CN104022911B - A kind of contents construction management method of pattern of fusion content distributing network - Google Patents
A kind of contents construction management method of pattern of fusion content distributing network Download PDFInfo
- Publication number
- CN104022911B CN104022911B CN201410299708.1A CN201410299708A CN104022911B CN 104022911 B CN104022911 B CN 104022911B CN 201410299708 A CN201410299708 A CN 201410299708A CN 104022911 B CN104022911 B CN 104022911B
- Authority
- CN
- China
- Prior art keywords
- node
- cost
- flow
- domain
- content
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
A kind of contents construction management method of pattern of fusion content distributing network, belongs to Internet technical field.The purpose of the present invention is to provide two layers of content node deployment mechanisms for ISP CDN, and design index and synchronization mechanism take into account flow settlement cost, data transmission quality to improve the contents construction strategy under CDN environment.Single treatment is carried out for network data size, it is only relevant with the route jumping figure of process once to transmit consumed flow cost, without considering size of data in itself, the interstitial content selected from core routing node set F is inverse relation with repeating flow, and synchronous cost is then directly proportional to interstitial content;Synchronous initial source is inter domain router node, for inter-domain routing node t, covers k content node by first minimum spanning tree TR of the wide searching algorithm generation using t as root, finds k node, RC+SC minimums are realized using the algorithm based on Greedy strategy.
Description
Technical field
The present invention relates to a kind of contents construction management method, belong to Internet technical field.
Background technology
With the continuous upgrading of Internet technologies, the performance of service end-network forwarding equipment-terminal machine is all significantly
The online real-time applications such as degree lifting, such as bandwidth, forwarding rate, disposal ability, HD video transmission have been popularized, and are promoted
Internet flows are doubled and redoubled.Growing flow and structure and server complicated and changeable and client end capacity and money
The asymmetry in source promotes the generation of content distribution techniques, so as to ensure that network edge user can obtain good access body
Test.It is scattered to network using DNS dynamic analysis by setting up content identical server in network edge in the CDN of early stage
Flowing of access, so as to reduce the access delay of user.Routing mechanism in content distributing network mainly covers two aspects:
(1) deployment way, i.e., content proxy server is deployed in which of network position, to meet that edge customer please to service
The service speed asked;(2) contents construction strategy, that is, which kind of contents construction algorithm is used to reduce network load and access delay.
1) deployment mechanisms
The deployment strategy of content distributing network is broadly divided into two classes:Marginalisation deployment is disposed with core domain.Marginalisation is disposed
Thought be that content proxy server is deployed in multiple ISP, so as to reduce the time delay of cross-domain network transmission, this deployment plan
Representative slightly is Akamai, its 78 national 1 in the world, and more than 105,00 servers are disposed in 000 network.
The thought of core exchange area deployment is that large-scale data center is deployed in into high speed fibre backbone network, high by high bandwidth
Disposal ability reduces user's access time delay, and the representative of this deployment strategy is Limelight, and 72 optical fiber are handed in the world for it
Regional deployment data center is changed, fiber direct connection is carried out with more than 578 ISP.
Marginalisation deployment mechanisms are the main flow deployment way of content distributing network, are mainly transmitted in view of global network
Core network congestion phenomenon is not can solve apart from long and conventional cache mechanism, in order to reduce data in net
" in last " (last mile) propagation delay time in network, by resource it is preposition in region closer to the user to being effectively improved
Access experience.Typically, this dispositions method is all to place server in substantial amounts of Network Access Point.To ensure content point
The service quality of hair, content distributing network manager need to safeguard the synchronization and renewal of a large amount of distributed server data.It is overall
For, the problem equivalent is in N number of different M server (N of position deployment>M) consumed with reaching minimum cost, i.e. K-
Median problems, and minimize K-center problems to have proved to be NP complete.
The position deployment strategy of content proxy server is a key technology in content distributing network, and it is directly affected
Distribute service quality, network traffic load and cost price.For ISP, suitable deployment mechanisms can reduce
Flow is repeated in flow settlement cost and domain between its domain, saves bandwidth consumption.Vakali et al. points out content distributing network management
Person generally according to the network type selection deployment strategy where user, is broadly divided into Single-ISP and the classes of Multi-ISP two.Pin
For Single-ISP, 40 content proxy servers are at least disposed in single ISP domain to ensure in geographical position overlying
Cover whole ISP edge networks.Multi-ISP deployment strategy is then to assume content agent in all ISP access point as far as possible
Server, the speed and reliability of content distribution are lifted, the tactful sharpest edges are can be by request flowing of access control
In this domain, so as to reduce flow between ISP domain.
