CN1968122A - Simulated annealing-based dynamically distributed multicast routing method - Google Patents

Abstract

The invention relates to a dynamic distribution multicast route method based on analogue anneal, wherein it uses distribution to build minimum-cost multicast route tree met time delay and time delay vibration limit, and supports dynamic recombination of multicast tree. Said invention is formed by initial solution construction and optimized solution construction, while the first step builds one initial multicast tree met end-to-end time delay and vibration limit without considering cost; and the second step iterates via analogue anneal, to reduce the cost of initial multicast tree, via exchange path, to build nearby solution, and obtain final multicast tree. The invention has better real-time property, etc.

Description

Dynamically distributed multicast routing method based on simulated annealing

One technical field

The invention belongs to interconnection path and have the multicast route technology of a plurality of QoS parameters, particularly a kind of dynamically distributed multicast routing method based on simulated annealing.

Two background technologies

As an important component part of multi-casting communication, (Quality-of-Service, QoS) Yue Shu multicast route technology has become the important content and the hot issue in network research field based on service quality.In recent years, appearance along with the real-time multimedia business, Steiner (Stanley) tree problem with QoS constraint has become the focus of multicast routing issue research, and this problem has been proved to be the complete problem into NP (Nondeterministic Polynomial, nondeterministic polynomial).

The initial research of multicast route mainly concentrates on unconfined minimum cost multicast routing issue and delay constraint minimum cost multicast routing issue.But along with the development of network application, some switch type real time communication are used except requiring and are also required delay variation to suffer restraints source node will suffer restraints to the time delay of destination node.Delay variation constraint is in order to guarantee that different recipients keeps synchronously when the information of reception, does not allow some recipient lag behind especially or shifts to an earlier date especially.For example, in the telecommunications meeting, the talker is heard it is very important simultaneously by all participants, otherwise, lack the sensation that interactive mode talks face to face.

The multicast routing issue of time delay and time delay jitter restriction mainly contains the research of two aspects at present: a class is not consider the problem of cost optimization, is called DVBMT problem (Delay and Delay Variation Bounded Multicast TreeProblem); Another kind of is the problem of considering cost optimization, is called DVBST problem (Delay and Delay VariationBounded Steiner Tree Problem).

Rouskas proposed in 1997 to satisfy the multicast routing issue of end-to-end time delay and time delay jitter restriction first, and had provided a kind of effective method DVMA (Delay Variation Multicast Algorithm).This method has obtained good Expected Results, and promptly the multicast tree delay variation that returns of method is very little, but computation complexity is very high, is O (klmn ⁴), be difficult to carry out effectively the multicast route.Because the very loaded down with trivial details complexity of this body structure of DVMA method, Rouskas fails further to propose to minimize the problem of network cost and solve, therefore the cost performance of method is not fine (Rouskas GN, Baldine I. Multicast Routing with End-to-End Delay and Delay Variation Constraints.IEEE Journal on Selected Areas in Communications, 1997,15 (3): 346-356.).

At the high complexity of DVMA method, people such as Sheu propose a kind of DDVCA of heuristic fast and effectively (Delay and Delay Variation Constraint Algorithm).The main thought of this method can fast and effeciently find the multicast tree that satisfies delay constraint and minimal time delay shake based on notion and the shortest cost path method that core router in the nuclear tree (CBT) is arranged, and the time complexity of method is O (mn ²) (Sheu P R, Chen S T. A Fast and EfficientHeuristic Algorithm for the Delay-and Delay Variation-bounded Multicast Tree Problem, Computer Communications, 2002,25 (8): 825-833.).

