CN1972206A - A method for building combination-based P2P system - Google Patents

Abstract

This invention relates to one combination P2P system method, which comprises the following steps: analyzing P2P system type and require and defining relative rules; b, organizing the points to system; c, estimating joints ability and durability; d, distributing roles and rules constitution method; e, fixing points for preset parts; f, establishing each degree intense index and rare index in data sharing process and fixing buffer; g, entering system maintenance.

Description

Method based on building combination P2P system

Technical field

The invention belongs to the method that a kind of P2P (peer-to-peer network) application system is set up, particularly a kind ofly can be applied to large-scale, complex network adaptively, relate to the method that mass file is shared, information system (comprising E-Government, ecommerce, all kinds of decision-making and old mis system) transformation is used based on building combination P2P system.

Background technology

The P2P system provides a kind of possibility for distributed system, can provide service to any terminal equipment of a common communication medium (as Internet) based on resource, thereby is reduced under the Client requirement to the special service supplier.So-called overlay network is a special virtual network of setting up on the physics communication network.It is that the pressure to physical network is abstract, both can satisfy functional requirement flexibly, and for example connective maintenance, index and route also can satisfy the non-functional requirement, for example autgmentability, fault-tolerance and load balance.When the design overlay network, an importance is exactly a customized solution, should satisfy above-mentioned requirements, also will avoid the potential conflict in the above-mentioned requirements.

Summary of the invention

The objective of the invention is on existing P2P network system basis, provide the expansion of a kind of discussing system better, improve the stability of a system, reliability, correctness and adaptive method, to overcome the deficiency that has the P2P network system now based on building combination P2P system.

To achieve these goals, concrete steps of the present invention are as follows:

First step: analyze type and the requirement of setting up the P2P system, these types are reduced following a few class:

1) extensive shared file system: storage, search efficiency, upgradability are had relatively high expectations, and node adds and withdraws from frequently, needs function supports such as special maintenance, buffer memory, index especially;

2) information system: higher to deal with data logic isomerism, the collecting and distributing property of data, data correctness, security requirement, node stability is higher, index not only relates to content but also relates to the logical construction of data, needs especially that central service, data are collecting and distributing, function support such as content indexing and configuration index, fail safe control;

3) multimedia application: information quality is had relatively high expectations, and the node stability situation is indefinite, needs function supports such as maintenance, buffer memory (load balance), route especially;

Relevant regular by their definition:

1) bunch scale (node maximum quantity);

2) dense index and sparse index way (how many positions are set up sparse index according to keywords);

3) node is replaced way (same bunch, the 1st or the 2nd identifies close) etc.;

Second step: one by one with node organization in system, it is according to being:

1) backbone network and host address situation are determined the 1st class sign;

2) if information system is determined the 2nd class sign according to organization, other system is determined the 2nd class sign according to de Brui jn figure;

Organize sub-clustering according to sign, set up bunch tree, set up bunch inner tissue according to the 2nd class sign according to the 1st class sign;

Third step: ability, the persistence of estimating node: the ability of node is the important evidence of distributing the role, need again by the match of lognormal Mixture Distribution Model, give ability, persistent quantitative data of egress according to experimental data.

The universal model curve is: $f_{\mathrm{Lgn}}\left(x\right)=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;x}}{e}^{\frac{{(\log \left(x\right)-\mathrm{\&mu;})}^2}{{2\mathrm{\&sigma;}}^2}}-\&infin;<X<+\&infin;$

Wherein μ, σ are two constants to be determined, describe life span respectively and bear and finish the work situation;

The 4th step: distribution role, regulation replacement way or redundancy strategy: according to the capability distribution role of node, most of nodes are the dispensing connector role only, the higher node of μ distributes the maintenance device role, the higher node of σ distributes buffer memory role, the collecting and distributing isogonism look of node allocation index, the data that μ, σ are all high;

The 5th step: predetermined parts are installed:, different universal adaptive parts are installed at dissimilar P2P networks relevant role;

The 6th step: set up dense indexs at different levels and sparse index in inquiry and shared data procedures, arrange buffer memory: set up dense index at the leaf submanifold, employing inverted list structure on the upper strata bunch, is set up sparse indexes at different levels according to predefined rule;

The 7th step: enter system maintenance: when key node withdraws from, the adding of maintenance device responsible node with withdraw from, add fashionable calculating and arrange to add corresponding bunch, and follow the tracks of and its ability of test, arrange potential role, in time note abnormalities when withdrawing from, replace by the Substitution Rules arrangement for the special role node.