2) routing policy
The one kind of content distributing network as content network, the quality of contents construction strategy directly affect entire content distribution
The performance of network.Contents construction is mainly responsible for asking to be redirected in user by certain routing algorithm by user
Hold on proxy server, and by given resource selection strategy so as to reaching the quick distribution of content.Contents construction mainly divides
For two parts:(1) resource routing algorithm, i.e., by monitoring current network property indices and service load pressure, select
Optimal content proxy server is responded;(2) forwarding mechanism is asked, i.e., service request is turned according to which kind of strategy
Hair.Typical content distributing network resource routing flow is (as shown in Figure 1):
(1) user carries out resource request to request analysis server;
(2) a global performance monitoring service group of planes is distributed in the service of the content proxy services group of planes in internet
Quality and network link various performance parameters;
(3) node selection server monitors the feedback information of service according to real-time performance, and dynamic adjusts selection algorithm and money
Source positioning strategy;
(4) user's request is forwarded to optimal content proxy server by resolution server by Resource orientation result;
(5) content proxy server sends data to user.
Request route belongs to application layer routing policy, and its main thought is according to application layer services demand, to resource request
Forwarded accordingly.Dramatically different with IP layers route is that resource is likely to be present in multiple positions, therefore for same money
The request in source may reach diverse location, and IP layers route what is be then to determine.Request routing algorithm is divided into static and two classes of dynamic:
Static routing refers to that after content distributing network is disposed the request for user is all according to given routing forwarding policy redirection
To content proxy server, the algorithm does not change with network state and service load;Dynamic routing refers to monitor net in real time
Network parameter loads with service end, and dynamic changes forwarding strategy, ensures that current service maintains optimum state as far as possible.
Due to the commercialization of content distributing network, industrial quarters is also more paid attention to resource routing algorithm.Point of Cisco System Co.
Cloth redirector (Distributed Director) is distance in AS (Autonomous System) domain, distance between AS domains
With calculating weights of the end-to-end delay as dynamic forwarding algorithm.Akamai companies carry out excellent in many levels to forwarding algorithm
Change, including content proxy server load, the bandwidth usage in transmission path, current service quality (reliability, delay etc.).
CDN system on heterogeneous networks position has different values, and it is to drop that CDN is disposed such as near terminal
Low server stress, lift the access rate of user, such as akamai;Position deployment CDN can reduce weight in net between in a network
Resurgent amount, reduce bandwidth consumption;Closed the account between the cross-domain position deployment CDN of ISP can reduce domain, such as iCODE;In terminal structure
Backbone network flow can be reduced by making P2P mixing CDN, improve Consumer's Experience, such as pplive.Because storage charges is with well below stream
The reality of amount expense and in order to reduce over long distances, cross-domain access delay promoted the generation of CDN technologies.Many business CDN
Company formulates suitable efficient distribution strategy purpose with data on flows by obtaining network topological information with ISP cooperations to reach.Together
When ISP in order to reduce pay cross-domain flow rate inflow caused by expense, improve constantly service quality and data source in domain,
Also tend to the cooperation with CDN.As the growth of the Internet index of discharge and user have promoted to the demand of network speed
Accuracy, availability, reliability of Internet resources distribution etc. is set to turn into the key issue of Internet technology.Content distributing network is made
To run on the nerve of a covering on whole internet, the service quality optimization being principally motivated in addressing in internet.In the prior art
Do not propose two layers of content node deployment mechanisms for ISP-CDN.
The content of the invention
It is an object of the invention to provide a kind of route management method of pattern of fusion content distributing network, provide for ISP-
CDN two layers of content node deployment mechanisms, design are indexed with synchronization mechanism to improve the contents construction strategy under CDN environment, and
Take into account flow settlement cost, data transmission quality.
The present invention adopts the technical scheme that to solve above-mentioned technical problem:
A kind of contents construction management method of pattern of fusion content distributing network, by three classes of routing node in ISP domain point:With master
The routing node that machine is joined directly together be defined as in edge routing node, ISP domain except edge routing node and inter-domain routing node with
The routing node that outer all routing nodes are defined as core routing node, are connected with routing node in other ISP domains is defined as
Inter-domain routing node, by being disposed to node-classification, and using flow cost as the benchmark of Greedy strategy;Methods described
Implementation process be:
Step 1: node deployment:
Single treatment is carried out for network data size, i.e., 1 time the consumed flow cost of transmission only considers the road of process
Footpath hop count, without considering size of data in itself, and it is defined as below:
Flow cost Cij(Pij):Flow cost caused by transmission data u from node i to node j is most short from i to j
The product of path length and size of data | pij|·sizeu;
The weight w of node ii:The user that node i is administered to node i path sum, i.e.,Represent node i
Shortest path sum of the r user having under its command to node i;
Repeat flow cost RC:When network data Size passes through shortest path P from node iijNode j is reached, if other are used
The access path at family overlaps with it, then data Size caused flows in path P are to repeat flow, i.e. RCij=
size·|Pij|;
Synchronous flow cost SC:Node i to node j synchronous cost for network data Size in PijFlow caused by upper,
That is SCij=size | Pij|;
The interstitial content that is selected from core routing node set F is inverse relation with repeating flow, synchronous cost then with
Interstitial content is directly proportional;It is assumed that the content node number that can be disposed is k, then flow cost is repeated in this time domain with disposing node
Relation is represented with formula 2-1:
Synchronous initial source is inter domain router node, for inter-domain routing node t, passes through first wide searching algorithm (BFS)
Minimum spanning tree TR of the generation using t as root covers k content node, then the formula 2-2 of the cost required for synchrodata Size
Represent:
K node is found, meets RC+SC minimums, realizes flow Least-cost:
Flow Least-cost is realized using the algorithm based on Greedy strategy, its process is:
A. initial phase:Calculate each minimum spanning tree using core router as root node, while obtain the minimum
The repetition flow cost of spanning tree;
B. calculate and repeat flow cost:In the case of nodes are disposed more than 1, select to repeat according to Greedy strategy first
Those spanning-tree root nodes of flow Least-cost, and pass through other each numbers of plies of core router node in every spanning tree
Determine which tree it is stayed in, so as to correct the repetition flow cost of each tree;
C. synchronous cost is calculated:For each k, it is root node to calculate successively by inter domain router, k core routing node
Synchronous cost required for the minimum spanning tree of composition;
Step 2: structure index structure, using dl-RCBF data structures as the index method of synchronization;
Step 3: setting synchronization mechanism, synchronization mechanism is used for paroxysmal caused by index information is broadcasted in reduction
High capacity and network traffics, it is additionally operable to reduce the frequent access for same resource.
