CN101800655A - Peer-to-peer service system establishing method for contributing resources to application of large-scale internet - Google Patents

Abstract

The invention discloses a peer-to-peer service system establishing method for contributing resources to the application of a large-scale internet, short for a peer-to-peer server, belonging to the integrated application of multiple distributed system technologies to the internet. The method comprises the following steps: enduing computation nodes with the capability of receiving and sending requests, and establishing internet peer-to-peer points of the computation nodes; adopting a transmission mode characterized by permanent flow as the basic mode for the data interchanging among the peer-to-peer points, and establishing dynamic clusters based on the common data and the service requirements; acquiring various performance characteristics of the clusters by the arrangement and the use of a node tracker, and determining the current state of the corresponding cluster; and finally, judging based on the monitoring results, wherein the peer-to-peer server uses different working strategies to assist the clusters so as to improve the performance of the entire system. The invention transfers a current internet server into a peer-to-peer server, and sufficiently and rationally contributes computing resources to the internet system, thereby improving the overall operating efficiency of the internet and promoting the development of the large-scale non-center internet application. The peer-to-peer mode enables the internet server to actively contact the client instead of passively waiting. Therefore, the internet server can actively assist the client in the condition of resource abundance, and can selectively help the client under the condition of resource shortage, thereby realizing the optimizing configuration of resources.

Description

A kind of reciprocity service system establishing method of using dedicate resources for large-scale internet

Technical field

Peer server among the present invention belongs to Internet technology, is the significant improvement to current internet server operational mode.Internet technology is a special case of distributed file system in essence, is the application of Distributed Calculation maximum-norm.During current internet was used, Internet Server had decisive action to the normal operation of whole system.Though the present invention has reduced the dependence of internet, applications to Internet Server, it still has important support effect to internet, applications.Peer server is the concentrated reflection that multiple distributed system technological synthesis is used.

Background technology

Internet Server [1] [2] is one of important foundation software of another internet, applications except browser [3].It holds the balance to the effect of the Internet, does not have current internet system without Internet Server.Such effect is owing to the current internet special construction causes, and promptly the Internet computational resource of division center comes from Internet Server fully.

The current internet server is in the space of prominence of decision systems performance.Yet,, must consider that the concrete condition of internet, applications comes server is designed, to avoid becoming the bottleneck of internet, applications as far as possible in order to reach this purpose.The greatest problem that Internet Server will solve is exactly the visit capacity of potential customers' end.This visit capacity is unforeseen.So in the design Internet Server, must optimize as far as possible and use the resource that server had, make the visit of client not be delayed even refuse; Server is unlikely to be unable to cope with when the visit peak.Some common servers all design with this thinking.Current for this reason Internet Server will be considered following factor.

At first the key factor of Kao Lving is the design of thread pool [4].Thread pool is a kind of solution that improves resource utilization ratio by concurrent mode.Current Internet Server all operates on the high performance computing equipment.The computational resource that these computing equipments had is far away more than general computing equipment.Make full use of these computational resources, must take high as far as possible concurrent mode to deal with a large amount of user's requests.Promptly when a large amount of access request wait, should the reasonable distribution computational resource make each request can both have suitable chance and obtain handling and response; And when not having relation of interdependence between the request, processing of request and response are carried out as far as possible simultaneously.The use of multithreading [4] has just become inevitable like this.Thread is the computational entity of independent operating in the computing system, and it need take computational resource, as central processing unit, internal memory, lasting memory and input-output equipment etc.; It can have no the asynchronous operation of relation ground at running other threads that neutralize, and also can get in touch by suitable mode and other threads in case of necessity; The cross-thread of asynchronous operation may be pind down owing to share above-mentioned resource fully, promptly needs to carry out Synchronization Control.Which kind of situation no matter, thread is to improve the necessary technology of systematic function.Yet because each thread all needs to take computational resource, if concurrency is infinitely enlarged, and the resource that each thread takies do not discharge, and must cause system resource to exhaust, and calculating can't be kept down.Therefore, suitable mechanism must be arranged is increasing without limitation of thread pool restriction resource use.Under this extreme case, thread pool can make the request that exceeds the computing equipment affordability meet with a response by suitable scheduling strategy.Common situation is that the resource request amount does not reach the computing equipment upper limit.Also need to take appropriate measure this time, rationally use resource.A general process has a thread pool, moves at the same time and have a plurality of processes in computing equipment.Even burden is little concerning each process, but adding up of a plurality of processes will form pressure to system resource.So thread pool must be under any circumstance, can both be to resource utilize rational management and distribution.Following principle is that the design thread pool will be considered.The first, the thread of Qi Donging is few as far as possible simultaneously.Be enough to deal with under the prerequisite of concurrent request, the thread of Qi Donging is few more simultaneously, and the consumption of system resource is also few more.The second, reuse the thread that has started as far as possible.Because each request processing time all is limited, after the intact request of a thread process, the request that can utilize this thread process waiting for, thus avoid starting new thread, increase the consumption of resource.The 3rd, the application schedules strategy reaches reasonable resources to be used.No matter be formation strategy, or timesharing strategy, can both make the utilization of limited resources reach justice and high usage.The 4th, in time reclaim resource.Because the request to resource is in fluctuation status, when the resource request amount was more and more lower, the thread of some startups just had no chance to be repeated to use; Need cancel these idle threads this time, and reclaim the resource that it takies.The 5th, according to the difference of concrete application, design corresponding thread scheduling scheme.Except some universal scheme, come the specialized designs scheduling strategy also must consider in conjunction with different application characteristics.Such as, as Internet Server, just should consider that the characteristics of internet access are formulated special scheduling strategy.The 6th, thread pool should possess adaptability.A thread pool has a plurality of parameters to come the behavior of standard thread pool usually, need be in the waiting time after preceding thread starts, the waiting time of idle thread recovery and the timeslice length in the timesharing strategy or the like such as size, new thread of startup of thread pool.These parameters can the mode with static state define before system operation, but the more important thing is and make thread self possess adaptive capacity, in time adjusted above-mentioned each parameter according to the history feature of request.The 7th, survey the running status of whole system and adjust the thread pool strategy.Because any thread pool is not an isolated operation, all be the resource that is within the computing equipment and consumes entire equipment, so also will having the ability to survey whole system resource operating position, a good thread pool also adjusts the strategy of self in view of the above.The whole system operation will become and coordinate more like this.

