CN101753457B - Method and system for predicating available bandwidth of end-to-end route between nodes at Internet end - Google Patents

Abstract

The invention discloses a method and system for predicating available bandwidth of end-to-end route between nodes at Internet end. The method includes the steps of implementing performance predication domain division according to the connection relationship of different ISP affiliated networks, and arranging reference nodes and measurement management servers in each predication domain. The reference nodes in different predication domain mutually measure and reach the available bandwidths of each; common nodes and the reference nodes in the located domain thereof mutually measure to reach the available bandwidth of each; the measuring results are recorded by the measurement management servers in corresponding domain, and the available bandwidth of end-to-end route between nodes at Internetend is obtained through calculation. The invention takes the output available bandwidth and input available bandwidth of each node as two independent values to process when predicating available bandwidth. The invention is simple to be operated, has smaller network infestation, is convenient for automation, and has the advantages of being scientific, objective, simple and easy to be operated.

Description

The method and system of prediction internet end node available bandwidth of end-to-end route between

Technical field

The invention belongs to the network data communication field, is to propose a kind of method and system of predicting arbitrary node available bandwidth of end-to-end route between in the internet specifically.

Background technology

The available bandwidth of internet end-to-end path has material impact to the Distributed Application performance, current many large-scale distributed application all need to rely on this information and carry out performance optimization, for example, in using, will directly influence video transmission quality to the height of peer node path available bandwidth based on the IPTV of P2P pattern; Based on the streaming media service of CDN, can select the optimal service device to provide service according to the network performance situation for the user; Use the via node bridge joint to be arranged in the VoIP system of NAT/ fire compartment wall aft engine node at needs such as Skype, the node that performance information can help to select to have optimal performance is as bridging nodes, thereby provides best speech quality etc. for the user.Yet the grouping delivery service that Internet service is defined as doing one's best, it does not provide the information such as available bandwidth in different paths to upper layer application, cause Distributed Application can't directly obtain information needed from network.

In view of the importance of network end-to-end path available bandwidth information, people have proposed multiple network available bandwidth measurement forecasting mechanism at present ^[1,2]Yet in essence, these mechanism all are to utilize the mode of directly measuring to obtain available bandwidth information, and measuring on the one hand needs the cost regular hour, causes system can't in time respond user's performance information query requests; If a large number of users all adopts this mode will seriously invade and harass network on the other hand.For network, for the sensing lead that obtains available bandwidth between any two nodes is O (N with N node ²), this method is excessive to the invasion that network causes, and can't be applied to such as in the such large scale network in internet.In addition, the network coordinate system based on the space embedding of current people's proposition can be based on the prediction of round-trip delay between part metrical information realization node, with O (N ²) the measurement complexity be reduced to O (N) ^[3]Yet it is different with time delay between node with additivity feature, available bandwidth is a kind of concavity parameter in essence, the available bandwidth in stretch footpath depends on that this path passes through the minimum value of available bandwidth in the link, that is available bandwidth of end-to-end route is by the decision of the bottleneck bandwidth in this path, and the euclidean distance between node pair feature has essential distinction in this character and the Euclidean space.In this case, simply adopt the prediction of Euclidean space embedding carrying out available bandwidth will cause bigger predicated error.Also do not have at present a kind of mechanism that effectively realizes the network availability bandwidth prediction, need our essential characteristic, building network available bandwidth prognoses system in conjunction with the Internet architecture and end-to-end available bandwidth with low load.

On structure, the internet forms complicated multi-level hierarchy, can be international connection, national trunk and regional access network network etc. with it brief overview.Fig. 1 has provided typical the Internet layer aggregated(particle) structure, wherein core network is made up of some Tier-1 and the interconnected institute of Tier-2ISP, the edge network access network then is made up of the ISP of some regional character, and domestic consumer's node is connected in the internet by the edge access network network.This hierarchical structure of internet causes network performance to present zone nonuniformity, and for example, in order to adapt to the network application of continuous development, ISP carried out configuration to core network usually in the process of planning network, so its performance is relatively stable.And at edge network, because resource-constraineds such as bandwidth, for example end subscriber is extensive use of internet access waies such as DSL and Cable at present, causes final jump performance bottleneck often to occur ^[4]In addition, outlet bandwidth is also limited usually between the different I SP network, and then occurs performance bottleneck easily.

Summary of the invention

Purpose of the present invention is exactly in order to address the above problem, and proposes available bandwidth Forecasting Methodology and system thereof between a kind of science, objective and simple internet end node.