In step 2, the process for building index structure is:
Assuming that set U={ key0,key1,…,keyn, | U |=n, and whole hash tables are divided into d sublist;It is if single
Hash functions meet:Hash (key) is made up of two parts, i.e. [B]dAnd remainder, it is preceding
Person will be determined in the bucket loaded in d sublist of remainder partial insertions in that minimum table, the latter remainder
Itself.
Increase count and record after remainder;Count effect is counter, for describing hash collision
Number;Record is used to describe the node location where the key, then whole dl-RCBF spaces are n+m (F+C+K), and wherein n is
Key number.
The implementation process of step 3 is:
By triggered synchro mode, shared resource indexes between servers, local when synchronizer trigger activates
Cache servers by the concordance list of the index information of broadcast transmission including local resource, RCBF-SYNCHRO-RECV () when
Between the complexity upper bound be O (dm2), wherein m represents cell number in 1 bucket, and d represents the Hash sublist number of division;
Dynamic cache mechanism is disposed in ISP domain, when cache servers receive a request, and local is not hit
When, dl-RCBF will be inquired about first, to determine the information whether on other servers;When reporting by mistake, then the request is turned
It is dealt into overseas;
When local miss, the record positions that will inquire about in dl-RCBF, if while the request of certain resource exceed threshold
Value, then by the resources duplication to long-range cache servers.
The beneficial effects of the invention are as follows:
The inventive method considers flow settlement cost, lifting data transmission service quality two from operator's angle
Aspect proposes pattern of fusion content distributing network CDN (Content Delivery Networks, abbreviation CDN) model.Exist merely
Position deployment content caching system e insufficient to the service quality of effectively lifting user between domain, because reconnaissance problem has been proved to
For np problem, based on this, the present invention proposes that the route management method of pattern of fusion content distributing network is obtained closely using Greedy strategy
Node deployment is carried out like optimal solution.The inventive method reduces flow settlement cost, greatly improves data transport service matter
Amount.
Brief description of the drawings
Fig. 1 is typical content distributing network resource routing flow chart;Fig. 2 is intra-area routes schematic diagram;Fig. 3 is spanning tree
Procedure chart is cut, wherein:Fig. 3 (a) be G figure, Fig. 3 (b) be figure G original minimum spanning tree, Fig. 3 (c) be revised figure G most
Small spanning tree;Fig. 4 is dl-RCBF structural representations;Fig. 5 is core router weight distribution figure, in figure:Abscissa represents route
Node ID, ordinate represent node degree;Fig. 6 is flow cost and time loss figure, wherein:Fig. 6 (a) is flow cost and section
The graph of a relation of points, abscissa represents nodes in figure, and ordinate represents flow cost, unit MB, and Fig. 6 (b) disappears for the time
The graph of a relation with nodes is consumed, abscissa represents nodes in figure, and ordinate represents time loss, unit ms;
Fig. 7 is hit rate comparison diagram (every kind of tactful hit rate under different cache sizes), in figure:Fig. 7 (a), Fig. 7
(b), Fig. 7 (c) is respectively zipfα=0.8、zipfα=0.8、zipfα=1.2When hit rate comparison diagram;Fig. 8 is a variety of cache strategies
Current capacity contrast's figure, Fig. 8 (a) discharge relation figures between various strategies and domains, horizontal sit represent policing type, and vertical sit between representative domain is flowed
Size is measured, its unit is GB;Fig. 8 (b) is discharge relation figure, horizontal sit represent policing type, indulge and sit table under various strategies and in domain
Show flow between domain, its unit is GB.
It is known in this area that all English implications can find corresponding Chinese implication in specification in above-mentioned accompanying drawing
General knowledge.The software used in being tested to the present invention is English interface, therefore contains English in sectional drawing.