Internet Server requires to possess a suitable application layer protocol.Internet Server operates in based on TCP/IP[5] under the network environment of agreement.Yet ICP/IP protocol is a puppy parc, does not consider the concrete feature of internet, applications.At first, ICP/IP protocol is a peer protocol, can be mutually with mode initiatively and the other side's contact between any two nodes that promptly connect; Either party both can be used as initiatively visitor, the person that also can be used as the passive response.Secondly, ICP/IP protocol is not done any definition at the mutual characteristics in the Internet.These characteristics are exactly WWW[3] service.The Internet normally carries out between browser and Internet Server alternately.Data are to pass through HTML[6 in the middle of the browser] display, it includes a large amount of link and multiple heterogeneous datas based on domain name.Being actually alternately on the Internet done the teledata request according to these links.In addition, also have the situation of some dynamic interactions, promptly the data returned of server need decide according to the parameter that the user provides.These problems, ICP/IP protocol all do not have to consider.In order to adapt to the requirement of internet, applications, based on the application layer protocol HTTP[7 of ICP/IP protocol] be suggested.Http protocol is not a peer protocol, promptly has only client to have the ability initiatively to send request to server end, and server end can only passive response and can not be initiatively and client contacts.Such restriction is in order to reduce the resource pressure of server end.The current internet system sets up for minority mechanism shared resource at the beginning of design.Comparatively speaking, Internet Server is in the minority status, and client terminal quantity is greater than server end.Thereby cause the resource of server end normally limited; These Limited resources should be mainly used in the visit of handling client, rather than go active and client contacts; Otherwise server resource can be nervous more.In addition, therefore normally the taker of data rather than the supplier of data of client also there is no need to allow server go initiatively access client at that time.Some structures on the current internet and the possibility also further limited such visit is set,, fire compartment wall [8] limited and NAT[8 as internet address] etc.In a word, this asymmetrical application layer protocol is to the adaptation of the Internet and center type network environment at that time.

The visit pressure of Internet Server buffer memory design also can alleviating server.The operating mechanism that the current internet server is relied on causes server resource always to be in finite state.In order further effectively to utilize computational resource, the buffer memory design is also extremely important.The basic ideas of buffer memory are the pressure of avoiding repeated accesses that server is formed.According to general statistical law, the client-access server data is concentrated relatively, and not all data all possess same accessed probability; And high access probability concentrates on the minority data usually.Be based on the recognition, after user's ever accessed was crossed some data, the possibility that these data are visited once more was very big.Like this, if independently caching mechanism is arranged, user's visit pressure will be transferred on these equipment so, and server is not impacted.

Internet Server does not generally all design condition managing.So-called state refers to the various intermediate object programs in the computing system running.The management of these data is prerequisites that computing system is able to operate as normal.Originally the Internet is used for sharing static file, visits these data and does not have intermediateness, therefore state is not carried out any management.But owing to there is the new internet, applications of state to occur, Internet Server must be considered this demand; This has increased the computational resource consumption of server.

Internet Server must be made the visit that a large number of users is satisfied in restriction to accessed size of data.Strictly speaking, it is irrational accessed size of data being limited.But, have to accessed data (file) are carried out the size restriction increases the server thread pond to prevent that thread running time is long pressure because the current internet server runs on the division center, computational resource always is in finite state.Another valuable source of Internet Server is exactly a bandwidth.The bandwidth that Internet Server had also is limited.A long-time thread of being responsible for the transmission data also can cause other client-access to be restricted to taking for a long time of bandwidth except the computational resource build-up of pressure to server itself.So general Internet Server is all made restriction to accessed size of data; Most of internet, applications are aided with a small amount of media file still based on the lightweight text.Except limiting accessed size of data, at the feature of http protocol, also having a common measure is that strictness limits the size that each HTTP responds.A common HTTP responds and comprises a plurality of heterogeneous files; Both made each file relatively little, if but the file that a HTTP comprises in responding is too much, thus its summation also can surpass proper standard to the server build-up of pressure.This can reach by limiting the file number that comprises in each HTTP response.

Internet Server must be made restriction to access customer quantity.In the data of an Internet service management, not all data can limit size; Especially share or the software shared system for multimedia file, these data are with respect to much bigger through the webpage of strictly regulating, can't the people for dwindling.Face this situation, server have to take the mode of limiting access client quantity to reduce the pressure that comes from these transfer of data, as strict limiting access Thread Count or forbid new user capture when the user reaches certain quantity.

Internet Server also needs to consider the physical distribution problem.For an Internet Server, the problem that another one influences internet access efficient is the network site between client and the server end.This network site is determined by network bottom layer router distribution situation.The distribution of these underlying device makes can pass through different physical pathways when connecting between the computing node on upper strata, the Internet.Both made and taked rational routing algorithm, can not guarantee any two internodal high efficiency visits.If client is in the relative long network environment of physical pathway with Internet Server, its access efficiency can be very low.At this time can't address this problem by server itself merely, make the node that comes from the different physical locations can both obtain acceptable access efficiency on the different physical locations thereby have only server is arranged into.

By above-mentioned discussion, can recognize the basic operational mode of current internet server and the many deficiencies that face.Because this leading position on current internet of the Internet, these defectives have restricted the development of internet, applications.The present invention is by transforming the basic operational mode of Internet Server, make Internet Server become a peer server, reach the purpose of fully rationally utilizing server resource, thereby improve whole the Internet operational efficiency, help the generation of novel internet, applications.Now the subject matter with the current internet server briefly is described below.

There is the problem of the wasting of resources in the current internet server.The current internet server is because the problem of distributed frame, dealing with possible extensive visit, except assurance high-quality on scheduling of resource and the allocation algorithm, generally need to increase great amount of hardware resources and be arranged in the corresponding network environment, thereby deal with the problem that the peak access time occurs.Because it is low that the peak access time accounts for total visit capacity ratio usually, therefore a large amount of computational resources that drop into are not fully utilized, and have formed the wasting of resources feature in the division center system.Meanwhile, client causes bandwidth availability ratio low poor with application system performance owing to being limited by the bottom route results.In other words, the idle and this contradiction situation of systematic function difference of computational resource can appear in the current internet server.More reasonable manner should be a server under resource situation more than needed, and slack resources is initiatively contributed client to poor performance, the node that does not make full use of as bandwidth.The another kind of the wasting of resources takes the form of the current internet server and only just can be used when having possessed client desired data resource or Service Source; And if certain Internet Server is not when having corresponding resource, and do not mean that this server to client without any help because its idle bandwidth, calculate can and storage resources etc. all can be utilized.It is irrational ignoring such slack resources based on the application of current internet server.

The current internet server exists the utilization of resources not have the problem of selection.One of current internet server resource waste feature reason is exactly that server self does not have initiatively to select, have only abilities to be visited such as passive.Even the current internet server in the peak access time, though the computational resource of self has been fully utilized, exists a large amount of clients of waiting for temporarily to can not get resource.And some nodes in these clients may be higher with the joint efficiency that server is set up, and they can obtain service very soon and leave, but take the server computational resource longways; Enjoying that some can take the server computational resource relatively longways because quality of connection is not high in the client of service.This also is a kind of form of expression of server resource waste.But existing server can not make one's options in these clients, thereby makes resource utilization higher.So-called initiatively select, be exactly server according to own resource situation and client state, carry out scheduling of resource and distribution.In data-oriented internet, applications, actual connection bandwidth is the leading indicator that is used for weighing client state.As a rule, Priority Service connects reasonable distribution and the use that the high node of bandwidth helps computational resource.Certainly, according to the connection bandwidth client being selected simply is not best mode.The present invention has done detailed design and realization to the selective power of server.

