 如果一个 染色体的适应度达到了某个预定阈值, 则停止遗传操作, 并将根据 得到的染色体排列和连接的双向洗牌网拓朴作为结果拓朴。
可以根据需要来确定遗传的代数。 遗传的代数越多, 得到的结 果越好, 但是, 需要更多的时间来计算。
发明人已经验证, 在遗传的最初阶段, 每代的平均偏差下降很 快, 并且经过 20代的遗传后, 平均偏差已经趋于稳定。 其后平均偏 差在特定的范围内波动。 在 30代以后平均偏差已经基本小于 20 %, 而其中的最优个体偏差已经基于维持在 10 %附近。 由此可见, 本发 明的拓朴抽象方法, 可以得到较为准确的拓朴, 可以为路由决策提 供更加准确和丰富的信息, 因此可以增加性能。
此外, 对于双向洗牌网拓朴的优化也可以使用本领域公知的其 他适当的算法, 诸如启发式算法等。
通过结合图 3-8给出的如上描述, 可以看出, 本发明的拓朴抽象 方法提供了适合光网络的详细的链路信息。 更具体地, 通过利用遗 传算法优化双向洗牌网拓朴, 用较少的逻辑链路数描述了较准确的 拓朴结构, 从而提高了性能。
在另一替代的实施例中, 利用得到的全连通拓朴的连通性矩阵 , 将全连通拓朴中具有较多链路分离路径数的节点对映射到欢向洗牌 网中直接连接的节点对, 利用启发式方法将全连通拓朴压缩成双向 洗牌网拓朴。
在又一替代的实施例中, 利用得到的全连通拓朴的连通性矩阵, 将全连通拓朴压缩成对称星型网络。
在网络状态发生改变时, 诸如链路的带宽可用性发生改变时, 需要进行重抽象操作来更新广播出去的抽象拓朴。
在本发明的一个实施例中, 基于预定时间间隔来执行重抽象。 在该实施例中, 预先定义了一个重抽象时间间隔, 基于该时间间隔 来周期性地执行重抽象, 而不考虑域内链路带宽的改变情况。
在本发明的另一个实施例中, 采用了一种基于事件的重抽象策 略, 即仅在发生预定事件时执行重抽象。 优选地, 可以采用迟钝重 抽象策略, 即仅当发生较大拓朴改变时或连接拒绝率达到预定阈值 时, 才执行重抽象。
在一个实施例中, 在达到预定连接拒绝率时, 执行重抽象。 例 如在预定的重抽象检测时间间隔内, 当连接拒绝率 r超过预定阔值 a 时立即执行重抽象, 其中预定阈值 α可以是基于客户 QoS需求的常 数, 例如, 可以是 20%。
在另一个实施例中, 在拓朴发生较大变化时, 执行重抽象。 在 该实施例中假设链路的容量为 个通道, 时刻的可用通道数为 ct。 当时刻 的可用通道数 等于预定值序列 中的元素值时, 执行重 抽象。
在一个实施例中, 如下构造预定值序列 ^: bk={ , La},
即当可用通道数 变成 0时或者变成 a时, 进行重抽象。
而在另一个实施例中, 预定值序列 ^具有更多元素,以使得在可 用通道数较多时, 拓朴再抽象频率较低, 而在可用通道数较少时拓 朴再抽象频率较高。 例如, 可以通过下式得到预定值序列 :
0 k = 0
La k = \,2,..., La 式子 ( 13 )
La + La k-L。 k = La + l,〜K
其中 是使在 。〉l时, 8且 。= 1时 A的最大下标。 例如, 对于每条链路 16 个通道的情况, 在 。 =2 时, 该预定值序列 bk = {0,2,4,6,10, 16} ; 在 La=3 时, bk = {0,3,6, 12,16} ; 在 La=4 时, = {0,4,8,16}。 上述式子仅仅是个例子, 也可以利用其他方式来得到预 定值序列 。
根据本发明的重抽象策略, 仅在出现较大的拓朴改变时, 才执 行重抽象, 而在拓朴改变较小时不执行重抽象。 因此, 在保证抽象 拓朴得到及时更新的前提下, 减少了重抽象的次数, 从而减少因执 行重抽象而消耗的资源, 提高了性能。
从上面的描述可以看出, 本发明的方法利用 "链路分离路径数" 来构造连通性矩阵, 不仅提供了可用波长的可用性信息, 而且提供 该波长相关的资源丰富度信息, 因此为路由决策提供了更详细的链 路信息来优化资源分配进而改善网络性能。 在一个优选实施例中, 进一步对得到的全连通拓朴进行压缩, 进而减少域间交换的信息量, 进一步提高性能。 在一个优选实施例中, 将全连通拓朴压缩成双向 洗牌网拓朴, 大大减少了链路数量, 进而减少了域间需要交换的路 由信息量并提高了性能。 在另一个优选实施例中, 釆用基于事件的 迟钝重抽象策略来执行重抽象, 从而减少因执行重抽象而消耗的资 源, 进一步提高了性能。