Technical scheme of the present invention is:

The Forecasting Methodology of available bandwidth of end-to-end route between a kind of internet arbitrary end node comprises the following step:

A. divide the performance prediction territory according to the annexation between the different I SP network in the internet, each performance prediction territory is corresponding to one or more continuous networks;

B. dispose i equally distributed datum node in the prediction territory in each prediction territory, the scope of i is 5～7, and each datum mark all has high access bandwidth, places available bandwidth measurement equipment at each datum node;

C. in each prediction territory a unique in logic management server is set, forms overlay network between the management server in different predictions territory, record datum node relevant information and metrical information in management server;

D. on network end node, place available bandwidth measurement equipment, adopt available bandwidth measurement equipment to measure arrival available bandwidth each other mutually, arbitrary end node H between m the datum node in each network end node and prediction territory, its place _AForm one according to measurement result and go out available bandwidth vector A _OutWith go into available bandwidth vector A _In, A wherein _Out(i) representative is by end node H _ATo i datum node LL _iThe available bandwidth measurement value, A _In(i) representative is by datum node LL _iTo H _AThe available bandwidth measurement value, and A _OutWith A _InAll be recorded in the management server in prediction territory;

E. adopt available bandwidth measurement equipment to measure arrival available bandwidth each other mutually between the datum node in the different predictions territory, this territory datum node of record management server in each prediction territory is to the available bandwidth measurement value of other prediction territory datum nodes;

F. arbitrary end node H _A, H _BBetween the result of available bandwidth of end-to-end route be:

H _ATo H _BAvailable bandwidth

For:

$\widetilde{\mathrm{avb}}(H_A,H_B)=\min \left\{\mathrm{avg}\right\{\mathrm{avb}(H_A,{\mathrm{LL}}_A)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_A,{\mathrm{LL}}_B)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_B,H_B)\left\}\right\}---\left(1\right)$

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place,

Avb (H _A, LL _A) be node H _ATo the available bandwidth measurement value set of its prediction territory, place datum node,

Avb (LL _A, LL _B) be node H _APrediction territory, place datum node is to H _BAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _B, H _B) be node H _BPrediction territory, place datum node is to H _BThe available bandwidth measurement value set,

Avg () is the function of averaging;

H _BTo H _AAvailable bandwidth

For

$\widetilde{\mathrm{avb}}(H_B,H_A)=\min \left\{\mathrm{avg}\right\{\mathrm{avb}(H_B,{\mathrm{LL}}_B)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_B,{\mathrm{LL}}_A)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_A,H_A)\left\}\right\}---\left(2\right)$

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place,

Avb (H _B, LL _B) be node H _BTo the available bandwidth measurement value set of its prediction territory, place datum node,

Avb (LL _B, LL _A) be node H _BPrediction territory, place datum node is to H _AAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _A, H _A) be node H _APrediction territory, place datum node is to H _AThe available bandwidth measurement value set,

Avg () is the function of averaging.

Of the present inventionly predict that respectively the management server in the territory is one, described management server connects a plurality of storage servers.

The prognoses system of a kind of internet end node available bandwidth of end-to-end route between, this system is made up of measuring basis node and management server two parts, be equipped with available bandwidth measurement equipment on datum node and the end node, management server is unique in logic in each prediction territory.

Prognoses system of the present invention: each predicts that the management server in the territory is one, and described management server connects a plurality of storage servers.

The present invention has the following advantages with respect to prior art:

Of the present invention simple to operate, less to the network invasion, be convenient to realize automation.After finishing initial configuration, just can measure and extract the network performance feature, make up forecast model, realize predicting the outcome and respond process such as user in predicting information inquiry.And in the available bandwidth forecasting process, the user can dynamically adjust measuring period and datum node and dispose number according to the needs of precision and sensing lead.For a network with N end node, if be divided into M performance prediction territory, dispose K datum node in each prediction territory, then total sensing lead is O (N*K)+O (M*K ²), generally M and K be much smaller than N, so sensing lead is also much smaller than the O (N of direct measurement ²), greatly reduce the expense of obtaining the node available bandwidth of end-to-end route between.

Available bandwidth accuracy for predicting height of the present invention.Than existing available bandwidth Forecasting Methodology, the present invention has fully utilized the architectural feature of internet and the basic distribution characteristics of available bandwidth, therefore has higher prediction accuracy.By in the internet, disposing some datum nodes and management server, the available bandwidth Predicting Technique is designed to a kind of general service in the network, can provide available bandwidth prediction service for various application.The present invention is for optimizing various Distributed Application performances, realizing that the win-win situation between network application and the ISP provides possibility based on available bandwidth information.