Embodiment
As shown in Fig. 2 present embodiment is to a kind of routing management side of pattern of fusion content distributing network of the present invention
Method is described in detail:
Routing node in ISP domain is commonly divided into three classes:Edge routing node (being designated as set E), core routing node
(being designated as set F) and inter-domain routing node (being designated as set T), as shown in Figure 2.The present invention is route to edge routing node, core
Node and inter-domain routing node definition are as follows:
Define 1 edge routing node:It is referred to as edge routing node with the routing node that main frame is joined directly together.
Define 2 inter-domain routing nodes:It is referred to as inter-domain routing section with the routing node that routing node is connected in other ISP domains
Point.
Define 3 core routing nodes:All route sections in ISP domain in addition to edge routing node and inter-domain routing node
Point.
If content server to be all deployed in edge router aspect, understand that content reaches nearest from user, but
Edge router is numerous, and maintenance cost is high, does not possess operability in level of practice, while this deployment way can not meet
To inter domain router path and most short.Therefore the present invention is it is not intended that node range of choice is positioned in edge route set,
But as far as possible in the reconnaissance of core router aspect with meet demand, and core router is typically smaller than 5 jumps to the hop count of user
(not being included in the route jumping figure after NAT).Table 1 is the primary symbols and implication that the inventive method is used, due to content node portion
It is deployed on router position, therefore for convenience of describing, the present invention does not make any distinction between router and content node.Assuming that set E elements
Number is n, and set F element numbers are m, and it is m ' that it, which neutralizes the direct-connected number of edge router, and set T element numbers are q.
The symbol of table 1 and implication
For apparent description flow cost problem, subscriber's main station is not considered to the route jumping figure of edge router, i.e.,
Edge router is the terminal node of network topology.And problem is simplified by following steps:
1. edge routing node is contracted into core routing node by tree-like contraction algorithm, while by core routing node
Cost weight setting be its edge routing node number for having under its command, now problem space be reduced to core router topology it is empty
Between, rather than whole ISP manifold.
2. inter-domain routing node set is considered as entirety, i.e., core routing node and inter-domain routing node phase in original topology
Path even merges into 1, due to calculating edge routing node to the most short of inter-domain routing node using first wide searching algorithm
During path, as long as it is not intended that whole routing nodes but search the 1st inter-domain routing node, algorithm can terminate, because
This this simplification does not reduce cost.
Step 1: node deployment strategy
Data type, size, number and the protocol type asked by terminal user can not be predicted, for simplification
Description, the present invention carry out single treatment for network data size, i.e., 1 time the consumed flow cost of transmission only considers process
Route jumping figure, without considering size of data in itself, and carry out following related definition.
Define 2-4 flow costs Cij(Pij):Flow cost caused by transmission data u from node i to node j is from i
To j shortest path lengths and the product of size of data | pij|·sizeu。
Define the weight w of 2-5 node isi:The user that node i is administered to node i path sum, i.e.,Table
Show r user that node i has under its command to the shortest path sum of node i.
Define 2-6 and repeat flow cost RC:When network data Size passes through shortest path P from node iijNode j is reached, if
The access path of other users overlaps with it, then data Size caused flows in path P are to repeat flow, i.e. RCij
=size | Pij|。
Define 2-7 synchronous flow costs SC:Node i to node j synchronous cost for network data Size in PijUpper generation
Flow, i.e. SCij=size | Pij|。
Obvious core routing node set F, the interstitial content selected from F are inverse relation with repeating flow, and synchronous
Cost is then directly proportional to interstitial content.If assuming, the content node number that can dispose be k, repeated in this time domain flow cost and
The relation of deployment node can be represented with formula 2-1:
By the data that the content node disposed on inter domain router is cached are newest, therefore synchronous initial source
For inter domain router node.For inter-domain routing node t, generated by first minimum of wide searching algorithm (BFS) generation using t as root
Set TR and cover k content node, then the cost required for synchrodata Size can be represented with formula 2-2:
Flow cost problem is now represented by:K node is found, to meet RC+SC minimums.Although in formalized description
Simultaneously priority is not present in middle RC and SC costs, but in practice, ISP is more likely to RC reduction, and synchronous node is got over
Few, the management cost consumed is smaller.Using the algorithm based on Greedy strategy, as shown in table 2.
Content node Deployment Algorithm in the domain of table 2
Key step can be summarized as in above-mentioned algorithm:
1. initial phase:Calculate each minimum spanning tree using core router as root node, while obtain the minimum
The repetition flow cost of spanning tree;
Flow cost is repeated 2. calculating:In the case of nodes are disposed more than 1, select to repeat according to Greedy strategy first
Those spanning-tree root nodes of flow Least-cost, and pass through other each numbers of plies of core router node in every spanning tree
Determine which tree it is stayed in, so as to correct the repetition flow cost of each tree, by taking Fig. 3 as an example.Initial topological diagram such as Fig. 3
(a) shown in, 5 nodes, k=2 and node 0 and node 2 are alternative content disposition nodes altogether, and Fig. 3 (b) is with node 0,2
The minimum spanning tree formed by root node and use BFS algorithms, then by (11-14) step in table 3 to minimum spanning tree
It is modified, such as Fig. 3 (c).