接下来, 将在下文中对本发明的拓朴抽象装置进行进一步的描 述。
图 9示出了根据本发明的一个实施例的拓朴抽象装置 900。 如图 9所示,所述拓朴抽象装置 900包括:链路分离路径数获取装置 910, 配置用于得到真实拓朴中每对边界节点之间的链路分离路径数 c(i, j, k ) ; 拓朴构造装置 920, 配置用于利用所述链路分离路径数 c( , ,A)获得连通性矩阵 C并构造相应全连通拓朴 GB
所述链路分离路径数获取装置 910可以根据上述对步骤 210-230 的描述来执行操作: 首先根据域的真实拓朴计算出邻接矩阵, 然后 通过计算边界节点之间的最大流来得到边界节点之间的链路分离路 径数。 在计算链路分离路径数时, 可以利用最高标号预流推进算法、 增广路算法或本领域公知的其他最大流算法。
所述拓朴构造装置 920可以根据上述对步骤 240的描述来执行操 作: 利用所述链路分离路径数获取装置 910得到的链路分离路径数 来构造连通性矩阵。
从上面的实施例可以看出,本发明的拓朴抽象装置 900通过使用 链路分离路径数来构造连通性矩阵, 不仅提供了可用波长信息, 而 且提供了与该波长丰富程度相关的资源信息, 因此可以有助于路由 选择, 使得性能得到提高。
在一个优选实施例中, 所述拓朴抽象装置 900进一步包括: 拓朴 压缩装置 930, 配置成用于对所述全连通拓朴进行压缩, 从而进一步 减少域间交换的信息量。
如图 10 所示, 在一个实施例中, 所述拓朴压缩装置 930 包括: 双向洗牌网拓朴构造装置 932,配置用于根据确定的双向洗牌网拓朴 的结构参数, 来初始化所述双向洗牌网拓朴; 和双向洗牌网拓朴优 化装置 934,配置用于利用所述全连通拓朴的连通性矩阵对所述双向 洗牌网拓朴进行优化。
所述双向洗牌网拓朴优化装置 934 可以进一步基于遗传算法来 对双向洗牌网拓朴进行优化。 图 1 1示出了基于遗传算法的双向洗牌 网拓朴优化装置 934 的一个实施例。 双向洗牌网拓朴优化装置 934 包括: 第一代生成装置 9342, 配置用于利用随机方式得到双向洗牌 网拓朴的第一代; 连通性矩阵获取装置 9344, 配置用于得到当前代 中的染色体的连通性矩阵; 适应度值计算装置 9346, 配置用于以真 实拓朴边界节点间的连通性矩阵为目标来计算当前代所述染色体的 适应度值; 以及遗传操作装置 9348, 配置用于根据适应度值对所述 染色体执行遗传操作, 以得到下一代的染色体, 其中遗传操作包括 选择操作、 变异操作、 杂交操作等; 映射装置 9350, 配置用于在满 足终止遗传操作的条件时利用得到的最优结果染色体将全连通拓朴 中的节点和逻辑链路映射到双向洗牌网拓朴中。 此外, 双向洗牌网 拓朴的优化也可以使用本领域公知的其他适当的算法, 诸如启发式 算法等。
在一个实施例中, 适应度值计算装置 9346计算的适应度值是所 述当前代染色体的连通性矩阵与所述真实拓朴边界节点间的连通性 矩阵之间差值的绝对值; 而在另一实施例中, 所述适应度值是所述 当前代染色体的连通性矩阵与所述真实拓朴边界节点间的连通性矩 阵之间的差值和所述当前代染色体的连通性矩阵与所述真实拓朴边 界节点间的连通性矩阵之和的比值的绝对值。
在一个实施例中,所述终止遗传操作的条件是遗传的代数超过预 定值或适应度值达到预定值。
在一个实施例中, 所述染色体基于一维文字排列编码。
从上面描述的实施例可以看出,通过进一步利用遗传算法优化双 向洗牌网拓朴, 显著降低了逻辑链路数并提供了更加准确的拓朴结 构, 从而提高了性能。
在一个优选的实施例中, 如图 12所示, 所述拓朴抽象装置 900 进一步包括重抽象触发装置 940,配置用于在网络状态发生改变时触 发重抽象或基于预定时间间隔来触发重抽象。 在一个实施例中, 所 述重抽象触发装置 940基于迟钝重抽象策略, 在发生预定拓朴改变 事件或连接拒绝率达到预定阈值时触发重抽象。 所述预定拓朴改变 事件可以是可用资源增加或减少至预定值的事件, 且所述重抽象在 可用资源较多时发生频率较低, 在可用资源较少时发生频率较高。