The problem that the current internet server exists resource pressure not have to share.The current internet server is when the request of customer in response end, and main calculating pressure all concentrates on himself.Though client can utilize buffer memory to slow down server stress, in case Data Update, the effect of buffer memory has not just had.This situation also is the drawback that central distribution formula structure is brought.If there is not suitable resource pressure to share mechanism, the scale of an internet, applications will be restricted in other words that cost can improve.

The current internet server is a kind of non-streaming server.So-called streaming server requirement permission client keeps the connection status with server for a long time, and data interrupted transmission in the meantime.The current internet server can only be accomplished this point limitedly, and very responsive to the variation of system scale.When scale surpassed restriction, this limited stream will be affected or require client to wait as long for.In the middle of current most of internet, applications, the connect hours of client and server all is limited, and transfer of data also is to carry out with the non-manifold formula that the lightweight data and the relative long time interval are feature during connecting.

Summary of the invention

The objective of the invention is to current Internet Server is transformed into peer server, make its computational resource can fully and reasonably contribute to internet system, thereby improve the Internet whole work efficiency, promote the development of large-scale non-central Internet internet, applications.Peer give Internet Server can be initiatively and client get in touch, and need not passive wait.This variation makes Internet Server under the situation that resource is had more than needed, the active assistance client; And under the situation of resource anxiety, can help client selectively, thereby reach distributing rationally of resource.

The first step makes Internet Server move with Peer.It owing to division center one of reason that produces current internet server subject matter, the present invention at first changes this, promptly change the feature of the simple customer in response end of Internet Server visit, server itself also possesses initiatively and client connects and transmit the ability of data.Such change is not what isolate, and it must be accompanied by the whole service environment and take place similarly to change; Otherwise, can't help the raising of systematic function.When Internet Server moved with Peer, its client of facing also should be moved by this way, and promptly all computing nodes all operate in the noncentral distribution formula environment with Peer.Even might not mean the optimized dispatching and the configuration of system resource like this, appropriate strategy and scheme are only key.If but the distributed environment of the non-central equity of neither one, the rational resource occupation mode that more becomes can not occur yet.The Internet Server of Peer operation is a prerequisite of improving the scheduling and the distribution of Internet Server resource.

In second step, reduce the requirement of Internet Server to resource distribution.After having set up reciprocity environment, the relation of current server and client has been changed fully.The importance of server in whole system has also descended, and promptly the resource that not exclusively is decided by server and is provided based on the application performance of extensive reciprocity the Internet is calculated the scale that participates in node and depend mainly on one.Opposite with current internet, the scalability of large-scale non-central Internet Internet application system is strong, and system scale is subjected to any restriction hardly; Just in small scale, deficiency can appear in systematic function, thereby server is produced dependence.As server, be fully operational in the different computing environments system's dedicate resources.Usually have only the calculating meeting of small scale that server resource is formed pressure; Even and such calculating because the difference of distributed frame, server stress also descends a lot.Therefore, under the prerequisite that reaches the same performance requirement, the peer server resource allocation request among the present invention is more much lower than current server resource distribution.

In the 3rd step, give the ability of Internet Server with the active dedicate resources.The Internet peer server among the present invention possesses the initiatively ability of contribution calculation resource.So-called initiatively dedicate resources is to carry out under the situation that server resource is had more than needed.Server itself have only server self just its resource consumption situation to be understood most, so should have the authority rather than the complete passive wait client requests of arranging resource.When server has needed data resource of client and Service Source, client meeting active and server contact; But this does not also mean that server can only be in passive position.Under the more than needed situation of server resource, server still can be initiatively and client contacts, sets up a plurality of connections, thereby reach making full use of and the stable transfer of data of client bandwidth; For towards the application of calculating, server is the amount of calculation of carrying client end initiatively also.In addition, another important improvement of the present invention is both to have made Internet Server not have required data resource of client and Service Source, and it still can participate in the calculating of client.Such application scenarios is significant to large-scale distributed the Internet.In such system, a large amount of calculating are not based on that the minority server carries out; The computing node that each is weak relatively as personal computer, becomes the dominant role in the system.These weak relatively computing nodes have formed the calculating cluster according to the peculiar agreement of large-scale non-central Internet system.With respect to the computing node a little less than these, server possesses higher computing capability and more computational resource, even more taller than some calculating computational resource summations that cluster had of these nodes compositions.When server did not have the desired data of client, because all computing nodes of comprising Internet Server in the middle of the large-scale non-central Internet the Internet are all with Peer work, server also can become a node in certain cluster.Like this, its not available data resource will progressively obtain in other node from cluster; In case server has had data, it just can go to assist weak relatively node and cluster to obtain data resource with active or passive mode.Similarly, server also can contribute to the Internet under the situation that does not possess Service Source.

The 4th step, give Internet Server and possess selective power, thus more effective computational resource that utilizes.The Internet peer server among the present invention is under the situation that resource own is fully utilized, still can show than current internet server resource is used higher reasonability, promptly limited resources be distributed to the node that server resource holding time weak point is had more helps to whole internet, applications performance raising.During current internet is used, the independent access server of each client and each other without any relation.Different with above-mentioned scene, in the large-scale non-central Internet the Internet, the computing node of visit Internet Server of the present invention has been formed the calculating cluster that scale does not wait in the mode of equity.Simultaneously, as computing capability than the strong a lot of server of ordinary node also as these clusters in the peer node assistance system to improve their performance.But as a rule, the performance of cluster has difference; And this difference mainly causes owing to cluster scale is different.In general, sweeping node has the ability still can efficiently move under the situation that does not need server to help; And the cluster of small scale is just had any problem, and operational efficiency is low.Should at first assist node in the cluster on a small scale as the server that contributes to system; Under the situation of resource-constrained, can not do to help or limited assistance for the node in the sweeping cluster.When assisting the cluster of small scale, also can select concrete node according to historical record or current running status.Selection mode at this time is not to select weak node to help, but selects in the cluster of small scale and the good node of server joint efficiency high-performance.Because the weak slightly nodes of these nodes and other have been formed the calculating cluster, obtaining the help of server when these nodes after, they can be assisted, and other do not obtain the node that server helps in the cluster.Because each node all has selfish feature in the large-scale non-central Internet the Internet, promptly each node is unwilling initiatively to help other nodes after obtaining service; This can cause to the help of slightly strong node small and weak cluster being benefited.On the other hand, both made and assisted strong node to help limited weak cluster, preferential to assist strong node also be reasonably, because these nodes can finish resource request to server quickly than weak node, thereby server can be transferred to limited resources on the weak node quickly.Certainly this is a kind of scheduling of resource or selection strategy.According to the difference of concrete application, can also adopt other scheduling strategies, as timesharing etc.In any case, the peer server operational mode among the present invention is different with current server, has possessed selective power in the process of dedicate resources.