3. calculate synchronous cost:For each k, it is root node to calculate successively by inter domain router, k core routing node
Synchronous cost required for the minimum spanning tree of composition.
The inventive method is analyzed the time complexity of algorithm according to 3 above-mentioned stages, the stage 1 consumed when
Between cost be O ((m+U|m|) m), the stage 2 needs to travel through all core routing nodes in the worst case, therefore when
Between complexity be O (m2), the worst time complexity in similar stage 3 is O (m+1+U|m+1|).It may thus be appreciated that whole algorithm when
Between complexity be O ((m+U|m|)·m)。
Clearly as the number of users that each edge router is administered is more, therefore as k increase repeats the generation of flow
Valency decline speed, then speed slow down;Opposite, the cost growth rate consumed in the initial procedure of synchronous cost
It is relatively slow, as gathering way for k accelerates.Greedy strategy can obtain the extreme point during whole cost function changes with k, and k compared with
Small, although minimum value be able to may not be searched out, less k operability in practice process is stronger.
Step 2: establish content indexing structure
Because each ISP is owned by itself special resource, therefore acquired resource on cross-domain router is general not
Together.In general, ISPs is all tended to flow control in domain.Indexed by cache sharing, ISP flows between can reducing domain
Amount.So source synchronous mechanism is by as the key of network performance in influence ISP domain.Index plan based on bloom strobe utilities
Slightly it is the minimum method of synchronization of flow.The present invention proposes a kind of dl-RCBF data structures as the index method of synchronization.
Assuming that set U={ key0,key1,…,keyn, | U |=n, and whole hash tables are divided into d sublist.For letter
Change statement, it is assumed that single hash functions meet:Therefore hash (key) is made up of 2 parts,
The former will determine in the bucket loaded in d sublist of remainder partial insertions in that minimum table that the latter is
Remainder is in itself.The present invention will increase count and record after remainder.Count effect and dl-cbf classes
Seemingly, and which content server record then describes the key on.As shown in figure 4, then whole dl-RCBF spaces are n+m (F+C
+ K), wherein n is key number;
Specific implementation process is as shown in table 3:
Table 3
Step 3: synchronization mechanism
Index information is carried out into broadcast can cause paroxysmal high capacity and network traffics.In order to reduce it is this it is unfavorable because
Element, the present invention will shared resource indexes between servers by triggered synchro mode.While in order to reduce for same
The frequent access of resource.As shown in table 4, when synchronizer trigger activates, local cache servers by broadcast transmission concordance list,
The table only includes the index information of local resource.In the worst case, RCBF-SYNCHRO-RECV () time complexity
For O (dm2), wherein m represents cell number in 1 bucket, and d represents the Hash sublist number of division.
The dl-RCBF synchronized update false codes of table 4
Domain-to-domain link flow is an important component of ISP costs, and collaborative cache is disposed in ISP domain can
Save flow cost between substantial amounts of domain.The mode interacted between cache servers have it is a variety of, such as CRISP agreements, still
Share same index information and would potentially result in Single Point of Faliure, while frequently renewal will bring substantial amounts of extra load.This hair
It is bright to propose a kind of dynamic cache mechanism:When cache servers receive a request, and do not hit locally, will inquire about first
Dl-RCBF, to determine the information whether on other servers.When reporting by mistake, then forward this request to overseas.Such as table
Shown in 5, when local miss, the record positions that will inquire about in dl-RCBF, if while the request of certain resource exceed threshold
Value, then by the resources duplication to long-range cache servers.Because consistency problem be present in distributed system, and the present invention
The optimistic method of synchronization that synchronization mechanism uses, when consistency problem occurs, receiving node will notify requesting node.
The dynamic cache mechanism of table 5
The experiment effect of the present invention:
For detailed assessment contents construction mechanism performance proposed by the invention, the inventive method is entered at following 2 aspects
Row experimental analysis:(1) hit rate of dl-RCBF index structures;(2) flow cost.In experimentation, it is necessary first to it is determined that interior
Hold deployment number and the position of node, therefore first carry out the performance evaluation of Deployment Algorithm.
The Deployment Algorithm experiment parameter of table 6
Beijing Netcom partial-topology of the topological data from Venus InfoTech's detection in 2009 is tested, altogether 342 routes
Device, 874 links, 251 edge routers.In order to more really analyze algorithm, specific experiment parameter is entered
Setting is gone, as shown in table 6.The number of host that wherein each edge router has under its command is uniformly random according to being carried out between 20-50
Specify, Fig. 5 show by the main frame that is had under its command edge router carry out it is tree-like be contracted to core router after, each core
The weight of router.