此外, 本发明还提供了一种路由控制器。 图 13示出了根据本发 明的一个实施例的路由控制器的示意方框图。 如图 13所示, 所述路 由控制器 1300包括: 真实拓朴获得装置 1310, 用于获得一个域的真 实拓朴; 根据本发明的拓朴抽象装置 900, 用于对一个域的真实拓朴 进行拓朴抽象; 链路状态发送装置 1320, 用于基于所述拓朴抽象装 置 900 得到的拓朴生成链路状态广播 ( Link-State Advertisement, LSA )信息并把该 LSA信息发送出去。 其中所述拓朴抽象装置 900 利用链路分离路径数来构造连通性矩阵, 不仅提供了可用波长信息, 而且提供了与该波长相关的资源信息。 因此可以有助于路由选择, 使得性能得到提高。
图 14示出了本发明的路由控制器在开始阶段的处理的流程图。 如图 14所示, 路由控制器开始操作后, 首先在步骤 1410, 真实拓朴 获得装置获得一个域的真实拓朴。 然后在步骤 1420, 拓朴抽象装置 对所述真实拓朴进行拓朴抽象。 接着在步骤 1430, 链路状态发送装 置基于抽象的拓朴生成 LSA并将其发送出去。
在工作阶段, 拓朴 象装置在满足重抽象条件时, 触发重抽象。 可以在网络状态发生改变时触发重抽象或者基于预定时间间隔来触 发重抽象。 在一个实施例中, 基于迟钝重抽象策略, 在发生预定拓 朴改变事件或连接拒绝率达到预定阈值时触发重抽象。 并且, 所述 预定拓朴改变事件可以是可用资源增加或减少至预定值的事件, 且 所述预定拓朴改变事件在可用资源较多时发生频率较低, 可用资源 较少时发生频率较高。
图 15示出了根据本发明一个实施例的路由控制器在工作阶段时 的处理的流程图。
如图 15所示, 在工作阶段, 当接收到请求消息时, 首先在步骤 1501判断该请求是否是连接建立请求还是连接拆除请求。 如果是连 接建立请求则过程进行至步骤 1502。
在步骤 1502, 检验指定入口节点和出口节点之间的资源, 若检 验表明在指定入口节点和出口节点之间没有足够资源来支持该连接 请求, 流程则进行至步骤 1504。在该步骤 1504, 发送消息给请求者, 以通知请求者连接失败。 然后在步骤 1505增加连接失败次数, 并计 算连接拒绝率, 即连接失败次数除以连接成功次数与连接失败次数 之和。
接着, 在步骤 1506判断拒绝率是否达到预定的拒绝率阈值。 如 果达到预定阈值, 则在步骤 1513触发重抽象, 以进行拓朴重抽象。 随后在步骤 1514将连接成功、 失败计数清零, 以便重新进行计数。 接着在步骤 1515, 基于重抽象的拓朴生成 LSA, 将该 LSA 发送出 去。 至此对于一个请求的处理结束。
如果在步骤 1502检验表明在指定入口节点和出口节点之间有足 够资源来支持该连接请求时,流程则进行至步骤 1507。在步骤 1507, 将该请求发送到下一个域, 并等待连接建立指示。 等到指示后在步 骤 1508在沿着域内的选定链路分配波长, 以建立连接。 随后在步骤 1509增加连接成功计数, 使连接成功次数加 1。 然后在步骤 1510, 触发再次泛洪, 以发送 LSA信息。 接着在步骤 151 1 , 检查入口节点 和出口节点之间的剩余资源, 并在步骤 1512判断剩余资源是否达到 重抽象的阈值 。。 如果达到阈值 , 流程则进入到步骤 1513, 执行 重抽象。 随后在步骤 1514使连接成功、 失败计数清零。 然后流程进 行至步骤 S1515 , 结束该请求的处理过程。 如果尚未到达阔值 La, 则直接结束该请求的处理过程。
在步骤 1501 , 如果请求是连接拆除请求, 流程则进行至步骤 1503。 在该步骤, 执行连接的拆除。 随后流程进行至步骤 1510, 执 行如前所述的后面的步骤直至结束。
应当注意, 为了使本发明更容易理解, 上面的描述省略了对于本 领域的技术人员来说是公知的、 并且对于本发明的实现可能是必需 的更具体的一些技术细节。
提供本发明的说明书的目的是为了进行说明和描述,而不是用来 穷举或将本发明限制为所公开的形式。 对本领域的普通技术人员而 言, 许多修改和变更都是显而易见的。
因此,选择并描述实施方式是为了更好地解释本发明的原理及其 实际应用, 并使本领域普通技术人员明白, 在不脱离本发明实质的 前提下, 斤 格
范围之内