Description of drawings

Fig. 1 is a Internet layer aggregated(particle) structure schematic diagram.

Fig. 2 is the present invention's concrete deployment way in network.

Fig. 3 is an available bandwidth sampled measurements flow chart between network node among the present invention.

Fig. 4 is the flow chart that the arbitrary end node predicts other end node available bandwidths among the present invention.

Fig. 5 is an embodiment of the invention operation schematic diagram.

Embodiment

Bottom is further described the present invention with embodiment in conjunction with the accompanying drawings.

A kind of method of predicting available bandwidth of end-to-end route between the internet arbitrary node comprises the following step:

A. divide the performance prediction territory according to the annexation between different I SP in the internet (Internet Service Provider, the ISP) network, each performance prediction territory is corresponding to one or more continuous networks;

In the partition process of performance prediction territory, requiring each prediction territory must be a continuous network configuration, therefore combination that the performance prediction territory can be an AS (Autonomous System, autonomous system), an ISP network or a plurality of adjacent ISP networks.For example, China Telecom can form a performance prediction territory as an ISP, and perhaps it with each provincial network independence component property prediction territory, even constitutes jointly a performance prediction territory etc. with ISP such as UNICOM according to network configuration.Concrete which kind of mode that adopts need consult to determine when disposing this prognoses system.

B. determine the measuring basis node location, and place the available bandwidth measurement facility that the principle of selection reference node is that 5～7 equally distributed datum nodes in the prediction territory are disposed in each prediction territory, and have high access bandwidth at this point;

The scope of datum node access bandwidth avb (L) depends on that the access bandwidth maximum that prediction territory, place provides for the common end user is avb (E) arbitrarily, for guaranteeing precision of prediction, requires avb (L) 〉=2*avb (E).

C. dispose management server, a unique in logic management server is safeguarded in each prediction territory, forms overlay network between the management server in different predictions territory, in record management server datum node relevant information and metrical information;

D. operation available bandwidth measurement program on network end node, and the mutual available bandwidth that arrives each other, the arbitrary node H of measuring between m the datum node in its prediction territory, place _AForm one according to measurement result and go out available bandwidth vector A _OutWith go into available bandwidth vector A _In, A wherein _Out(i) representative is by end node H _ATo i datum node LL _iThe available bandwidth measurement value, A _In(i) representative is by datum node LL _iTo H _AThe available bandwidth measurement value, and A _OutWith A _InAll be sent to field managing server;

E. measure the available bandwidth that arrives each other mutually between the datum node in the different predictions territory, the management server in each prediction territory is safeguarded the available bandwidth measurement value of this territory datum node to other prediction territory datum nodes;

F. arbitrary end node H _A, H _BBetween the result of available bandwidth of end-to-end route be:

H _ATo H _BAvailable bandwidth

For:

$\widetilde{\mathrm{avb}}(H_A,H_B)=\min \left\{\mathrm{avg}\right\{\mathrm{avb}(H_A,{\mathrm{LL}}_A)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_A,{\mathrm{LL}}_B)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_B,H_B)\left\}\right\}---\left(1\right)$

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place,

Avb (H _A, LL _A) be node H _ATo the available bandwidth measurement value set of its prediction territory, place datum node,

Avb (LL _A, LL _B) be node H _APrediction territory, place datum node is to H _BAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _B, H _B) be node H _BPrediction territory, place datum node is to H _BThe available bandwidth measurement value set,

Avg () is the function of averaging;

H _BTo H _AAvailable bandwidth

For

$\widetilde{\mathrm{avb}}(H_B,H_A)=\min \left\{\mathrm{avg}\right\{\mathrm{avb}(H_B,{\mathrm{LL}}_B)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_B,{\mathrm{LL}}_A)\},\mathrm{avg}\{\mathrm{avb}({\mathrm{LL}}_A,H_A)\left\}\right\}---\left(2\right)$

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place,

Avb (H _B, LL _B) be node H _BTo the available bandwidth measurement value set of its prediction territory, place datum node,

Avb (LL _B, LL _A) be node H _BPrediction territory, place datum node is to H _AAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _A, H _A) be node H _APrediction territory, place datum node is to H _AThe available bandwidth measurement value set,

Avg () is the function of averaging.

Final result is got the concavity feature that minimum value has shown available bandwidth, and it then is a kind of weighting to all datum node measurement results that every section available bandwidth is averaged.