For the quality of evaluation arrangement strategy, k values that the present invention is generated by different synchronous cost observation algorithms and
Corresponding total cost loss.Can be seen that from Fig. 6 (a) overall cost early stage decrease speed quickly, this is due to early stage every portion
The flow cost that 1 content node of administration is reduced is larger, and as node gradually increases, caused income also gradually slows down,
And now the cost of synchronous consumption is rising, after nearly all node all turns into content node, repeating flow cost reduces
For 0, and total cost consumption is whole synchronous costs.
From Fig. 6 (a) it is also seen that under the conditions of different synchronous costs, the extreme point obtained by Greedy strategy also differs,
General trend shows as that synchronous cost is smaller, and the k values obtained are bigger, and the phenomenon shows synchronous cost and produced in total cost
Raw influence is lower, and when synchronous cost is larger, the extreme point reach of curve.When the operation of whole algorithm is can be seen that in Fig. 6 (b)
Between with the increase of interstitial content, the time consumed close to power function cost, early time when being less than 10 nodes, consumed
Cost varies less, and when this is due to that number is few, each subtree generation and the time contrasted are also less.Greedy strategy in general
It is more effective in node deployment selection mechanism, and in the case that core router is more, can only needs less
Time can find more suitable deployment number.
The inventive method is analyzed the contents construction mechanism of proposition at following two aspects:
1. analysis uses the storage hit rate after dl-RCBF;
2. analysis carries out deployment strategy and indexes the situation of change of flow between synchronization mechanism converse domain;
Establish while dl-RCBF index structure purposes are to reduce synchronous cost to improve hit rate.And flowed between domain
The reduction of amount is the objective for the CDN for establishing ISP fusions.
Index structure is tested
The present invention is by dl-RCBF information compared with following 3 class cache algorithm:
1. local cache mechanism;
2.BPR modes, which in stationary window access frequency by being used as the threshold value broadcasted;
3.ESC strategies, pass through the server where ESC abstract structure query resources.
In comparative experiments, algorithm is replaced as cache more new strategies using LRU cache herein, using zipf letters
Number produces 10,000 request of different alpha parameters respectively, and content node is 3 between the domain disposed.
URL rank and times of acquisition meets Zipf distributions.Global hit rate is to evaluate the weight of distributed cache mechanism
Want index.Every kind of tactful hit rate under different cache sizes is tested respectively.From figure 7 it can be seen that collaborative cache
Mechanism is substantially better than non-cooperating formula.ESC modes are because local miss request can be transmitted to other sections by it better than BPR
Point.Because our dl-RCBF structures possess nodal information, therefore when forwarding request with higher accuracy.While with
Distribution function α increase, after every kind of tactful hit rate is all being lifted, but the hit rate of local policy reaches 43%, i.e.,
The space for caching increase can not also continue to improve hit rate, and three kinds of cooperative caching mechanism are with Parameters variation in addition, between
Difference gradually reduce.From Fig. 7, it can also be seen that cache policy effect in the case of parameter is bigger is better, and this is due to ginseng
Number is bigger to represent more to the repetitive requests of same resource, and the number of hit is also higher.
Flux experiment
In order to evaluate the performance of entire content routing mechanism, the present invention is using OMNet++ as analog platform to whole system
The performance of system is verified.Analog parameter is as shown in table 7, and the topological data arrived used in it is consistent with table 2-1.Entirely counting
According to concentration, including 14,287 requests of Isosorbide-5-Nitrae, wherein different requests amounts to 610,000.
The analog parameter of table 7
The inventive method considers two Key Performance Indicators.1st important performance indications be net in flow, the 2nd
It is the flow between domain.As shown in figure 8, wherein no-cache modes are represented not using cache strategies.As can be seen from Figure 8
As long as employing after content node stores to Internet resources, the flow in domain between domain has declined.Plan is locally stored
Slightly reduce the flow in domain between domain to a certain extent, yet with its limited storage space and do not have and other nodes
Cooperated, therefore the income brought is less.Mechanism proposed by the present invention not only greatly reduces the flow between domain, while significantly
Degree, which reduces, repeats flow in domain, it contrasts ESC main advantage and is that local miss request is used probabilistic manner by ESC
It is forwarded to other nodes.
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto,
Any one skilled in the art the invention discloses technical scope in, technique according to the invention scheme and its
Inventive concept is subject to equivalent substitution or change, should all be included within the scope of the present invention.