Claims

权 利 要 求
1 . 一种拓朴抽象方法, 包括:
得到真实拓朴 GR 中每对边界节点之间的链路分离路径数 和
利用所述链路分离路径数 c(i, , y获得连通性矩阵 C并构造相应 全连通拓朴 GB
2. 根据权利要求 1 所述的方法, 其中通过计算边界节点之间的 最大流来得到链路分离路径数。
3. 根据权利要求 2所述的方法, 其中计算边界节点之间的最大 流通过最高标号预流推进算法来进行。
4. 根据权利要求 1所述的方法, 进一步包括:
对所述全连通拓朴进行拓朴压缩。
5. 根据权利要求 4所述的方法, 其中所述拓朴压缩包括: 确定一个双向洗牌网拓朴的结构参数;
利用所述结构参数初始化所述双向洗牌网拓朴; 和
利用所述连通性矩阵对所述双向洗牌网拓朴进行优化。
6. 根据权利要求 5所述的方法, 其中对所述双向洗牌网拓朴进 行优化基于遗传算法, 并包括:
利用随机方式得到双向洗牌网拓朴的第一代:
得到当前代中的染色体对应拓朴的连通性矩阵;
以真实拓朴边界节点间的连通性矩阵为目标来计算当前代染色 体的适应度值;
根据适应度值对当前代染色体执行遗传操作,以得到下一代的染 色体;
重复得到染色体对应拓朴的连通性矩阵、计算适应度值、执行遗 传操作直至满足终止遗传的条件; 和
利用得到的结果染色体将全连通拓朴中的节点映射到双向洗牌 网拓朴中, 以得到优化的双向洗牌网拓朴。
7. 根据权利要求 6所述的方法, 其中所述适应度值是偏差度值 总和的相反数, 所述偏差度值为所述全连通拓朴的连通性矩阵与染 色体对应拓朴的连通性矩阵之间的差值和所述全连通拓朴的连通性
8 根据权利要求 6所述的方法,其中终止遗传的条件是遗传的代 数达到预定值。
9. 根据权利要求 6所述的方法, 其中所述染色体采用了一维文 字排列编码。
10. 根据权利要求 1所述的方法, 进一步包括:
在网络状态发生改变时执行拓朴的重抽象。
1 1. 根据权利要求 10所述的方法, 其中基于迟钝重抽象策略, 在发生预定拓朴改变事件或连接拒绝率达到预定阈值时, 执行拓朴 的重抽象。
12. 根据权利要求 1 1 所述的方法, 其中所述预定拓朴改变事件 是可用资源变成预定值的事件, 且所述预定拓朴改变事件在可用资 源较多时发生频率较低, 在可用资源较少时发生频率较高。
13. 一种拓朴抽象装置, 包括:
链路分离路径数获取装置, 配置用于得到真实拓朴 中每对边 界节点之间的链路分离路径数 cO A) ; 和
拓朴构造装置,配置用于利用所述链路分离路径数 Φ·,_/·Α)获得连 通性矩阵 C并构造相应全连通拓朴 GB
14. 根据权利要求 13 所述的拓朴抽象装置, 其中所述链路分离 路径数获取装置进一步配置成通过计算所述边界节点之间的最大流 来得到链路分离路径数。
15. 根据权利要求 14所述的拓朴抽象装置, 其中计算边界节点 之间的最大流通过最高标号预流推进算法来进行。
16. 根据权利要求 13所述的拓朴抽象装置, 进一步包括: 拓朴压缩装置, 配置成用于对所述全连通拓朴进行拓朴压缩。
17. 根据权利要求 16所述的拓朴抽象装置, 其中所述拓朴压缩 装置包括:
双向洗牌网拓朴构造装置,配置用于根据确定的双向洗牌网拓朴 的结构参数, 来初始化所述双向洗牌网拓朴; 和
双向洗牌网拓朴优化装置,配置用于利用所述真实拓朴边界节点 间的连通性矩阵为目标对所述欢向洗牌网拓朴进行优化。