In the 5th step, set up the streaming Internet Server.The essential characteristic of large-scale non-central Internet internet system transfer of data is exactly to carry out in the mode of stream, rather than request in the internet system of center and response mode.This stream is that all participation computing nodes form under the support of large-scale non-central Internet the Internet transport protocol in the system.Changed the basic outlook of the Internet with the system of stream mode operation, also made the operational mode of current server that variation has taken place simultaneously.No matter be initiatively dedicate resources or dedicate resources selectively, the peer server among the present invention all is in the transmission of supporting stream and the processing of convection current.

Good effect of the present invention is:

The present invention proposes the notion of peer server, making in the Internet with the dedicate resources is the Internet Server that purpose exists, can appropriately use its computational resource, with active and passive or multiple modes such as selection are arranged to the Internet dedicate resources, significantly reduced the server resource waste, set up primary condition for improving whole the Internet calculated performance.

Description of drawings

Fig. 1. peer server is carried out passive assistance with the request response modes to cluster;

Fig. 2. the weak cluster of peer server active assistance;

Fig. 3. peer server is assisted cluster after the resource instant the acquisition.

Embodiment

The first step is set up the Internet equity point.In the large-scale non-central Internet internet system, equity refers to possess between any two computing nodes not only initiatively to the other side sends connection request but also ability that can passive response the other side request.This variation makes computational resource in the Internet participate in the node that calculates by become all from single or minority node.In such distributed frame, there is not the saying of server strictly speaking.As computing capability will be the computing equipment of target far above ordinary node and with the contribution own resource, so-called peer server.Still it being called server among the present invention is a saying sanctified by usage, and its mode of operation is different fully with traditional server.The foundation of equity point is to increase to prerequisite with the computational resource that current computing node performance generally strengthens, had.Under such prerequisite, it is not only consumption to system resource that any computing node enters system, and also is the contribution to system resource.Based on this thought, distributed frame must become non-central formula by center type.No matter be consumption of natural resource or dedicate resources, computing node all can with initiatively or passive mode and other nodes carry out alternately.Each common computing node all is under terminal use's control in the large-scale non-central Internet internet system, and any one such node all only can satisfy dedicate resources under the prerequisite of self-demand.In common calculation process, a node and other nodes cooperatively interact, and share computational resource each other, satisfy demand separately simultaneously; For this reason, forming above-mentioned equity point is to utilize the computational resource on all nodes necessary.Different with ordinary node, the target that the peer server that is transformed into by server enters system is the computational resource of contribution self, rather than consume system resources.In this case, as if peer server promptly needs only the pattern work with traditional server as long as the request of passive response ordinary node is just passable.If but server has possessed the ability of the request of initiatively sending, it can initiatively contribute the computational resource of self when ordinary node is not asked.Since ordinary node only certain node possess under the prerequisite of data resource or Service Source just can and its contact.Peer server is not to possess such data resource or Service Source at any time; Initiatively participate in calculating to possess such data resource or Service Source rapidly but the strong computing capability that it had can make it, thereby become the utilizable node of ordinary node.Secondly, peer server also can be carried out the reasonable resource scheduling according to the concrete condition of correlation computations, and needn't respond the resource request of all nodes under the prerequisite that possesses the active interaction capabilities.In brief, in the large-scale non-central Internet the Internet, when node moves in the mode of equity point, for fully rationally utilizing of resource provides precondition.

In second step, formulate the large-scale non-central Internet the Internet transport protocol.Distribution transport protocol is that the uniqueness according to its computing environment proposes.After the restriction of having broken away from the division center the Internet, new host-host protocol is effectively to utilize computational resource necessary under new the Internet architecture.For ordinary node, this party to the agreement's face will guarantee it to the elementary cell integrality that a plurality of heterogeneous datas are formed, and will adapt to its feature efficent use of resources on the other hand and guarantee that a plurality of heterogeneous datas form efficient stream.This agreement also will make full use of the computational resource of peer server, guarantees that weak current can access reliable support.When realizing this agreement, be based on existing non-central non-flow data host-host protocol, as BT[9], BiToS[10] and BASS[11].But these agreements just considered single character data in the non-headed construction system transmission problem and the perfect desired target of host-host protocol fall far short.In a large-scale non-central Internet the Internet, a perfect host-host protocol will solve following problem.

(1) monitoring stream has the variation of computing node quantity.In the large-scale non-central Internet internet system, can find computing node according to the feature of social network system.But because the transmission of stream is a long relatively process of a time, the number of computing node is also ceaselessly changing in this process.This variation or because the user normally adds or leaves to cause, or because the taste of convection current causes.In issuing process, because stream carries out in a particular cluster, as long as the current line node of this group is enough, stream just can form.But form after the stream, can not guarantee that this quantity is constant always.Under the ideal situation, this quantity can increase.This may be because new node adds formation in the group, also may be to cause owing to the expressed implication of stream is welcome to recommend to cause.If this recommended amount is big, one will be disseminated in whole large-scale non-central Internet internet system rapidly at valuable stream on the implication so.Can think that thus valuable circulation does not often need to worry the problem of its computing node number on the implication.Concerning issue, the quantity of stream is also few so usually.Certainly may have such situation, promptly a lot of users reorganize the stream that forms oneself to valuable stream, cause valuable fluxion amount to increase; Yet this simple imitation can make the value of stream descend gradually.More general situation is, the stream influence power that much is published is flat, and the flowmeter operator node is shared in the group at most; As long as group's scale is not too little, stream still can form.During the worst situation, stream is received the user and interrupts in issuing process, and this number of interruptions to account for whole computing node ratio big so that remaining computing node is not enough to keep the issue fully of stream.The situation of common this interrupt flow can occur after having browsed a period of time in the group in member's convection current.At this time host-host protocol need provide support to the residue computing node.In when search, flow needed computing node and come from the group or outside the group; Its quantity is decided by the popularity degree of searched stream.In addition, can require the node relevant with search node to add during owing to search, this two parts node has formed the support to search stream.Usually search stream is had the many drift net objects of computing node and is come the front by popularity degree (amount of the being replicated) ordering according to stream, comes the back and have the few node of computing node quantity.Can occur flowing in the incipient stage relatively stable and be worth the trend that descends and cause computing node to reduce along with the minimizing and the stream of replica node thus at back segment.At this moment need host-host protocol to provide support, guarantee the smooth transmission of stream.When visit, host-host protocol can require visitor or interviewee's neighbor node to become the computing node of supporting stream; These nodes also are among the different groups.According to the knowledge of community network, the interviewee that can become super node in the middle of a group seldom and most visit concentrates on these interviewees.Therefore, as a rule, the stream that forms based on visit can find enough computing nodes.And small part visit ordinary node and the stream that forms needs host-host protocol to provide support.