Suppose end node H _ATo H _BBetween end-to-end path experience n bar link altogether, and the available bandwidth of every section link is avb (i), H so _AWith H _BBetween available bandwidth of end-to-end route depend on minimum value among the avb (i), promptly

$\mathrm{avb}(H_A,H_B)=\underset{1\&le;i\&le;n}{\mathrm{<figure-callout\; id="1"\; label="min"\; filenames="H2009102648728E00011.png"\; state="\{\{state\}\}">min</figure-callout>}}\left\{\mathrm{avb}\left(i\right)\right\}$

This character is called the concavity feature of available bandwidth.

End-to-end available bandwidth prognoses system between a kind of the Internet nodes, this system is made up of measuring basis node and management server two parts.The feature of ordering by B in the aforementioned available bandwidth Forecasting Methodology in the position of measuring basis node is determined, operation can be initiated bandwidth measurement and the response to network node bandwidth measurement program to datum node to network node on it, it can measure arrival available bandwidth each other mutually according to the ordinary node in certain scheduling strategy and territory, place or the datum node in other territories, is used for finishing the function that available bandwidth Forecasting Methodology D is ordered.Management server is unique in logic in territory in each prediction, but considers to dispose physically a plurality of for robustness, is in charge of the performance queries request of common network node in datum node and the response territory, place.

During concrete enforcement:

At first provide the needed running environment of end-to-end available bandwidth Predicting Technique that the present invention proposes.As shown in Figure 2, edge network is divided into separate performance prediction territory, and in each performance prediction territory, disposes management server and datum node according to the annexation between the ISP network.In each performance prediction territory, as new node H _ADuring the adding system, it is at first to field managing server inquiry datum node information, with posterior nodal point H _AAnd periodically carry out available bandwidth measurement in the territory between these datum nodes, and measurement result is sent to field managing server.Simultaneously, also carry out the measurement of available bandwidth mutually between the datum node in the same area not, this measuring period can be according to dynamically adjusting.

As needs prediction end node H _ATo H _BAvailable bandwidth the time, H _AWith H _BInquire about its field managing server respectively and obtain H _AAvailable bandwidth measurement value set avb (H to its territory, place datum node _A, LL _A), end node H _BTerritory, place datum node is to H _BAvailable bandwidth measurement value set avb (LL _B, H _B), and end node H _A, H _BAvailable bandwidth measurement value set avb (LL between the datum node of territory, place _A, LL _B), can calculate H according to formula (1) then _ATo H _BAvailable bandwidth, in like manner can calculate H _BTo H _AAvailable bandwidth.

Fig. 3 and Fig. 4 have provided the surveying work flow chart and the available bandwidth evaluation work flow process figure of internet of the present invention available bandwidth of end-to-end route Forecasting Methodology respectively.Measurement flow process wherein shown in Figure 3 comprises datum node measurement and common end node measurement two parts, and flow process starts from initialization step S101, enters performance prediction territory partiting step S102 then, divides the performance prediction territory according to the network annexation.Next enter the deploying step S103 of field managing server and datum node.Enter step S104 then, the datum node in the different predictions territory is measured arrival available bandwidth each other mutually, enters step S105 subsequently, and measurement result is sent to corresponding field managing server.Enter step S106 subsequently, wait for a predetermined period T, measure again after overtime.

After the ordinary node adding system, execution in step S107 at first, datum node information in the management server queries territory in territory, place, enter step S108 subsequently, measure the available bandwidth that arrives each other mutually between the datum node in ordinary node and the prediction territory, its place, enter step S109 subsequently, measurement result is sent to field managing server.Enter step S110 after finishing this process, wait for a predetermined period T, measure again after overtime.

In available bandwidth evaluation work flow process shown in Figure 4, in order to predict node H _ATo H _BAvailable bandwidth, flow process starts from initialization step S101, enters step S102 then, H _ATo the management server queries avb in its territory, place (H _A, LL _A) and avb (LL _A, LL _B), enter step S103 subsequently, H _ATo H _BInquiry avb (LL _B, H _B), enter step S104 at last, calculate avb (H according to formula (1) _A, H _B), forecasting process finishes at S105.

Embodiment one:

Present embodiment has provided and has utilized content of the present invention to realize available bandwidth prediction service between network node in the such network environment of Fig. 5.