1 8. 根据权利要求 1 7所述的拓朴抽象装置, 其中所述双向洗牌 网拓朴优装置基于遗传算法, 并包括:
第一代生成装置,配置用于利用随机方式得到双向洗牌网拓朴的 第一代;
连通性矩阵获取装置,配置用于得到当前代中的染色体的对应拓 朴的连通性矩阵;
适应度值计算装置,配置用于以真实拓朴边界节点间的连通性矩 阵为目标来计算当前代染色体的适应度值; 以及
遗传操作装置,配置用于根据适应度值对当前代染色体执行遗传 操作, 以得到下一代的染色体; 和 '
映射装置,配置用于在满足终止遗传的条件时利用得到的结果染 色体将全连通拓朴的节点映射到双向洗牌网拓朴中。
19. 根据权利要求 18所述的拓朴抽象装置, 其中所述适应度值 是偏差度值总和的相反数, 所述偏差度值为所述全连通拓朴的连通 性矩阵与染色体对应拓朴的连通性矩阵之间的差值和所述全连通拓 值。、 ' 、 一 、 ' ' 、
20. 根据权利要求 18所述的拓朴抽象装置, 其中终止遗传的条 件是遗传的代数超过预定值。
21. 根据权利要求 18 所述的拓朴抽象装置, 其中所述染色体基 于一维文字排列编码。
22. 根据权利要求 13所述的拓朴抽象装置, 进一步包括: 重抽象触发装置, 配置用于在网络状态发生改变时触发重抽象。
23. 根据权利要求 22所述的拓朴抽象装置, 其中所述重抽象触 发装置基于迟钝重抽象策略, 在发生预定拓朴改变事件或连接拒绝 率达到预定阈值时触发重抽象。
24. 根据权利要求 23 所述的拓朴抽象装置, 其中所述预定拓朴 改变事件是可用资源增加或减少至预定值的事件, 且所述预定拓朴 改变事件在可用资源较多时发生频率较低, 在可用资源较少时发生 频率较高。
25. 一种路由控制器, 包括:
根据权利要求 13至 24任一项所述的拓朴抽象装置, 以及 链路状态发送装置,配置成用于基于所述拓朴抽象装置得到的拓 朴生成链路状态广播 LSA信息, 并将所述 LSA信息发送出去。
PCT/CN2008/000741 2008-04-10 2008-04-10 拓扑抽象方法、拓扑抽象装置以及路由控制器 WO2009124419A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
US12/735,732 US8406154B2 (en) 2008-04-10 2008-04-10 Method and apparatus for topology aggregation and routing controller
JP2011503322A JP2011517220A (ja) 2008-04-10 2008-04-10 トポロジ抽出方法、トポロジ抽出装置、およびルートコントローラ
EP08733944A EP2271028A1 (en) 2008-04-10 2008-04-10 Topology abstraction method, topology abstraction apparatus and route controller
PCT/CN2008/000741 WO2009124419A1 (zh) 2008-04-10 2008-04-10 拓扑抽象方法、拓扑抽象装置以及路由控制器
KR1020107025146A KR20100133003A (ko) 2008-04-10 2008-04-10 토폴로지 축약 방법, 토폴로지 축약 장치 및 라우트 제어기