(2) weak current is supported.Because the intensity of stream in the large-scale non-central Internet the Internet is to be in the middle of the continuous change, costlyly fail to be convened for lack of a quorum very popular and intensity stabilization, being worth low stream then can be by dying down by force, so need consider the support to weak current in agreement; This is the requirement of system's fairness.This help to weak current can be taked dual mode, promptly supports by the super node and the more stream of the searching node of specific placement.The support of super node is similar to the mode among the BASS.Weak current generally is to participate in the few relatively stream of node, and these participation nodes are not isolated the existence.Can obtain more computing node by its position at community network.Here there are system recommendation and the artificial dual mode of recommending.No matter be the sort of recommendation, all be the stream node of seeking in community network current influenza interest exploratoryly.In ideal conditions, this searching can bring stable computing node support, has promptly found a plurality of computing nodes to this influenza interest finally.Under the poorest situation, agreement is difficult to search out in community network to the interested node of this weak current, has to seek the super node of specific placement thus and helps.Because the large-scale non-central Internet the Internet is a resource sharing system in essence, most system actions all are to find the pairing demand of similar behavior in community network, so common this situation probability of happening is not high.In a privately owned environment, this situation can be very general; But not to be the present invention will pay close attention to for this.After this situation took place, the super node of specific placement can be carried out effectively this stream and help.

(3) in the transmission course of stream, take relevant scheduling strategy according to data character.After the formation that has enough computational resource supports stream, when concrete transport stream, to consider that the file scheduling strategy is with efficiency of transmission that improves stream and the performance of improving receiving terminal system.Scheduling strategy puts forward at data character, size in the stream and to the influence that presentation layer causes.Stream among the present invention comprises a plurality of data units that formed by multiple heterogeneous data, can be called the drift net object; The having different transmission and show requirement of these data; The size differences of these heterogeneous datas is also very big in addition.At these differences, when the transmission data, should take corresponding strategy to make and make full use of under the prerequisite of computational resource, can consider above-mentioned different factor, take different transmission policys to guarantee that the user obtains more satisfied bandwagon effect.For a concrete stream, its data of description is the most important to whole transmission; By analysis to data of description, can obtain the basic condition of whole stream, show still that for transport stream stream all has directive function; And data of description all is the lightweight data little to system pressure usually.Based on above-mentioned cognition, the data of description of stream will at first be transmitted.Secondly, comprise some text datas again in each stream, these data are that the implication of expression stream is necessary; And these data can not pose a big pressure to transmission yet.Usually after the transmission data of description, can transmit the text data of each drift net object.This transmission can be finished very soon.After the data of description end of transmission and stream Chinese version data began transmission, system just can be the living view roughly of the miscarriage of transmitting at presentation layer, make user's convection current included in have preliminary understanding, and carry out limited interactive operation.Subsequently, system begins to handle the transmission of heavyweight data, such as picture, Voice ﹠ Video.At this moment transmission policy focuses on according to intrinsic relation between the drift net object, and the remaining heavyweight transfer of data of each drift net object is put in place.Wherein picture transmission and audio frequency and video are inequality again for these.Picture size is generally also less than audio frequency and video, and picture does not have the requirement of sequential.In scheduling strategy, picture has the priority higher than audio frequency and video.At last, because the large-scale non-central Internet the Internet has high dynamic feature, require a plurality of drift net objects are transferred in the short as far as possible time on the stream node; But a plurality of drift net objects have size differences, and this difference may cause when transmission system resource allocation unreasonable.When the big order of drift net object in stream differs too great disparity prior to a little drift net object and this size, if will cause small data to wait for the phenomenon of big data according to the intrinsic sequential delivery of drift net object; And the behavioral characteristics of system might make the small data of back lose after big transfer of data finishes.This just requires to consider time-sharing operation in scheduling strategy, promptly transmit the period for each drift net object in the stream all provides in the unit interval section, avoids occurring the unreasonable of system resource allocation.

(4) determine the relation between the drift net object in the stream.Be sequential between the drift net object in the stream, and this ordinal relation is can not be destroyed at presentation layer, causes the user when long-range receiving and stream inconsistent at source node.But this relation does not also require that transmission strictly observes the time, and key is can look consistent with the stream on the source node just passable when showing.This relation is actually a kind of structure of knowledge of artificial regulation.From topological relation, can be linear, tree-shaped with netted.But no matter which kind of form all requires host-host protocol to guarantee to duplicate with same form on long-range stream node, thus reflection stream creator's original meaning intactly.Except the topological relation of static state, the author of stream also must specify the dynamic relationship of stream, i.e. context of showing between the drift net object.For randomly topologically structured, this is equivalent to indicate the traversal order of stream.In issue and access process, must guarantee to flow with this traversal order and on the stream node, show.And in search procedure, because not only node of source of stream, can reformulate new static state and dynamic order according to relevant informations such as searching order standards, i.e. search is equivalent to reconfiguring knowledge.

(5) data character is taked related transmission method in the analysis stream.Because brand-new definition has been carried out in convection current of the present invention, thereby there is the several data of different in kind in the stream.Thereby host-host protocol of the present invention must have the ability to differentiate data of different nature and take different transmission meanss.In the large-scale non-central Internet the Internet, mainly need identify whether the heavyweight data have sequential in transmission course requirement.Can take the BT algorithm to finish for the data that do not have sequential to require, and the data that have sequential to require take BiToS algorithm and BASS algorithm to realize transmission.

In the 3rd step, set up large-scale non-central Internet the Internet dynamic calculation cluster.The present invention has taked to set up the way of high-efficiency dynamic cluster and has disperseed computational resources to be organized into a plurality of clusters these, in each cluster computational resource can obtain fully shared, thereby finish corresponding calculated.What should be noted that a bit is, because each node all is to be controlled by relative users, such cluster is in the middle of the high dynamic change, and promptly each node all may add at any time, also may leave at any time.This brings difficulty for setting up the efficient calculation cluster.The present invention has adopted the efficient operation that data are duplicated in complex networks system [12], netted multicast is made [13] mutually, dynamically screened node and guarantee cluster based on the routing algorithm means such as [14] of Complex Networks Theory [12].The large-scale non-central Internet the Internet transport protocol is set up with regard to being based on above-mentioned theory.Behind the host-host protocol of having formulated at the large-scale non-central Internet the Internet, the node in the system will form a plurality of calculating clusters that scale does not wait.For peer server among the present invention or super node, its at computing environment be not mutually independent irrelevant each other client, but the resource-sharing calculating cluster that each takes what he needs each other.Host-host protocol is exactly the key mechanism that impels cluster to form.In the large-scale non-central Internet the Internet, have only according to concrete application characteristic to form the calculating cluster, thereby being transmitted in the mode of stream, data form new internet, applications.

In the 4th step, judge that the large-scale non-central Internet the Internet calculates the cluster performance state.After large-scale internet calculating cluster formed, a key issue was to judge the performance state of these clusters.Have only the performance that correctly monitors cluster, the peer server among the present invention could be assisted to finish calculating efficiently it.Usually the performance of judging a cluster can be according to following index.

(1) monitoring cluster scale.Cluster scale is to judge a straightforward procedure calculating the trunking efficiency height.In the middle of the large-scale non-central Internet the Internet, form the calculating of high-performance based on stream, it is necessary that the node of sufficient amount adds, unless it is little to flow the data volume that is comprised in the related application.All be based on this prerequisite of heavyweight data for main application in the large-scale non-central Internet the Internet, if there is not abundant node, one to calculate cluster normally inefficient.This be since any node that participate in to calculate all must with the webbed topological structure of other groups of nodes, finish the formation and the correlation computations of stream thereby each takes what he needs.In a high dynamic the Internet, find the node that can efficiently cooperate and be not easy.If it is very few to participate in the interstitial content of a calculating, wherein arbitrary node can find the chance of appropriate node just to descend comparatively speaking, thereby causes whole calculated performance based on stream to lower.Therefore, can roughly judge the performance of a cluster by cluster scale.Yet this wherein also has problem to exist.In the middle of internet, applications, the situation that small number of nodes is formed HPCC might appear, such as in local area network (LAN) of they coexistences, network distance is short or wherein certain node possess superpower computing capability.Judge that by cluster scale its performance will have error this time.In addition, can be protected relevantly, can't provide a preset parameter with concrete application when cluster scale reaches its performance of great degree; This parameter can be called the qualified parameter of cluster.Little and the less demanding stream application of sequential for data volume, scale does not need very big; For big data quantity but sequential is irrelevant, it is big that scale is wanted; Big and sequential is responsive for data volume, as Streaming Media, at this moment higher to the requirement meeting of scale.Based on above-mentioned discussion, when judging the cluster performance, under this prerequisite of same application, carry out according to cluster scale; Different application, the qualified parameter of cluster is different.But even so, because the network behavioral characteristics, also can only provide a relative performance comparative result usually, promptly scale all under the situation less than the qualified parameter of cluster sweeping cluster performance may be higher than the low cluster of scale; All under the situation greater than the qualified parameter of cluster, scale is inoperative to the cluster performance in scale.This relative value may cause that server resource contributes to does not need the cluster supported.Though, there are the problems referred to above in cluster scale as the index of judging the cluster performance, but since cluster scale obtains easily and in most of the cases cluster scale systematic function is played decisive action, it is also very convenient to gather this numerical value, being used as the cluster performance index can adopt.

(2) monitor the ratio that spreads.Speed that flows in the cluster and streaming media playing speed ratio.As the index of judging the cluster performance, cluster scale is a relative indicatrix just, and is not accurate enough.In order to remedy this deficiency, can utilize the speed that flows in the cluster and flow data playback rate ratio as index.In the large-scale non-central Internet the Internet, the speed of stream is decided by the efficient of respective cluster, but whether this efficient can satisfy the requirement of system, just needs an index and weighs.In the middle of stream, there is multiple heterogeneous data; The speed of stream medium data convection current is the most responsive.When flow rate satisfied the speed that streaming media playing requires, Streaming Media can normal play, and overall system efficiency just reaches requirement; When flow rate was lower than the speed of streaming media playing requirement, streaming media playing can be stagnated or have some setbacks.So, if the ratio that spreads (speed that flows in the cluster and streaming media playing speed ratio) can judge that greater than 1 o'clock this cluster performance meets the demands; Ratio is less than 1 o'clock if spread, and judges that then this cluster performance is discontented with sufficient requirement.The ratio that spreads is an absolute value, can appropriately express the performance of a cluster.But its problem is that the ratio that spreads in the cluster may be different, and different nodes has the different ratio that spreads, and need do a comprehensive judgement; Secondly, gather that these data need and all clusters in the node contact, it is big that cost is wanted; In addition, the ratio that spreads in addition is dynamic change, needs to use the mode of appropriate image data; At last, the stream in some cluster does not comprise stream medium data, and the ratio that at this time spreads is just inoperative.

(3) monitoring node is replaced frequency.In the large-scale non-central Internet the Internet, calculate and to be based on all that cluster carries out; And cluster itself is not a static node set.All nodes all are in the middle of the continuous variation, and this variation is the inevitable outcome that network fluctuation and node fluctuation bring.Each node that still is in the cluster must adapt to this behavioral characteristics: when the fluctuation of network fluctuation or node results in stream that certain node forms when hard to carry on, must searching may replace the alternative node of current cooperative nodes.In the middle of the host-host protocol of large-scale non-central Internet the Internet, considered this problem.This replacement through and existing cooperative nodes relatively and new cooperative nodes relatively and relatively and suitably just final decision after a series of means such as delays of historical record, i.e. node replacement is a careful behavior in the large-scale non-central Internet the Internet.Careful reason is to find an appropriate nodes cost big in a large-scale cluster; Unless confirm that by multiple monitoring its efficient is low excessively, generally can not replace.When replacement occurrence frequency such in the cluster is too high, illustrate that the node in this cluster is also too high for the extra cost of selecting node to pay, and this cluster dynamic is strong excessively, plays pendulum.Calculating is difficult to carry out in the middle of the frequent cluster that takes place to replace, and stream also is difficult to form; Reason is that a lot of nodes in the cluster can not find lasting cooperative nodes.Therefore, by node being replaced the statistics of frequency, can tentatively judge the performance quality of a cluster.This index has overcome this shortcoming of existence that the ratio that spreads must rely on stream medium data.But similar with the ratio that spreads, replace the data acquisition of frequency and also gather difficulty than cluster scale.Also have a deficiency to be, the difference that node is replaced frequency has also just reflected the relative value of performance difference between the cluster, and not exclusively reflects the absolute situation of cluster internal calculation performance.In addition, judge that with utilizing cluster scale the cluster performance is similar, can not provide concrete replacement frequency values and judge the cluster performance, because this value is diverse to stream of different nature.

(4) data sync difference between monitoring node.One of large-scale non-central Internet internet system and current internet are significantly different is that data are with the form transmission of stream; And the formation of stream is to depend on the formed cluster of node.When a cluster is in when moving under the high performance state, cooperate closely between the node, stream can transmit in whole cluster relatively swimmingly.Because what form between the node is a kind of netted effective multicast model,, can keep synchronism in other words a little less than the asynchronous nature between the node substantially if the cluster performance is good.If the synchronism of the inner stream of cluster is poor excessively, illustrate that so fluctuation is excessive in the cluster, cooperate efficient low between node, the cost that whole cluster is used to keep the trunking efficiency cost is big.Opposite extreme situations be when synchronism when node can not remain within the tolerance interval, cluster can be split into a plurality of according to synchronous difference; This can cause cluster scale to reduce, thereby the cluster performance of division further is affected.In a word, by data synchronization difference between acquisition node, can judge the performance of a cluster.And synchronism difference has nothing to do with concrete the application, and promptly this is the absolute value of reflection cluster performance.The another one advantage is that synchronism difference does not rely on stream medium data yet.Weak point is to gather these data than cluster scale complexity equally.

In the 5th step, set up the node tracking device and be used for the variation of monitoring system performance index.Peer server among the present invention can effectively be worked, at first will basic understanding be arranged whole large-scale non-central Internet the Internet.This task is finished by the node tracking device.Because need obtain the overall situation of node, so similar to server, this node is with the center type arrangement works.All nodes that are present in the large-scale non-central Internet internet system all will be in back and the node tracking device contact of changing, and the node tracking device when variation that node needs also can take place on node active get in touch with multicast mode and node.Mutual and concrete application and the present invention between node and the node tracking device judge that the index of cluster performance has relation.Concerning most of large-scale non-central Internet internet, applications, all be to be essential characteristic with heterogeneous heavyweight data.At this time the node tracking device need be followed the tracks of the work in the whole system, and main data of gathering are as follows.

(1) stream of all existence in the acquisition system.A stream has one section life cycle: stream is at first generated by a node; Progressively have node to add the transmission that becomes a mandarin, up to a peak, i.e. the formation that has a great deal of nodes convection current at the animated period of stream helps; Afterwards, along with the node that obtains all flow datas increases, the node that participates in flow transmission reduces gradually, and until there not being node to participate in, promptly stream is withered away.The node tracking device wants the whole life of convection current to monitor.

(2) cluster in the acquisition system and scale thereof.The life cycle of convection current is monitored, and is of equal value in fact with to the monitoring of cluster.As a rule, the power of stream is accompanied by the ups and downs of cluster scale.Therefore, the life cycle to the monitoring meeting convection current of cluster scale has a clear assurance.After any one new node adds the transmission that becomes a mandarin, all want proactive notification node tracking device; When a node withdraws from the transmission of stream, also need to notify the node tracker.To be that node is improper in the stream withdraw from a problem that exists, and the node tracking device just can in time not obtain the situation of this node like this.This problem can be assisted by other nodes.Because any one node does not isolate in cluster, need to reticulate the multicast structure with other groups of nodes.After improper withdrawing from of node, certainly will influence other with its mutual node.These nodes can proactive notification node tracking device after finding that certain node is responded its request, has so just solved the improper contingent inconsistent problem that withdraws from.

(3) gather the cluster state.Except cluster scale is monitored, according to the requirement of peer server, also need to gather, as the ratio that spreads, node are replaced between frequency and node data sync difference etc. to some other parameter.The collection of these data mainly comes from the proactive notification of node in the cluster.Because these data variation are much more frequent than cluster scale, therefore can cause big pressure to the node tracking device.At this time, the way that can take a plurality of node tracking devices are set solves; But this scheme can be brought inconsistent problem between the tracker, thereby must be by setting up promptness and the consistency of rational multicast mechanism maintenance to system keeps track betwixt.

Except gathering related data, the node tracking device also have an important function be exactly with node and the present invention in peer server (reciprocity super node) carry out alternately.Must form cluster between the node just understands the transmission of enhanced flow and improves the performance of corresponding calculating; But owing to be in released state physically between the node, each node can only the perception localized variation, and this can limit effective growth of cluster scale.At this time, node can be sought from the node tracking device and newly add the node that becomes a mandarin, and attempts setting up the relation of netted multicast with them.By the optimum organization that this and new node carry out, the cluster performance can effectively improve, thereby the transmission of stream also is enhanced.In addition, the node tracking device also will carry out alternately with the peer server among the present invention, and server can obtain the cluster state above the node tracking device; These states carry out work for direct server decisive action.At last, the node tracking device can be treated server usually as a node that can add any stream; Like this, the node in the cluster will all have an opportunity to obtain the support from this peer server, and this has remarkable help to improving the cluster performance.

In the 6th step, peer server is taked the relevant work strategy according to different situations.After the cluster performance that node is formed had had judgement, the peer server among the present invention just can help weak cluster targetedly, thereby improves the performance of whole large-scale non-central Internet the Internet.The working strategies of peer server has following mode.

(1) assists with the pattern of passive response based on request.The request response modes is unique working strategies of traditional server; Peer server among the present invention also possesses such ability.The a plurality of clusters that in the large-scale non-central Internet the Internet, have great deal of nodes and form between them; Yet in the middle of all these nodes, peer server has abundant computational resource.If each cluster can both obtain the auxiliary of it in its starting stage, the transmission of stream is strengthened; In the final stage of cluster, scale can reduce gradually equally, and at this time the effect of peer server also clearly; Only be in and descend even need not the help of peer server to the dependence of peer server in prosperous at cluster.At any time, all should be to the authority of any node in the cluster with the request peer server.But this ability can not allow node abuse again, especially is in the performance height and peer server is under the state of resource anxiety at cluster.In the present invention, peer server is notified the node tracker its other resources user mode, and is responsible for distributing peer server to give ordinary node by the node tracking device.Have when the peer server resource under the situation of free time, the node tracking device initiatively to ordinary node notice peer server idle state, impels ordinary node that it is conducted interviews; When the peer server inadequate resource, the node tracking device can be hidden peer server, and ordinary node just can not have been asked it like this.Fig. 1 represent peer server with the request response modes cluster is carried out passive assistance.

(2) peer server active assistance strategy under resource situation more than needed.The resource that peer server had among the present invention always is in a dynamic changing process; This dynamic process is relevant with current application.If peer server resource under the situation of passive response is not fully utilized, it can the mode of taking the initiative assist respective nodes or cluster.Method commonly used is according to the monitoring result of node tracking device for the cluster state, selects weak cluster to assist.Secondly, in this process, peer server initiatively to the open access resources that allow of ordinary node, as increasing the connecting line number of passes of setting up simultaneously more.For data-oriented internet, applications, the Thread Count increase that connects simultaneously can make the utilance of node bandwidth improve usually, thereby strengthens corresponding internet, applications performance.For the internet, applications towards calculating, peer server can initiatively be born more amount of calculation.Why can form peer server resource situation more than needed, one of reason is that peer server generally all can be done suitably restriction to it when the request of passive response node resource, is taken by the high cluster of calculated performance to prevent resource.During this time, peer server should be initiatively to all by the open source request restriction gradually of the cluster of its support and poor performance.This process can carry out progressively rising to the best or peer server resource exhaustion up to the performance of these clusters constantly.Fig. 2 represents the weak cluster of peer server active assistance.

(3) active assistance strategy under the situation that does not have the required direct resource of node.Even do not have in peer server under the situation of data that node needs or Service Source, peer server also can be accepted the resource request from node.Certainly, at this time requested resource is not direct data resource or Service Source, but the resource of other the Internet necessity that peer server had such as computational resource, bandwidth resources and storage resources etc.In this case, the work that at first will do of peer server is copy data resource or Service Source.Only after having had the node resource requirement, peer server could really be brought into play the effect of assisting cluster and respective nodes.Because peer server is based on rare priority principle [9] and duplicates carrying out data, its can be fast with assisted cluster to form to cooperate, reach the purpose of raising cluster performance.Peer server among the present invention is mainly served the transmission of flowing in the large-scale non-central Internet the Internet.In the large-scale non-central Internet the Internet peer server can and a plurality of node set up simultaneously far more than the connection number between the ordinary node, thereby it can obtain enough flow datas quickly and helps relevant weak cluster like this.Work by forefathers, rare in this case priority principle all is effective to sequential sensitivity or the multicast of insensitive heavyweight data, so peer server can obtain to be assisted node to have the stream resource of needs when replication stream very soon, thereby forms the strong cluster that can cooperate.After peer server has enough stream resources, the strategy that can take the initiative and assist and select assistance.At this time one does not have the peer server of the required direct resource of node just to be converted to a peer server that possesses direct resource and can assist node with active and passive and selection mode.Fig. 3 illustrates that peer server obtains after the resource cluster to be assisted immediately.

List of references:

【1】Renu?Tewari；Architecture?and?Algorithms?for?Scalable?Wide-Area?Information?System，PhD?Dissertation，University?of?Texas?at?Austin，1998

【2】Eric?Dean?Katz，Michelle?Butler，Robert?McGrath；A?Scalable?HTTP?Server：the?NCSA?Prototype，Volume?27，Issue?2，November?1994，Page(s)：155-164

【3】NCSA?Mosaic；ftp://ftp.ncsa.uiuc.edu/Mosaic

【4】Amibar；Smart?Thread?Pool；http://www.codeplex.com/smartthreadpool

【5】Douglas?E.Comer；Internetworking?with?TCP/IP-Principles，Protocols?and?Architecture，ISBN?86-7991-142-9

【6】HTML?4.0Specification；http://www.w3.org/TR/REC-html40-971218/

【7】Hypertext?Transfer?Protocol；http://www.w3.org/Protocols/

【8】Paul?Francis；Is?the?Internet?Going NUTSS？IEEE Internet?Computing，November/December?2003，Pages：97-99

【9】Cohen?B.；Incentives?Build?Robustness?in?BitTorrent；in?Workshop?on?Economics?of?Peer-to-Peer?Systems，Berkeley?USA，May?2003

【10】Vlavianos?Aggelos，et?al；BiToS：Enhancing?BitTorrent?for?Supporting?Streaming?Applications；INFOCOM 2006，25 ^th?IEEE International Conference?on?Computer?Communications?Proceedings，April?2006，Page(s)：1-6

【11】Dana?C.，Li?D.，Harrison?D.，Chuah?C.N.；BASS：BitTorrent?Assisted?Streaming?System?for?Video-on-Demand，Multimedia?Signal?Processing，2005IEEE?7 ^th?Workshop，Page(s)1-4

【12】Newman?M?E?J.；The?Structure?and?Function?of?Complex?Networks，SIAM?Review，2003，45，Page(s)：167-256

【13】Obraczka，K.；Multicast?Transport Protocols：a Survey and Taxonomy，?Communication?Magazine，IEEE，Volume?36，Issue?1，Jan.1998Page(s)：94-102

【14】Hari?Prasad?Thadakamalla；Decentralized?Algorithms?for?Search?and?Routing?m?Large-Scale?Networks，PhD?Dissertation，Pennsylvania?State?University，December?2007

Claims (8)

1. the reciprocity service system establishing method for large-scale internet application dedicate resources is called for short peer server.It is characterized in that described peer server moves with Peer; Described peer server is efficiently moved with low relatively computational resource; Described peer server possesses the initiatively ability of contribution calculation resource; Described peer server is the contribution calculation resource selectively; Described peer server sends data in the mode that continues to flow.

2. the reciprocity service establishing method for large-scale internet application dedicate resources the steps include:

1) adopt the tcp/ip communication agreement to make on the Internet any computing node promptly set up the Internet equity point except possessing the ability that the ability that accepts request also possesses the request of transmission;

2) adopting to continue to flow is the basic mode of the transmission means of feature as above-mentioned equity point interaction data;

3) in internet, applications, set up based on the dynamic cluster under corporate data and the demand for services prerequisite;

4) gather the various status flags of cluster, judge the current performance of respective cluster;

5) use the node tracking device to gather each cluster status data of current existence;

6) according to the different characteristic of cluster, described peer server is taked the relevant work strategy.

3. as described in the claim 2, it is characterized in that using the method for setting up the server and client side in the ICP/IP protocol, give the server and client side for the computing equipment on any the Internet, make it to become reciprocity point, and utilize the computing equipment own resources to tackle visit as much as possible or request.

4. as described in the claim 2, it is characterized in that data can be transmitted based on TCP/IP in mobile mode before control equity point user stops between equity point.

5. as described in the claim 2, it is characterized in that guaranteeing under the prerequisite that the demand data of each reciprocity point own satisfies, the help of transfer of data is provided for other equity points in concurrent mode.

6. as described in the claim 2, it is characterized in that judging the performance of calculating cluster that its method is according to concrete technical indicator:

1) utilize centralized servers to gather the number of equity point in each cluster, the variation of monitoring cluster scale;

2) utilize centralized servers to collect in the cluster speed and the stream medium data playback rate of flow-data in each equity point, and try to achieve the two ratio;

3) utilize centralized servers to collect each reciprocity putting in reciprocal process in the cluster, equity is put the frequency that is replaced on every side;

4) utilize centralized servers collect in the cluster between the equity point with the sequential related data transmission in the synchronism difference that exists.

7. as described in the claim 2, it is characterized in that making peer server to work, thereby obtain the ability of monitoring cluster state, become the node tracking device in the mode of centralized servers.

8. as described in the claim 2, it is characterized in that the state current according to cluster, make peer server take corresponding strategies work, its method is:

1) after the service request that receives common equity point, described peer server is its other resources behaviour in service notice node tracker; The node tracking device carries out resource allocation according to the state of peer server self; When the peer server resource was idle, the node tracking device was initiatively informed common reciprocity point with the peer server idle state, impelled common equity point that it is conducted interviews; When the peer server inadequate resource, the node tracking device stashes peer server, forbids that common equity point is to its request;

2) peer server is under himself resource situation more than needed, for relevant equity point initiatively provides assistance; Peer server is put cluster state of living in according to equity to judge whether decision provides service when passive response equity point resource request; When request equity point cluster performance of living in is normal, refuse this request; Otherwise, provide request desired service;

3) peer server is under himself resource situation more than needed, should initiatively supported and the cluster open source request restriction gradually of poor performance by its to all, up to the performance of these clusters progressively rise to the best or with its resource exhaustion till;

4) do not have under the situation of required data of equity point or Service Source in peer server, peer server can utilize himself computational resource, bandwidth resources and storage resources to take the initiative in offering a hand in common reciprocity point; Peer server will be carried out duplicating of data resource or Service Source based on rare priority principle; After peer server has enough stream resources, can be active and passive or selection mode assist common reciprocity point.

Images