Claims (3)
1. a kind of contents construction management method of pattern of fusion content distributing network, it is characterised in that by routing node in ISP domain point
Three classes:Edge routing node is defined as with the routing node that main frame is joined directly together, removes edge routing node and Yu Jian roads in ISP domain
Core routing node, the route section being connected with routing node in other ISP domains are defined as by all routing nodes beyond node
Point is defined as inter-domain routing node, by being disposed to node-classification, and using flow cost as the benchmark of Greedy strategy;
The implementation process of methods described is:
Step 1: node deployment:
Single treatment is carried out for transmission data u sizes, i.e., 1 time the consumed flow cost of transmission only considers that the path of process is jumped
Number, without considering size of data in itself, and it is defined as below:
Flow cost Cij(Pij):Flow cost caused by transmission data u from node i to node j is shortest path from i to j
The product of length and size of data | pij|·size;Wherein size is transmission data u size of data;
The weight w of node ii:The user that node i is administered to node i path sum, i.e.,Represent that node i has under its command
R user to node i shortest path sum;
Repeat flow cost RCij:When transmission data u passes through shortest path P from node iijNode j is reached, if the visit of other users
Footpath of asking the way overlaps with it, then the flowing of access of transmission data u caused other users in path P is to repeat flow, i.e.,
RCij=size | Pij|;
Synchronous cost SCij:Node i is to transmit data u in P when occurring synchronous to node j synchronous costijFlowed caused by upper
Amount, i.e. SCij=size | Pij|;
The interstitial content that is selected from core routing node set F is inverse relation with repeating flow, synchronous cost then with node
Number is directly proportional;
It is assumed that the content node number that can be disposed is k, m ' is that in core routing node set F and edge router is direct-connected
Number, then flow cost is repeated in this time domain and represented with disposing the relation of node with formula 2-1:
Synchronous initial source is inter domain router node, for inter-domain routing node t, by first wide searching algorithm BFS generations with t
K content node is covered for the minimum spanning tree TR of root, then the cost required for synchrodata Size is represented with formula 2-2:
K node is found, meets RC+SC minimums, realizes flow Least-cost:
Flow Least-cost is realized using the algorithm based on Greedy strategy, its process is:
A. initial phase:Each minimum spanning tree using core router as root node is calculated, while obtains minimum generation
The repetition flow cost of tree;
B. calculate and repeat flow cost:In the case of nodes are disposed more than 1, selected to repeat flow first according to Greedy strategy
Those spanning-tree root nodes of Least-cost, and determined by the number of plies of other each core router nodes in every spanning tree
Which tree it is stayed in, so as to correct the repetition flow cost of each tree;
C. synchronous cost is calculated:For each k, it is root node to calculate successively by inter domain router, and k core routing node forms
Minimum spanning tree required for synchronous cost;
Step 2: structure index structure, using dl-RCBF data structures as the index method of synchronization;
Step 3: setting synchronization mechanism, synchronization mechanism is used for paroxysmal high negative caused by index information is broadcasted in reduction
Load and network traffics, it is additionally operable to reduce the frequent access for same resource.
A kind of 2. contents construction management method of pattern of fusion content distributing network according to claim 1, it is characterised in that:
In step 2, the process for building index structure is:
Assuming that set U={ key0,key1,…,keyn, | U |=n, and whole hash tables are divided into d sublist;It is if single
Hash functions meet:Hash (key) is made up of two parts, i.e. [B]dAnd remainder, the former
It will determine in the bucket loaded in d sublist of remainder partial insertions in that minimum table, the latter is remainder sheets
Body;
Increase count and record after remainder;Count effect is counter, for describing hash collision
Number;Record is used to describe the node location where the key, then whole dl-RCBF spaces are n+m (F+C+K), and wherein n is key
Number.
A kind of 3. contents construction management method of pattern of fusion content distributing network according to claim 2, it is characterised in that:
It is in the implementation process of step 3:
By triggered synchro mode, shared resource indexes between servers, when synchronizer trigger activates, local cache clothes
Business device includes broadcast transmission the concordance list of the index information of local resource, RCBF-SYNCHRO-RECV () time complexity
The upper bound is O (dm2), wherein m represents cell number in 1 bucket, and d represents the Hash sublist number of division, RCBF-SYNCHRO-
RECV () is the Hash table to all long-range transmissions successively contrast function compared with local Hash table;
Dynamic cache mechanism is disposed in ISP domain, will when cache servers receive a request, and do not hit locally
Dl-RCBF is inquired about first, to determine the information whether on other servers;When reporting by mistake, then forward this request to
It is overseas;
When local miss, the record positions that will inquire about in dl-RCBF, if while the request of certain resource exceed threshold value,