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2008/000741 WO2009124419A1 (zh) 2008-04-10 2008-04-10 拓扑抽象方法、拓扑抽象装置以及路由控制器

Publications (1)

Publication Number Publication Date
WO2009124419A1 true WO2009124419A1 (zh) 2009-10-15

Family

ID=41161512

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2008/000741 WO2009124419A1 (zh) 2008-04-10 2008-04-10 拓扑抽象方法、拓扑抽象装置以及路由控制器

Country Status (5)

Country Link
US (1) US8406154B2 (zh)
EP (1) EP2271028A1 (zh)
JP (1) JP2011517220A (zh)
KR (1) KR20100133003A (zh)
WO (1) WO2009124419A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10108616B2 (en) * 2009-07-17 2018-10-23 International Business Machines Corporation Probabilistic link strength reduction
US9467348B2 (en) * 2014-01-31 2016-10-11 Viasat, Inc. Systems, methods, and devices for reducing overhead messaging in networks
JP6492977B2 (ja) * 2015-06-01 2019-04-03 富士通株式会社 並列演算装置、並列演算システム、ノード割当プログラム及びノード割当方法
US10038623B2 (en) * 2016-10-24 2018-07-31 Microsoft Technology Licensing, Llc Reducing flooding of link state changes in networks
US11184245B2 (en) 2020-03-06 2021-11-23 International Business Machines Corporation Configuring computing nodes in a three-dimensional mesh topology
CN113572690B (zh) * 2021-06-11 2023-02-24 深圳市国电科技通信有限公司 面向可靠性的用电信息采集业务的数据传输方法
CN114448863B (zh) * 2022-01-06 2022-11-22 武汉烽火技术服务有限公司 一种寻找跨域路径的计算方法和装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030193898A1 (en) * 2002-04-15 2003-10-16 Wong Vincent Chi Chiu Method and apparatus for selecting maximally disjoint shortest paths in a network
CN1787470A (zh) * 2005-11-25 2006-06-14 北京邮电大学 一种应用于不对称网络中的生成树拓扑抽象方法
CN1816000A (zh) * 2005-02-02 2006-08-09 华为技术有限公司 一次路由计算实现层次路由的拓扑方法
GB2440287A (en) * 2005-09-27 2008-01-23 Roke Manor Research Disjoint pair formation in a network using shortest path determination