Above-mentioned extensive shared file system, information system, multimedia application require by upgradability, retractility, correctness, reliability, the mistake proofing characteristic value tolerance different with fault-tolerant ability, isomerism, data integrity, consistency, bandwidth, link transmission delay, packet loss rate, fail safe etc.

In the second above-mentioned step,, add according to deBrui jn figure or two kinds of signs of information system institutional framework definition according to the IP address of backbone network and main frame, and based on two kinds of methods that identify sub-clustering, set up network topology.

In the above-mentioned third step, ask for node life span and the method for bearing task quantity by the model match.

In the fourth, fifth above-mentioned step, design multiclass role reaches according to ability and system requirements and distributes the role to set up the method for system flexibly with arranging parts.

In the 6th above-mentioned step, about in the P2P system, setting up the method for dense index and sparse index.

The invention provides a kind of new architecture that is used for designing effective P2P system overlay network, be called large complicated composite type P2P network (LSCC).Consider the most important characteristic of large scale system, particularly consider more most important characteristics of a large amount of participants' system on a large scale, for example upgradability and autgmentability requirement, some of consideration overlay network topology cause the uncontrollable and unpredictable behavior of dynamic change, need establish flexible systems stabilisation, enable to keep efficient operation with minimum maintenance cost.In addition, also need to consider its inherent isomerism physically, and the user behavior that can increase the weight of workload distribution imbalance problem.

The new space that the present invention seeks to develop according to the more already present mechanism that meets above-mentioned requirements.As large-scale P2P system, be not only to be used for ripe at present multimedia shared system and instantaneous communication system, also be not only to be used at present, also will consider systems such as words are gradually many at present for example ecommerce, E-Government, teleconference and remote teaching, decision support, the integration of SOA data also at the shared file system of very fast development.Must consider to expand the notion of original P2P aspect strengthening management, the present invention follows a kind of method based on the role: common, the most basic operation is to each node distribution role according to ability and the behavior estimated to the P2P overlay network.Especially, be defined as follows core roles: maintenance device, router, buffer, index, connector, safety check device, the collecting and distributing device of data etc.Design universal component and served as these roles.These different roles will make each node can effectively participate in network, improve overall performance effectively.The variation of role and parts and the refinement of function enable to set up the types of applications system flexibly according to application need, satisfy needs of different applications, enlarge range of application.The rule that has defined some realizes fair solution automatically, guarantees that node contribution balance and stability is reliable.Node not playing role, can bear a role, also can bear a plurality of roles, composition mechanism can be used among the different P2P systems LSCC flexibly.Consider the key effect of index in resource query, large-scale directory system requires very big memory capacity, occupies very big bandwidth, is the bottleneck of inquiry work, and the structure that meets that design adopts dense index to add sparse index makes scatteredload, raises the efficiency.Consider that heterogeneous system had both comprised the isomerism of physics aspect, the isomerism that also comprises the logic aspect, for information system, the latter's research also to be put on agenda, the special proposition is provided with the scheme that automatic realization Different Logic structural system is integrated by the parameter of a fixed structure.

Aspect network topology structure, consideration by backbone network and host address combine with use de Brui jn figure cluster structured, exponential autgmentability is the intrinsic characteristic of de Brui jn figure, have diameter and be the notable feature that logarithmic series increases, even the degree of node is constant, also can spend minimum cost to satisfy the upgradability demand.But will note adopting progressive extended attribute to define deBrui jn variable during design, in addition, the degree of node is fewer to be reduced the structure maintenance cost better as far as possible, but wants to satisfy the needs of using.Another content of design is grouping, makes to guarantee that its reliability and multilayer topological structure, De Brui jn structure are applicable to that striding the group contact constitutes one and mix topological structure with adopting bunch inner mode that combines that is connected of collection, provides a tight network of constructing.Particularly, guarantee to give near equity point to select freedom near several bunch members.This selection may be enabled effectively to be mapped to the network and the satisfied safety of bottom and to trust requirement by various strategies and metric constraint.