Then by the resources duplication to long-range cache servers.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410299708.1A CN104022911B (en) | 2014-06-27 | 2014-06-27 | A kind of contents construction management method of pattern of fusion content distributing network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410299708.1A CN104022911B (en) | 2014-06-27 | 2014-06-27 | A kind of contents construction management method of pattern of fusion content distributing network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104022911A CN104022911A (en) | 2014-09-03 |
CN104022911B true CN104022911B (en) | 2018-03-30 |
Family
ID=51439493
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410299708.1A Active CN104022911B (en) | 2014-06-27 | 2014-06-27 | A kind of contents construction management method of pattern of fusion content distributing network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104022911B (en) |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104244020B (en) * | 2014-09-22 | 2017-12-12 | 内蒙古工业大学 | A kind of request method for routing of cloud environment net cast service |
CN104506656B (en) * | 2015-01-09 | 2018-03-23 | 中国联合网络通信集团有限公司 | A kind of method and system of the set content distribution network node in mobile Internet |
CN105681438B (en) * | 2016-01-26 | 2019-12-13 | 南京航空航天大学 | centralized content center network cache decision method |
CN105933226A (en) * | 2016-04-20 | 2016-09-07 | 乐视控股(北京)有限公司 | Content distributing method and system |
CN105933234A (en) * | 2016-04-20 | 2016-09-07 | 乐视控股(北京)有限公司 | Node management method and system in CDN network |
CN106060605B (en) * | 2016-05-25 | 2019-03-26 | 清华大学深圳研究生院 | A kind of live stream adaptive approach and system based on CDN |
CN106487910B (en) * | 2016-10-26 | 2019-07-19 | 苏州大学 | A kind of content distributing network creation method |
CN108093496B (en) * | 2017-12-23 | 2020-03-13 | 西安交通大学 | ISA100.11a standard-based consistency networking method |
CN108280024B (en) * | 2017-12-29 | 2021-06-22 | 五八有限公司 | Flow distribution strategy testing method and device and electronic equipment |
CN109600432B (en) * | 2018-12-03 | 2021-08-03 | 上海科技大学 | Dynamic access method for content center mobile edge network user |
CN112751890B (en) * | 2019-10-29 | 2023-05-05 | 贵州白山云科技股份有限公司 | Data transmission control method and device |
CN112738149B (en) * | 2019-10-29 | 2023-04-25 | 贵州白山云科技股份有限公司 | Data transmission system and method |
CN113395208B (en) * | 2021-06-21 | 2022-06-14 | 哈尔滨工业大学 | BGP route leakage detection method and system based on block chain |
CN114501161B (en) * | 2022-04-11 | 2022-08-02 | 中兴通讯股份有限公司 | Audio and video system, access method, route forwarding method and readable storage medium |
CN115834594B (en) * | 2022-11-16 | 2024-04-19 | 贵州电网有限责任公司 | Data collection method for improving high-performance computing application |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101883052A (en) * | 2010-06-25 | 2010-11-10 | 中兴通讯股份有限公司 | Method and system for realizing traffic optimization in peer-to-peer network |
CN103001870A (en) * | 2012-12-24 | 2013-03-27 | 中国科学院声学研究所 | Collaboration caching method and system for content center network |
-
2014
- 2014-06-27 CN CN201410299708.1A patent/CN104022911B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101883052A (en) * | 2010-06-25 | 2010-11-10 | 中兴通讯股份有限公司 | Method and system for realizing traffic optimization in peer-to-peer network |
CN103001870A (en) * | 2012-12-24 | 2013-03-27 | 中国科学院声学研究所 | Collaboration caching method and system for content center network |
Non-Patent Citations (1)
Title |
---|
内容分发网络研究;李乔 等;《电子学报》;20130831 * |
Also Published As
Publication number | Publication date |
---|---|
CN104022911A (en) | 2014-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104022911B (en) | A kind of contents construction management method of pattern of fusion content distributing network | |
USRE49943E1 (en) | System and method for a context layer switch | |
JP5745169B2 (en) | Content processing method, content processing device, and content processing system | |
CN103475719B (en) | The content distribution method that in a kind of CDN-P2P UNE, cross-domain flow rate minimizes | |
Haßlinger et al. | Content delivery and caching from a network provider’s perspective | |
CN114090244B (en) | Service arrangement method, device, system and storage medium | |
Rajahalme et al. | Incentive-compatible caching and peering in data-oriented networks | |
CN101217565B (en) | A network organization method of classification retrieval in peer-to-peer network video sharing system | |
CN101420337A (en) | Method, device and indexing server for resource positioning | |
CN102546728A (en) | Peer-to-peer network resource downloading method | |
CN102075359A (en) | Network coordinate-based server deployment method and device | |
US20120259861A1 (en) | Method and apparatus for storing and searching for index information | |
Yamamoto | A survey of caching networks in content oriented networks | |
Tuncer et al. | Scalable cache management for ISP-operated content delivery services | |
CN101854387B (en) | P2P flow optimization method and system based on distributed index server architecture | |
CN101969458B (en) | P2P traffic optimization method supportive of hierarchical network topology | |
Ghalehtaki et al. | A bee colony-based algorithm for micro-cache placement close to end users in fog-based content delivery networks | |
CN108768690A (en) | A kind of the P2P self-organization network structures and resource search method of structuring | |
Claeys et al. | Hybrid multi-tenant cache management for virtualized ISP networks | |
Sarkar et al. | Content delivery networks: Insights and recent advancement | |
CN107302571B (en) | The routing of information centre's network and buffer memory management method based on drosophila algorithm | |
Amani et al. | Towards optimal content replication and request routing in content delivery networks | |
Li et al. | A smart routing scheme for named data networks | |
EP2211525B1 (en) | Method for distributing in a self-organizing, distributed overlay network a reference to an object | |
Reshadinezhad et al. | An efficient adaptive cache management scheme for named data networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |