CN102035672B - Distributed network measurement method based on hierarchical coordination multi-agent system - Google Patents

Abstract

The invention relates to a distributed network measurement method based on a hierarchical coordination multi-agent system. By adopting the method, the distributed network measurement can be realized by a cooperative multi-agent system consisting of a plurality of measurement nodes. The measurement nodes are divided into different logical levels according to the network position per se; the measurement nodes form clusters according to the physical topology, wherein the measurement nodes in the clusters are adjacent in the physical topology and each cluster is provided with a cluster head; the nodes in the cluster are only used for measuring chain information of other measurement nodes which are positioned at the same cluster with the measurement nodes; the cluster heads can be respectively positioned in two clusters; the roles in the bottom layer cluster are used as the cluster heads; and the roles in the upper layer cluster are used as general nodes in the cluster. Therefore, the real-time chain state information of two adjacent layers of network can be acquired by implementing the measurement on the nodes of the cluster nodes. Meanwhile, the network measurement data of two adjacent layers can be relayed by using the nodes of the cluster heads as a bridge for information interaction of measurement nodes in two adjacent nodes, thereby ensuring that the measurement nodes in the layers can get the state information of the network outside the layers.

Description

A kind of distributed network measurement method based on stratification cooperation multiagent system

Technical field

The present invention relates to field of computer network administration, particularly a kind of distributed network measurement method based on stratification cooperation multiagent system.

Background technology

The network measure model generally is divided into centralized measurement model and distributed measurement model.

Traditional network measuring system is mostly based on the centralized measurement model.Fig. 1 is traditional centralized network measurement model figure.As shown in Figure 1, measuring system generally comprises central server, measured node, central database and Analysis server.Central server is responsible for whole network measure process is managed, and comprises initiatively measuring and passive measurement.Measured node is responsible for carrying out concrete test assignment, and the data of collecting are sent to central database carries out statistics and analysis.This metering system has adopted typical C/S structure.

The flow process that the tradition centralized network is measured is as follows:

(1) measured node is initiated request to server, and after the authentication by server, server will obtain corresponding test assignment from tactful configuration database, and will send to measured node;

(2) measured node is implemented network measure according to the configuration parameter that obtains;

(3) measure complete after, whether measured node sends to Analysis server with the measurement data of obtaining, and select to sleep according to local collocation strategy, or directly begins next time test loop;

(4) Analysis server measurement data that measured node is uploaded is carried out statistics and analysis, and analysis result is stored in the central database, and in general, Analysis server and central database are on the same physical server.

(5) central database externally provides query interface, and in many ways query function is provided.

Distributed network measurement is independently finished corresponding surveying work by at tested Nodes the intelligent measure program being installed by program.There is not at present unified distributed network measurement structure.After measured node was finished test assignment, measurement result can be stored in this locality and carry out analyzing and processing, also measurement result can be sent to centralized Analysis server and process.Distributed network measurement has been avoided all surveying works all are placed on the drawback that causes server load overweight on the server, alleviates the burden of server.When network topology structure was expanded, the distributed network measurement system only need to be at newly-increased node deploy process of measurement, and increases corresponding configuration information, can adapt to new network environment.

Along with the appearance of P2P (Peer to Peer, point-to-point) technology, the distributed network measurement system presents P2Pization gradually.The P2P network has following characteristics:

(1) the permission node dynamically adds and withdraws from, and the impact that in this process system is caused is very little, has fabulous extensibility;

(2) the P2P network all is to set up in the mode of self-organizing usually, and when part of nodes lost efficacy, network can be adjusted whole topology automatically, keeps other internodal connectednesses, has anti-attack, high fault-tolerant advantage;

(3) the P2P system has good load balancing characteristics.

With the distributed network measurement system of the distributed Hash table of employing DHT (Distributed Hash Table) technology as networking plan, take Chord as example, carry out the explanation of distributed network measurement scheme.Fig. 2 is the distributed network measurement system diagram based on Chord.As shown in Figure 2, comprise 6 measured node in the distributed network measurement system, measured node makes up the DHT network by Chord networking algorithm.The measurement procedure of distributed network with measured node 1 for for example lower:

(1) measured node 1 operation DHT networking agreement adds in the distributed measurement network;

(2) measured node 1 is inquired about other measured node in the Chord ring by the XM vlan query protocol VLAN;

(3) measured node 1 is implemented to measure to other tested nodes, obtains measured node 1 to the link performance data of other tested nodes;

(4) analysis and the processing of data carried out in measured node 1 this locality, and with result store in local data base.