Analytical method of the present invention is the data according to random acquisition, analyzes and assess the working condition of the P2P system that extensively exists.Utilization system journal and behavior tracking method are stipulated each part-time role's task, utilize analogy method to obtain the interaction of the dynamic and P2P agreement that comprised of the essential characteristic of network, user's behavior exactly.

Description of drawings

Fig. 1 is the tack schematic diagram of embodiment of the invention query requests service.

Fig. 2 is embodiment of the invention router and the two related work schematic diagram of index.

Fig. 3 is an embodiment of the invention equity point adition process schematic diagram.

Among Fig. 1: 1-query requests step, 2-judge whether responsibility bunch step, 3-transmit query requests step, 4-judge whether visible step, 5-judge whether buffer memory step, 6-index provide respond step, 7-buffer provide respond step, 8-outwards transmits query requests step, 9-query response step;

Among Fig. 2: 1-issue request step, 2-judge whether that responsibility bunch step, 3-transmission issue request step, 4-index entry step, 5-index determining step, 6-renewal index request step, 7-issue transmission step, 8-issue determining step;

Among Fig. 3: join request step, 2-random walk step, 3-of 1-judges whether to accept step, 4-and accepts that step, 5-Connection Step, 6-are upgraded bunch map step, 7-upgrades the routing table step.

Embodiment

The invention will be further described below in conjunction with drawings and Examples, but this embodiment should not be construed as limitation of the present invention.

If overlay network that shared file system is used.At first initiate, set up bunch, have Centroid to give special duty simultaneously by a part of node machine: whole network is managed, the adding of responsible node with withdraw from, bunch establishment and maintenance, index mechanism and maintenance thereof.At first consider when setting up bunch by network the ABC class, by the network address, by the subnet address cluster dividing.Network bunch in topological structure follow de Brui jn directed graph structure.Add as new node and fashionablely to identify then at first by application and carry out safety inspection, press dual mode sign, one is to try to achieve by hash function calculating, is that the binary number mode according to de Brui jn directed graph identifies in addition.To adding the node maximum number 2 of new node above current de Brui jn directed graph ^N-1The time, node in the division de Brui jn directed graph selects the little node of the identifier degree of depth of node to implement division as far as possible.The characteristics that as far as possible keep original de Brui jn directed graph during division.When bunch scale (this scale is by the relevant rule predetermining of Centroid) when reaching maximum, set up new bunch.If upper level bunch is arranged when setting up new bunch thereon, is integrated among the upper level bunch administration; If there is not upper level bunch, newly-built its upper level bunch, newly-built bunch all is integrated into upper level bunch administration with original bunch.

Inquiry for convenience and the management of strengthening inquiry need be set up index structure.Bunch set up dense index and sparse index respectively to each layer of system centre node at each layer, dense index refers to each file, each keyword one index entry, generally build in resource source node or the resource source node place leaf submanifold.If such index comprises the information of all resources of all nodes of total system, this index file is incited somebody to action and is huge, to index place node is that CPU or network service all are white elephants very, the huge easy formation network congestion of flow, and search efficiency is extremely low.If adjacent node only built in the resource information index of each node, inquiry will be very slow.If adopt random fashion to be distributed in some nodes, be not best solution for large complicated network, there is not effective solution for range query.LSCC considers to set up successively sparse index, sparse mode is according to the keyword figure place, if for example divide 3 layers, index set up in the 1st layer of according to keywords preceding 2 character, index set up in the 2nd layer of according to keywords preceding 4 character, and index or the like set up in the 3rd layer of according to keywords preceding 10 character.Concrete sparse way is by the relevant rule file regulation of Centroid.The territory of index comprises: bunch number, nodal scheme, domain name, file name, keyword.The destination address of last layer index search for following one deck bunch bunch number; The destination address of leaf submanifold is pointed to relevant resource.

In the leaf submanifold, to set up key table, when a fresh target inquiry takes place on the network when, should inquiry to all node broadcasts.Each node is checked this keyword whether in key table, or is not contained by content in the key table, if "Yes" is not just replenished key table, but will be searched for the file that produces recently in this node.If not in key table, it need be added in the key table, again to all are searched for for the file of sharing in this node.If find the file that is mated, then revise dense index in this bunch, add an index entry.Can apply manual intervention to the interpolation process, reduce the situation of duplicate contents or close content indexing as far as possible.Cluster knot point relevant procedures will be checked the interpolation process, check this document whether be other of index with the copy of keyed file or part roughly the same, for example can stipulate that above certain with index of 80% or 500 words of file is identical with keyed file, just think roughly the same part, will not add index, or in index, mark especially.Inspection is regular by artificial formulation, fills out to be placed in the system convention storehouse.To when range query, effectively increase hit rate like this.If add new index, send message for simultaneously upper level cluster knot point, the sparse index in the modification upper level bunch is up to the system centre index node.

Sparse index is set up process and is taked splitting algorithm, when when a sparse index is added fresh content, if scale full (each index scale is by the rule decision of the artificial formulation of building the system centre node in), then the predetermined potential empty indexing role node of alternative is set up new sparse index, by certain rule (artificial formulation, build in the system centre node), former sparse index content separately is stored in two indexing role nodes.For example, former index comprises d, e, and the content that f, g etc. take the lead can be with d, and the index that e takes the lead is stayed in the origin node, and with f, and the index that g takes the lead changes and is put in the new node, upgrades last layer index node index content simultaneously.

Some key players' the course of work will be described especially: the router role is the main individuality of query requests service, and Fig. 1 shows its tack.The first step, inquiry request message trigger its work.Local message dispatcher passes to router with inquiry request message, includes the key content (step 1) of bunch identifier and request in this message.In second step,, determine whether it is problem (step 2) in local bunch by bunch identifier in the message and keyword request being compared and judging.Request in local bunch is then according to keywords searched downwards from the system index node, from level to level up to leaf node.If the request in local bunch is just write instructions and transfer and is given with the reciprocity point (step 3) of having distributed the index role in the cluster.After this, three selections can be arranged.The first, if the index role is assigned to local reciprocity point really, to index role inquiry and confirm whether the object of being asked has come forth and be now just can obtain (step 4).If can obtain then, take out destination address and reply router (step 6).The second, if Cacher role is assigned to local reciprocity point, (step 5) will provide to reply to turn back to router (step 7) to need to confirm whether be buffered in local buffer memory role node by the object of being asked.The 3rd, if, request is write instructions and transfer from the next reciprocity point (step 8) when prevariety if it is not at local buffer memory role node that Cacher role is not assigned to the object of local equity point or request.If the project of request is included in local index or buffer memory role node, institute's information requested (step 9) is got in touch and obtained to requestor's equity point with supplier's equity point.

Fig. 2 shows router and the two related work situation of index.This plot triggers (step 1) by the request notice message.Similar in appearance to previous plot, message is passed to local router, checks that target is whether in current place bunch (step 2).Comprise bunch identifier and correspondent keyword in request in the notice message, these are one of contents of announcement, compare according to bunch identifier and correspondent keyword and can obtain conclusion.If comparative result determines that request is just write instructions and transfer and given near same bunch reciprocity point (step 3) of having distributed the index role.Then, the index role calls out and checks the object (step 4) of having done announcement recently new in the local index.The request of dispatching of local index revise index messages near belong to same bunch index (step 6).Because query task is frequent the generation, other inquiry also may take place in the same stage.Index will successfully have been done object router explanation and announce that (step 4) provides subsequently and confirms to get back to request notice message source (step 8).If router is not received the information of successfully searching object, should be with message repeater system center index node.

If final notice message source node can not obtain to search the message of success, illustrate in existing index and do not find target, should send to total system once more and newly search message.Other nodes are had a responsibility for searching in the shared file of this node according to the keyword in this message in the system.Should be transmitted to notice message source node if find relevant resource, to place bunch index node report, add new index entry simultaneously, and have place bunch index node to initiate upwards successively to upgrade each layer sparse index content.

Equity point adition process is different with previous process, and Fig. 3 shows its course of work: it adds message trigger (step 1) by request.Request adds message and is issued the reciprocity point of serving as the maintenance device role by the equity point.Near the member of bunch selected maintenance device (step 2) the maintenance device random access.Select to accept new equity point more suitable bunch, then, the reciprocity point that the maintenance device notice adds.But, be installed in the role of equity point and may be only get in touch with to the similar role of remote equity point, because new reciprocity point can not be assigned with the maintenance device role, it can not be a new reciprocity point of accepting the maintenance device role.It is blank that the connector role is used to fill up this design.Connector may provide three critical functions: 1. connectivity functionality, 2. new role facility function and 3. bunches of figure update functions.In these cases, maintenance device agreement connector is accepted new reciprocity point (step 3).Local connector is got in touch and is made with remote connector again and adopts decision (step 4).Afterwards, new equity is put to the maintenance device request as the same bunch of (member of step 5).Maintenance device provides near bunch figure this machine, and (new reciprocity point in the step 6) writes its effective address (step 7) in its routing table.

The content that is not described in detail in this specification belongs to this area professional and technical personnel's known prior art.

Claims (6)

1, a kind of method based on building combination P2P system, its concrete steps are:

First step: analyze type and the requirement of setting up the P2P system, these types are reduced following a few class:

1) extensive shared file system: storage, search efficiency, upgradability are had relatively high expectations, and node adds and withdraws from frequently, needs function supports such as special maintenance, buffer memory, index especially;

2) information system: higher to deal with data logic isomerism, the collecting and distributing property of data, data correctness, security requirement, node stability is higher, index not only relates to content but also relates to the logical construction of data, needs especially that central service, data are collecting and distributing, function support such as content indexing and configuration index, fail safe control;

3) multimedia application: information quality is had relatively high expectations, and the node stability situation is indefinite, needs maintenance, buffer memory, routing function support especially;

Relevant regular by their definition:

1) bunch scale;

2) dense index and sparse index way;

3) node is replaced way;

Second step: one by one with node organization in system, it is according to being:

1) backbone network and host address situation are determined the 1st class sign;

2) if information system is determined the 2nd class sign according to organization, other system is determined the 2nd class sign according to de Brui jn figure;

Organize sub-clustering according to sign, set up bunch tree, set up bunch inner tissue according to the 2nd class sign according to the 1st class sign;

Third step: ability, the persistence of estimating node: the ability of node is the important evidence of distributing the role, according to experimental data, again by the match of lognormal Mixture Distribution Model, gives ability, persistent quantitative data of egress,

The universal model curve is: $f_{\mathrm{Lgn}}\left(x\right)=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;x}}{e}^{-\frac{{(\log \left(x\right)-\mathrm{\&mu;})}^2}{2{\mathrm{\&sigma;}}^2}}-\&infin;<X<+\&infin;$

Wherein μ, σ are two constants to be determined, describe life span respectively and bear and finish the work situation;

The 4th step: distribution role, regulation replacement way or redundancy strategy: according to the capability distribution role of node, most of nodes are the dispensing connector role only, the higher node of μ distributes the maintenance device role, the higher node of σ distributes buffer memory role, the collecting and distributing isogonism look of node allocation index, the data that μ, σ are all high;

The 5th step: predetermined parts are installed:, different universal adaptive parts are installed at dissimilar P2P networks relevant role;

The 6th step: set up dense indexs at different levels and sparse index in inquiry and shared data procedures, arrange buffer memory: set up dense index at the leaf submanifold, employing inverted list structure on the upper strata bunch, is set up sparse indexes at different levels according to predefined rule;

The 7th step: enter system maintenance: when key node withdraws from, the adding of maintenance device responsible node with withdraw from, add fashionable calculating and arrange to add corresponding bunch, and follow the tracks of and its ability of test, arrange potential role, in time note abnormalities when withdrawing from, replace by the Substitution Rules arrangement for the special role node.

2, the method based on building combination P2P system as claimed in claim 1 is characterized in that: extensive shared file system, information system, multimedia application require to be measured by upgradability, retractility, correctness, reliability, mistake proofing and fault-tolerant ability, isomerism, data integrity, consistency, bandwidth, link transmission delay, characteristic value that packet loss rate is different with fail safe.

3, the method based on building combination P2P system as claimed in claim 1, it is characterized in that: in second step, IP address according to backbone network and main frame, add according to de Brui jn figure or two kinds of signs of information system institutional framework definition, and based on two kinds of methods that identify sub-clustering, set up network topology.

4, the method based on building combination P2P system as claimed in claim 1 is characterized in that: in the third step, ask for the node life span and bear task quantity by the model match.

5, the method based on building combination P2P system as claimed in claim 1 is characterized in that: in fourth, fifth step, design multiclass role reaches according to ability and system requirements and distributes the role to set up the method for system with arranging parts.

6, the method based on building combination P2P system as claimed in claim 1 is characterized in that: in the 6th step, set up the method for dense index and sparse index in the P2P system.

Images