As shown in Figure 5, two edge networks are formed respectively and are independently predicted the territory, be referred to as territory A and territory B respectively, in each prediction territory, dispose 1 field managing server and 2 datum nodes, and suppose that edge network is 1000Mbps to the access bandwidth of core network, the access bandwidth of datum node is 100Mbps, H _AWith H _BAccess bandwidth be respectively 10Mbps and 5Mbps.If available bandwidth is by the access bandwidth decision of node between the interior nodes of same prediction territory, then at first calculates and to obtain by measuring

avg{avb(H _A，LL _A)}＝10Mbps

avg{avb(LL _A，LL _B)}＝100Mbps

avg{avb(LL _B，H _B)}＝5Mbps

So

$\widetilde{\mathrm{avb}}(H_A,H_B)=\min \left\{\mathrm{10,100,5}\right\}=5\mathrm{Mbps}$

As seen from Figure 5, above-mentioned predict the outcome and real network between node available bandwidth match.And, also can be easy to predict node H according to measurement result _AAvailable bandwidth to other nodes.Based on predicting the outcome, the equity side that can select to have higher available bandwidth such as the such Distributed Application of P2P is as the neighbor node data download, thereby can improve the data download efficiency, and reduces expending Internet resources.

List of references

[1]D?Antoniades，M?Athanatos，A?Papadogiannakis，et?al.Available?Bandwidth?MeasurementAs?Simple?As?Running?Wget.In?Proceedings?of?Passive?Active?Measurement?Conference2006.

[2]X?Jin，Q?Xia，S.Chan.A?Cost-based?Evaluation?of?End-to-End?Network?Measurements?inOverlay?Multicast.In?Proceedings?of?IEEE?Infocom?2007.

[3]C.Xing，M.Chen.A?Virtual?Node?Based?Network?Distance?Prediction?Mechanism.InProceedings?of?IEEE?GlobeCom?2008.

[4]M?Dischinger，A?Haeberlen，K?Gummadi，et?al.Characterizing?Residential?BroadbandNetworks.In?Proceedings?ofACM?Internet?Measurement?Conference?2007.

Claims (2)

1. the Forecasting Methodology of available bandwidth of end-to-end route between the internet arbitrary end node is characterized in that, comprises the following step:

A. divide the performance prediction territory according to the annexation between the different I SP network in the internet, each performance prediction territory is corresponding to one or more continuous networks;

B. dispose i equally distributed datum node in the prediction territory in each prediction territory, the scope of i is 5～7, and each datum mark all has high access bandwidth, places available bandwidth measurement equipment at each datum node;

C. in each prediction territory a unique in logic management server is set, forms overlay network between the management server in different predictions territory, record datum node relevant information and metrical information in management server;

D. on network end node, place available bandwidth measurement equipment, adopt available bandwidth measurement equipment to measure arrival available bandwidth each other mutually, arbitrary end node H between m the datum node in each network end node and prediction territory, its place _AForm one according to measurement result and go out available bandwidth vector A _OutWith go into available bandwidth vector A _In, A wherein _Out(i) representative is by end node H _ATo i datum node LL _iThe available bandwidth measurement value, A _In(i) representative is by datum node LL _iTo H _AThe available bandwidth measurement value, and A _OutWith _InAll be recorded in the management server in prediction territory;

E. adopt available bandwidth measurement equipment to measure arrival available bandwidth each other mutually between the datum node in the different predictions territory, this territory datum node of record management server in each prediction territory is to the available bandwidth measurement value of other prediction territory datum nodes;

F. arbitrary end node H _A, H _BBetween the result of available bandwidth of end-to-end route be:

H _ATo H _BAvailable bandwidth For:

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place, avb (H _A, LL _A) be node H _ATo the available bandwidth measurement value set of its prediction territory, place datum node, avb (LL _A, LL _B) be node H _APrediction territory, place datum node is to H _BAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _B, H _B) be node H _BPrediction territory, place datum node is to H _BThe available bandwidth measurement value set,

Avg () is the function of averaging;

H _BTo H _AAvailable bandwidth

For

Wherein, LL _A, LL _BRepresent end node H respectively _AWith H _BThe datum node collection in prediction territory, place,

Avb (H _B, LL _B) be node H _BTo the available bandwidth measurement value set of its prediction territory, place datum node,

Avb (LL _B, LL _A) be node H _BPrediction territory, place datum node is to H _AAvailable bandwidth measurement value set between the datum node of prediction territory, place,

Avb (LL _A, H _A) be node H _APrediction territory, place datum node is to H _AThe available bandwidth measurement value set,

Avg () is the function of averaging.

2. the Forecasting Methodology of available bandwidth of end-to-end route between the internet according to claim 1 arbitrary end node is characterized in that: each predicts that the management server in the territory is one, and described management server connects a plurality of storage servers.

Images