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH1098477A (ja) * 1996-09-25 1998-04-14 Toshiba Corp 経路選択方法および通信システム
US7813270B2 (en) * 2003-05-15 2010-10-12 Alcatel-Lucent Usa Inc. Route precomputation method and apparatus for bandwidth guaranteed traffic
US7463579B2 (en) * 2003-07-11 2008-12-09 Nortel Networks Limited Routed split multilink trunking
US7639688B2 (en) * 2005-07-18 2009-12-29 Cisco Technology, Inc. Automatic protection of an SP infrastructure against exterior traffic
JP4494357B2 (ja) * 2006-03-08 2010-06-30 富士通株式会社 パス経路計算方法及び,この方法を適用する光通信ネットワーク
US20070300239A1 (en) * 2006-06-23 2007-12-27 International Business Machines Corporation Dynamic application instance placement in data center environments
WO2008055539A1 (en) * 2006-11-06 2008-05-15 Telefonaktiebolaget Lm Ericsson (Publ) Multi-domain network and method for multi-domain network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030193898A1 (en) * 2002-04-15 2003-10-16 Wong Vincent Chi Chiu Method and apparatus for selecting maximally disjoint shortest paths in a network
CN1816000A (zh) * 2005-02-02 2006-08-09 华为技术有限公司 一次路由计算实现层次路由的拓扑方法
GB2440287A (en) * 2005-09-27 2008-01-23 Roke Manor Research Disjoint pair formation in a network using shortest path determination
CN1787470A (zh) * 2005-11-25 2006-06-14 北京邮电大学 一种应用于不对称网络中的生成树拓扑抽象方法

Also Published As

Publication number Publication date
EP2271028A1 (en) 2011-01-05
US8406154B2 (en) 2013-03-26
JP2011517220A (ja) 2011-05-26
KR20100133003A (ko) 2010-12-20
US20110080851A1 (en) 2011-04-07

Similar Documents

Publication Publication Date Title
WO2009124419A1 (zh) 拓扑抽象方法、拓扑抽象装置以及路由控制器
CN110351286B (zh) 一种软件定义网络中链路洪泛攻击检测响应机制
CN103634842B (zh) 一种分布式卫星网络群间路由方法
CN103001875A (zh) 一种量子密码网络动态路由方法
CN103139069B (zh) 基于层次分析法的多度量参数的通信网路由方法
KR100950423B1 (ko) 다개체 유전자 알고리즘을 이용한 라우팅 경로 검색 방법 및 그에 따른 센서 네트워크 시스템
CN101557300B (zh) 一种网络中的拓扑重构方法、装置及设备
CN109067758A (zh) 一种基于多路径的sdn网络数据传输隐私保护系统及其方法
JP2000134199A (ja) マルチキャスト樹を蓄積するための効果的な手段
CN101729385A (zh) 一种路径计算及建立方法、装置和系统
CN113794638B (zh) 基于差分进化算法的sdn数据中心网络大象流调度方法
CN101588287A (zh) 对等网络数据调度和下载的方法、装置和系统
CN110661704B (zh) 转发路径的计算方法及sdn控制器
CN105472484A (zh) 一种电力骨干光传输网波道均衡路由波长分配方法
Kuhn et al. The complexity of data aggregation in directed networks
Bai et al. Effective hybrid link-adding strategy to enhance network transport efficiency for scale-free networks
Lin et al. Priority-based genetic algorithm for shortest path routing problem in OSPF
CN106792971A (zh) 基于蚁群算法的网络节点选择方法
CN115130044B (zh) 一种基于二阶h指数的影响力节点识别方法和系统
Singh et al. Crossover behavior in a communication network
Moza et al. Finding K shortest paths in a network using genetic algorithm
Cavendish et al. On the construction of low cost multicast trees with bandwith reservation
CN114697002A (zh) 一种分布式量子密码网络组密钥分发方法及系统
CN101102616A (zh) 自动交换光网络中多约束条件下最短路径查找方法及装置
CN113572690B (zh) 面向可靠性的用电信息采集业务的数据传输方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 08733944

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 12735732

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 2011503322

Country of ref document: JP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 20107025146

Country of ref document: KR

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2008733944

Country of ref document: EP