Inquiry can encircle interior nodes from Chord, also can encircle exterior node from Chord, such as monitoring server.When measured node receives inquiry, will retrieve local data base, and return result for retrieval.

The centralized network measurement model adopts the C/S structure, the disadvantage of this model is that the task of each measured node is unified by central server initiation and management in the system, cause easily the overweight bottleneck that becomes system of task load of central server, and can not directly carry out data communication between measured node, cause the inefficiency of whole system, and be difficult to adapt to large-scale network environment.Equally, for Analysis server and the central database concentrated, exist too load heavy, inefficient problem.

Although distributed network measurement has solved the bottleneck problem of central server, also there is certain limitation.At first, the test target of distributed network measurement is the end-to-end link performance between measured node often, because the cross-domain many sub-links of end-to-end link, and the state of sublink is real-time change, therefore, the end-to-end measurement result can accurately not reflect the real-time link state; Secondly, end-to-end link is measured the performance data that can't obtain each sub-links, so measurement result can't be located the bottleneck zone of end-to-end link; At last, because end-to-end link comprises many sub-links, therefore many existence that end-to-end link has overlapping sublink, thus when implementing end-to-end measurement, overlapping sublink is repeated to measure, the problem such as caused that network burden increases the weight of.

Summary of the invention

The object of the invention is to, a kind of distributed network measurement method based on stratification cooperation multiagent system is provided.To use hierarchical structure to realize distributed network measurement, eliminate the bottleneck problem that traditional centralized measurement exists, can press close to the actual physical network topology simultaneously, adapt to large-scale network environment.

For achieving the above object, provide a kind of distributed network measurement method based on stratification cooperation multiagent system, the method adopts the cooperation multiagent system that is comprised of a plurality of measured node to realize distributed network measurement; Described measured node is client computer; The concrete steps of the method comprise:

Step 1): at least 1 described measured node forms bunch according to physical topology according to own residing network site, the ground floor that bunch forms described stratification cooperation multiagent system of acquisition;

Step 2): each of ground floor bunch selected a cluster head as the measured node of last layer, and this measured node forms bunch according to physical topology according to own residing network site again, the second layer that bunch forms described stratification cooperation multiagent system of acquisition; And each of the second layer bunch selected a cluster head again as the measured node of last layer, by that analogy, until comprise the top layer of physical network topology of living in;

Step 3): the measured node in every cluster is only measured and other of one bunch of oneself existing together comprise network performance parameter between the measured node of cluster head, and network performance parameter is transferred to the cluster head of this bunch; The cluster head of every one deck is as the measured node of last layer, also only measure and one bunch other measured node of oneself existing together between network performance parameter, and network performance parameter is transferred to the cluster head of this bunch, the cluster head node can obtain the real-time link state information of adjacent two layers network by implementing measurement.

The method also comprises step 4); Described step 4): according to described step 3 the end to end real-time network performance parameter that obtains) is oriented the bottleneck zone of end-to-end link.

Described network performance parameter comprises: bandwidth, time delay and packet loss.

Described step 4) according to formula (1), formula (2) and formula (3) computation and measurement node A ₁₁And A ₂₁Bandwidth, time delay and packet loss:

$f\left(B_{A_{11}-A_{21}}\right)=\min (f\left(B_{A_{11}-A_1}\right),f\left(B_{A_1-A_2}\right),f\left(B_{A_2-A_{21}}\right))---\left(1\right)$

$f\left(D_{A_{11}-A_{21}}\right)=\mathrm{\Σ}(f\left(D_{A_{11}-A_1}\right),f\left(D_{A_1-A_2}\right),f\left(D_{A_2-A_{21}}\right))---\left(2\right)$

$f\left(L_{A_{11}-A_{21}}\right)=\mathrm{\Π}(f\left(L_{A_{11}-A_1}\right),f\left(L_{A_1-A_2}\right),f\left(L_{A_2-A_{21}}\right))---\left(3\right).$

The invention has the advantages that, the invention provides a kind of distributed network measurement method based on stratification cooperation multiagent system.To use hierarchical structure to realize distributed network measurement, eliminate the bottleneck problem that traditional centralized measurement exists, can press close to the actual physical network topology simultaneously, adapt to large-scale network environment.Employing realizes data interaction between measured node based on the cooperation multiagent system, and measured node can local stores it, can finish sharing of measurement result by the cooperation between multiagent system simultaneously, thereby avoid the drawback of traditional central database; End-to-end measurement is cut into many sub-links measures, each measured node only measures the link information of own adjacent node, thereby has avoided the problem that sublink is repeated to measure in end-to-end measurement repeatedly, has reduced network burden; Go out the end-to-end link result according to each sub-links test result calculations, thereby can orient easily the bottleneck zone of end-to-end link, make things convenient for the manager according to measurement result network to be optimized.

Description of drawings

Fig. 1 is traditional centralized network measurement model figure;

Fig. 2 is the distributed network measurement system diagram based on Chord;

Fig. 3 is stratification cooperation multiagent network architecture figure of the present invention;

Fig. 4 is distributed network measurement data flow diagram of the present invention.

Embodiment

The present invention will be described in detail below in conjunction with the drawings and specific embodiments.

The invention provides a kind of distributed network measurement method based on stratification cooperation multiagent system, the method adopts the cooperation multiagent system that is comprised of a plurality of measured node to realize distributed network measurement.Measured node is divided in the different logic levels according to own residing network site.Measured node forms bunch according to physical topology, and measured node is contiguous at physical topology in bunch, and each cocooning tool has a cluster head.Measured node is only measured and oneself is existed together one bunch the link information of other measured node in bunch.Cluster head can be in respectively in two bunches as the super node of each bunch, and the role in bottom bunch is cluster head, the role in upper strata bunch as bunch in ordinary node.Therefore, the cluster head node can obtain the real-time link state information of adjacent two layers network by implementing measurement.Simultaneously, the cluster head measured node is carried out transfer with the network measure data in the adjacent layer, thereby is made layer interior measured node can know the state information of layer outer network as measuring the mutual bridge of nodal information between adjacent layer.

Fig. 3 is stratification cooperation multiagent network architecture figure.As shown in Figure 3.Take 3 layer network frameworks as example, comprise 3 measured node in each bunch.Wherein, the cluster head in every one deck connects a cluster knot point in the upper strata by a dotted line, and dotted line represents mapping, in live network is disposed, only has a measured node.

Measurement data flow graph when Fig. 4 is distributed network measurement, as shown in Figure 4.Suppose to comprise among Fig. 42 levels, bunch 1 (Cluster1), bunch 2 (Cluster2), bunch 3 (Cluster3), bunch 4 (Cluster4) are in the same layer, and bunch 5 (Cluster5) are in the upper strata.Comprise 3 nodes in each bunch, be designated as respectively Cluster1 (A ₁, A ₁₁, A ₁₂, Cluster2{A ₂, A ₂₁, A ₂₂, Cluster3{A ₃, A ₃₁, A ₃₂, Cluster4 (A ₄, A ₄₁, A ₄₂, cluster head is respectively A ₁, A ₂, A ₃, A ₄4 bunches cluster head has formed Cluster5.

Flow process when system moves is as follows:

(1) in the Cluster1 measured node measure and bunch in network performance parameter between other measured node;

(2) in the Cluster2 measured node measure and bunch in network performance parameter between other measured node;

(3) in the Cluster3 measured node measure and bunch in network performance parameter between other measured node;

(4) in the Cluster4 measured node measure and bunch in network performance parameter between other measured node;

(5) in the Cluster5 measured node measure and bunch in network performance parameter between other measured node;

(6) Cluster1, Cluster2, Cluster3, Cluster4 sends local measurements by cluster head;

In this way, every link is only once measured in the one-shot measurement task, and test result obtains propagating in the whole network, thereby so that each measured node can be obtained local link-state information to other measured node.

The end-to-end link metrical information can be calculated by formula.With measured node A ₁₁And A ₂₁Bandwidth, time delay and packet loss are example:

$f\left(B_{A_{11}-A_{21}}\right)=\min (f\left(B_{A_{11}-A_1}\right),f\left(B_{A_1-A_2}\right),f\left(B_{A_2-A_{21}}\right))---\left(1\right)$

$f\left(D_{A_{11}-A_{21}}\right)=\mathrm{\Σ}(f\left(D_{A_{11}-A_1}\right),f\left(D_{A_1-A_2}\right),f\left(D_{A_2-A_{21}}\right))---\left(2\right)$

$f\left(L_{A_{11}-A_{21}}\right)=\mathrm{\Π}(f\left(L_{A_{11}-A_1}\right),f\left(L_{A_1-A_2}\right),f\left(L_{A_2-A_{21}}\right))---\left(3\right).$

By above formula, we can easier orient the bottleneck zone of end-to-end link.

It should be noted last that above embodiment is only unrestricted in order to technical scheme of the present invention to be described.Although with reference to embodiment the present invention is had been described in detail, those of ordinary skill in the art is to be understood that, technical scheme of the present invention is made amendment or is equal to replacement, do not break away from the spirit and scope of technical solution of the present invention, it all should be encompassed in the middle of the claim scope of the present invention.

Claims (4)

1. distributed network measurement method based on stratification cooperation multiagent system, the method adopts the cooperation multiagent system realization distributed network measurement that is comprised of a plurality of measured node; Described measured node is client computer; The concrete steps of the method comprise:

Step 1): at least 1 described measured node forms bunch according to physical topology according to own residing network site, the ground floor that bunch forms described stratification cooperation multiagent system of acquisition;

Step 2): each of ground floor bunch selected a cluster head as the measured node of last layer, and this measured node forms bunch according to physical topology according to own residing network site again, the second layer that bunch forms described stratification cooperation multiagent system of acquisition; And each of the second layer bunch selected a cluster head again as the measured node of last layer, by that analogy, until comprise the top layer of physical network topology of living in;

Step 3): the measured node in every cluster is only measured and other of one bunch of oneself existing together comprise network performance parameter between the measured node of cluster head, and network performance parameter is transferred to the cluster head of this bunch; The cluster head of every one deck is as the measured node of last layer, also only measure and one bunch other measured node of oneself existing together between network performance parameter, and network performance parameter is transferred to the cluster head of this bunch, the cluster head node can obtain the real-time link state information of adjacent layer network by implementing measurement.

2. the distributed network measurement method based on stratification cooperation multiagent system according to claim 1 is characterized in that the method also comprises step 4);

Described step 4): according to described step 3 the end to end real-time network performance parameter that obtains) is oriented the bottleneck zone of end-to-end link.

3. the distributed network measurement method based on stratification cooperation multiagent system according to claim 2 is characterized in that described network performance parameter comprises: bandwidth, time delay and packet loss.

4. the distributed network measurement method based on stratification cooperation multiagent system according to claim 3 is characterized in that described step 4) according to formula (1), formula (2) and formula (3) computation and measurement node A ₁₁And A ₂₁Bandwidth, time delay and packet loss:

$f\left(B_{A_{11}-A_{21}}\right)=\min (f\left(B_{A_{11}-A_1}\right),f\left(B_{A_1-A_2}\right),f\left(B_{A_2-A_{21}}\right))---\left(1\right)$

$f\left(D_{A_{11}-A_{21}}\right)=\mathrm{\Σ}(f\left(D_{A_{11}-A_1}\right),f\left(D_{A_1-A_2}\right),f\left(D_{A_2-A_{21}}\right))---\left(2\right)$

$f\left(L_{A_{11}-A_{21}}\right)=\mathrm{\Π}(f\left(L_{A_{11}-A_1}\right),f\left(L_{A_1-A_2}\right),f\left(L_{A_2-A_{21}}\right))---\left(3\right);$

Wherein, A ₁Cluster head node for bunch 1, A ₂Cluster head node for bunches 2, A ₁₁Be the common survey node of a non-cluster head in bunch 1, A ₂₁Common survey node for a non-cluster head in bunches 2.

Images