DDVBM (the Distributed Delay and Delay Variation BoundedMulticast Algorithm) method that people such as Low and Lee proposes is first distributed method that solves the DVBST problem, and its computation complexity is O (n ³).This method mainly is divided into two stages.Phase I is mainly sought the minimum cost multicast tree that satisfies delay constraint, if should violate the delay variation restriction by tree, then enters second stage, promptly seeks the minimum cost tree of satisfying time delay and time delay jitter restriction simultaneously by calling the route searching process.But the destination node that the DDVBM method has been ignored in the multicast group also can be used as intermediate node forwarding grouping, therefore when calculating auxiliary Steiner Node Tree, the multicast tree of source node and all Steiner nodes might can't connect, cause method can't continue to carry out (Low C P, Lee Y J.Distributed Multicast Routing, with End-to-End Delay and Delay Variation Constrains.Computer Communications, 2000,23 (9): 848-862.).

The medium people of Wang Ming is in the process of research DVBST problem, find current numerous method two best link choice functions generally using can not embody the dynamic process of route fully, and under certain specific situation, possibly can't realize " multicast can reach ", promptly by may not comprising all destination nodes in the multicast tree as a result of its generation.For this reason, he proposes a qualification and the new best link choice function about " multicast can reach ", and having designed a new heuristic DDVBMRA (Delay and Delay Variation Bounded Multicast RoutingAlgorithm) on this basis, the time complexity of method is O (mn ²).But this method is in operation and all accessed mistake of all network nodes can occur, and Sub this moment (M) still is not equal to the situation of destination node collection, thereby cause the DDVBMRA method can't obtain feasible multicast tree (Wang Mingzhong, Xie Jianying, open and respect shafts. the minimum cost multicast routing policy of time delay and time delay jitter limits. Chinese journal of computers, 2002,25 (5): 534-541.).

Yu Yanping has proposed another kind of heuristic LCDVMA (Low-cost DelayVariation-constrained Multicast Algorithm) in its thesis for the doctorate, make up the multicast tree of the network cost optimization that satisfies time delay and delay variation constraint.The main thought of this method is at every turn by the shortest cost path or prolong the path the most in short-term and link arbitrary node on the tree, and by relatively finding out optimal path and adding on the tree, the method complexity is O (mkn again ³).Because this method has increased the scope of feasible path, thereby cause the delay variation of multicast tree to increase (Yu Yanping. Multicast Routing Algorithm research. Zhejiang University's thesis for the doctorate, 2002.).

People such as Guo Wei mention the multicast routing issue of time delay and time delay jitter restriction on the level of optimization, have proved that the DVBST problem is a np complete problem, propose a kind of heuristic genetic method based on the dynamic penalty function method then to find the solution this problem.This method uses genetic algorithm to select the Steiner point, and to selected Steiner point, uses the method for similar minimum spanning tree heuristic to find the solution the Steiner tree.Because this method adopts penalty function method with feasibleization of infeasible solution, thereby enlarged the region of search, increased computing time (Guo Wei, Xi Yugeng. have time delay in time to prolong the minimum cost multicast routing problem of difference constraint. the communication journal, 2001,22 (6): 13-20.).

There are following four problems in the multi-broadcast routing method of existing these QoS constraints in being applied to the real time communication network:

(1) the existing method of great majority all is centralized, and centralized approach needs Centroid (or each node) to be responsible for calculating whole routing tree, and this node must be grasped the details of whole network configuration.In large scale network, can there be problems such as the overweight and routing iinformation of the calculated load of poor fault tolerance, Centroid is inaccurate.

(2) the existing distributed method still needs the partial status information of the whole network of each node maintenance, more or less can have the problem of centralized approach.And these distributed methods have that communication cost height, connection setup time are long, the shortcoming such as of poor quality of routing tree.

(3) have only a few methods to consider multicast member's dynamic change problem.Because in many multicasts were used, the multicast member can freely dynamically add or leave multicast conversation.Therefore, guarantee multicast member's dynamic change and do not influence current connection proper communication, to make the destructiveness minimum of multicast tree be very important.

(4) it is quite difficult seeking an optimum multicast tree under time delay and delay variation double constraints.In the multicast route, minimize the network cost and tend to and time delay or delay variation restriction conflict mutually.Therefore, concerning the researcher, the Steiner tree method, particularly distributed method of time delay and delay variation constraint satisfied in design simultaneously, and be just very difficult.

Three summary of the invention

The object of the present invention is to provide a kind of dynamically distributed multicast routing method based on simulated annealing, this method not only can construct the minimum cost multicast router tree that satisfies time delay and delay variation constraint with a kind of distributed way, and when multicast member's dynamic change, carry out the dynamic reorganization of multicast tree, and guarantee the destructiveness minimum of multicast conversation.

The technical solution that realizes the object of the invention is: a kind of dynamically distributed multicast routing method based on simulated annealing comprises initial solution construction process and optimal solution construction process, promptly

1.1 the step of described initial solution construction process is as follows:

1.1.1 the preceding k bar that uses distributed k-Bellman-Ford k bar shortest-path method to calculate from source node s to each destination node m ∈ M prolongs the path the most in short-term, as the alternative path collection, wherein M is the destination node collection, claims the multicast group again;

1.1.2 select a destination node at random, and add alternative path concentrated source node s to initial multicast tree T to the path of prolonging the most in short-term of this destination node ₀In, T wherein ₀Initial condition be an empty tree;

1.1.3 concentrate to select to satisfy formula max{0, maxd-δ from source node s destination node m outside tree at alternative path }≤d (p (s, m))≤min{mind+ δ, Δ } article one alternative path, this m ∈ M, m  T ₀, and add it to T ₀In, wherein (p (s, m)) represents the path delay of time from source node s to destination node m to d; Maxd and mind represent respectively from source node s to T ₀In the maximum delay and the minimal time delay in path in the path of the destination node that comprises, Δ and δ are respectively the time delay upper limit and the delay variation upper limit of multicast conversation;

1.1.4 repeating step 1.1.3, all destination nodes in M all are included in T ₀In, so far the initial solution construction process finishes, the initial multicast tree T of this procedure construction ₀Will be as the initial solution of optimal solution construction process;

1.2 the step of described optimal solution construction process is as follows:

1.2.1 T is set _Best=T _Now=T ₀, T wherein _BestRepresent the multicast tree of cost minimum up to the present, i.e. optimal solution, T _NowRepresent current multicast tree, promptly current separating;

1.2.2 set initial temperature t=t ₀

1.2.3, then enter step 1.2.8 as if loop termination criterion in this temperature reaches; Otherwise enter next step 1.2.4;

Produce the current T of separating 1.2.4 separate the structure process with neighborhood _NowNeighborhood disaggregation N (T _Now), and therefrom select a multicast tree T at random _Next, calculate Δ c=cost (T _Next)-cost (T _Now), cost (T wherein _Next) and cost (T _Now) represent multicast tree T respectively _NextWith the current T that separates _NowCost;

If 1.2.5 Δ c≤0 or exp (Δ c/t)＞random (0,1), T _Now=T _Next

If 1.2.6 cost (T _Now)＜cost (T _Best), cost (T wherein _Best) expression optimal solution cost, T _Best=T _Now

1.2.7 repeating step 1.2.3～step 1.2.6;

1.2.8 move back temperature, t=λ t, wherein λ is for moving back warm speed; If satisfy termination rules, stop calculating, otherwise get back to step 1.2.3; So far the optimal solution construction process finishes, and the output of this process is exactly a minimum cost multicast tree that finally satisfies time delay and delay variation constraint.

The neighborhood that the present invention is based on the dynamically distributed multicast routing method of simulated annealing is separated the structure process and is constructed the neighborhood disaggregation by switching path in the feasible solution scope, and its step is as follows:

Step 1: at current multicast tree T _NowIn select a leaf destination node m at random;

Step 2:, delete current multicast tree T from destination node m _NowIn lead to the link of source node s, up to running into another destination node or source node, form stalk tree T _Sub

Step 3: calculate current subtree T _SubIn maximum delay maxd and minimal time delay mind from source node s to each destination node path;

Step 4: concentrate at alternative path, with time delay between [max{0, maxd-δ }, min{mind+ δ, Δ }] and the alternative path from source node s to destination node m join current subtree T _SubIn; If the new tree that forms is acyclic, then become current multicast tree T _NowA neighborhood separate;

Step 5: repeating step 4, all checked up to all k bar alternative paths.

After the dynamically distributed multicast routing method that the present invention is based on simulated annealing satisfies a minimum cost multicast tree of time delay and delay variation constraint at structure, the situation that the multicast member dynamically adds or leave multicast conversation can appear, at this moment need the multicast tree destination node of dynamically recombinating, and the multicast tree after guaranteeing to recombinate still satisfies time delay and delay variation retrains, and described dynamic reconfiguration method is as follows:

When destination node member m ∈ M request " leaving " multicast group, in two kinds of situation: (1) if destination node m is not a leaf node, this situation is not made any change to multicast tree, as long as allow the user forwarding packet data of this node stop to oneself; (2) if destination node m is a leaf node, it sends leave request message to the source node direction of multicast tree so, and a node connects a node, arrives source node or another destination node up to this message; Message the link of process with deleted;

When new node v  M request " adding " multicast group, it sends join request message to source node so, divide three kinds of situations: (1) is not if v is the node in the tree, the preceding k bar that then obtains from source node to node v prolongs the path the most in short-term, and select article one time delay at [max{0, maxd-δ }, min{mind+ δ, Δ] between the path join in the current multicast tree; If can't find such path, then abandon joining request of node v; (2) if v be the tree in node, and new multicast group M ∪ { v} satisfies delay variation upper limit δ, and then this multicast tree itself is exactly a feasible tree; (3) if v is a tree node, and new multicast group M ∪ { v} does not satisfy delay variation upper limit δ, and the intermediate node that node v must be a source node to one or more destination nodes path is described; In this case, delete the path that this comprises v and these destination nodes, again node v and these destination nodes are joined in the routing tree successively.

The present invention compared with prior art, its remarkable advantage is: (1) constructs the multicast tree that satisfies time delay and delay variation minimum cost in complete distributed mode; (2) according to the change of destination node, the multicast tree of dynamically recombinating, and guarantee destructiveness minimum to multicast tree; (3) neighborhood that is undertaken by switching path is separated the shortcoming that the structure process has overcome existing methods such as region of search expansion, increase computing time, and guarantees that optimal tree satisfies time delay and delay variation constraint; (4) have fast convergence rate, encoding and decoding are simple and real-time is good characteristics; (5) cost is little, and the performance of time delay, delay variation and route success rate is also fairly good.

Four description of drawings

Fig. 1 is the flow chart that the present invention is based on the dynamically distributed multicast routing method of simulated annealing.

Fig. 2 is that the neighborhood that the present invention is based on the dynamically distributed multicast routing method of simulated annealing is separated the flow chart of structure process.

Fig. 3 is the network topological diagram that the present invention is based on the dynamically distributed multicast routing method utilization example of simulated annealing.

Fig. 4 is the initial solution construction process schematic diagram that the present invention uses example.

Fig. 5 is that the present invention uses the neighborhood of example to separate structure process schematic diagram.

Fig. 6 is the relation of experimental simulation result-number of network node of the present invention and multicast tree cost.

Fig. 7 is the relation of experimental simulation result of the present invention-destination node number and multicast tree cost.

Five embodiments

Below in conjunction with accompanying drawing the present invention is described in further detail.

In conjunction with Fig. 1, satisfy the minimum cost multicast router tree of time delay and delay variation constraint for searching, the dynamically distributed multicast routing method that the present invention is based on simulated annealing comprises initial solution construction process and optimal solution construction process, that is:

1.1 described initial solution construction process is not consider under the cost prerequisite, constructs a multicast tree that satisfies end-to-end time delay and delay variation constraint, with its initial feasible solution as the optimal solution construction process, its step is as follows:

1.1.1 the preceding k bar that uses distributed k-Bellman-Ford k bar shortest-path method to calculate from source node s to each destination node m ∈ M prolongs the path the most in short-term, as the alternative path collection, wherein M is the destination node collection, claims the multicast group again;

1.1.2 select a destination node at random, and add alternative path concentrated source node s to initial multicast tree T to the path of prolonging the most in short-term of this destination node ₀In, T wherein ₀Initial condition be an empty tree;

1.1.3 concentrate to select to satisfy formula max{0, maxd-δ from source node s destination node m outside tree at alternative path }≤d (p (s, m))≤min{mind+ δ, Δ } article one alternative path, this m ∈ M, m  T ₀, and add it to T ₀In, wherein (p (s, m)) represents the path delay of time from source node s to destination node m to d; Maxd and mind represent respectively from source node s to T ₀In the maximum delay and the minimal time delay in path in the path of the destination node that comprises, Δ and δ are respectively the time delay upper limit and the delay variation upper limit of multicast conversation;

1.1.4 repeating step 1.1.3, all destination nodes in M all are included in T ₀In, so far the initial solution construction process finishes, the initial multicast tree T of this procedure construction ₀Will be as the initial solution of optimal solution construction process;

1.2 described optimal solution construction process is the cost that reduces initial multicast tree by the continuous iteration of simulated annealing, but the restriction of not violating time delay and delay variation, finally obtains the multicast tree of minimum cost, its step is as follows:

1.2.1 T is set _Best=T _Now=T ₀, T wherein _BestRepresent the multicast tree of cost minimum up to the present, i.e. optimal solution, T _NowRepresent current multicast tree, promptly current separating;

1.2.2 set initial temperature t=t ₀

1.2.3, then enter step 1.2.8 as if loop termination criterion in this temperature reaches; Otherwise enter next step 1.2.4;

Produce the current T of separating 1.2.4 separate the structure process with neighborhood _NowNeighborhood disaggregation N (T _Now), and therefrom select a multicast tree T at random _Next, calculate Δ c=cost (T _Next)-cost (T _Now), cost (T wherein _Next) and cost (T _Now) represent multicast tree T respectively _NextWith the current T that separates _NowCost;

If 1.2.5 Δ c≤0 or exp (Δ c/t)＞random (0,1), T _Now=T _Next

If 1.2.6 cost (T _Now)＜cost (T _Best), cost (T wherein _Best) expression optimal solution cost, T _Best=T _Now

1.2.7 repeating step 1.2.3～step 1.2.6:

1.2.8 move back temperature, t=λ t, wherein λ is for moving back warm speed; If satisfy termination rules, stop calculating, otherwise get back to step 1.2.3; So far the optimal solution construction process finishes, and the output of this process is exactly a minimum cost multicast tree that finally satisfies time delay and delay variation constraint.

Being described in detail as follows of said method.

Suppose (V, E) network of expression with the simple non-directed graph G=of cum rights.Wherein, V, E are respectively network node and link set.For every link e ∈ E, define two arithmetic number functions: cost function cost (e): E → R ⁺With time delay function d elay (e): E → R ⁺, R wherein ⁺Expression arithmetic number collection.The source node s ∈ V of given multicast conversation, one group of destination node M  V-{s}, m represents a destination node among the destination node collection M, multicast tree T (T  G) is that root is s, and comprises the one tree of destination node set M.

Definition: time delay and delay variation constrained minimum cost multicast tree

Time delay and delay variation constrained minimum cost multicast tree be one from source node s, connect the multicast router tree T (V of all destination node M _T, E _T), and satisfy:

● delay constraint: ${\mathrm{\Δ}}_T==\max \left(\underset{e\∈p(s,m)}{\mathrm{\Σ}}d\left(e\right)\right)\≤\mathrm{\Δ}---\left(1\right)$

● the delay variation constraint: ${\mathrm{\δ}}_T=\underset{u,v\∈M}{\max }\left\{\right|\underset{e\∈p(s,u)}{\mathrm{\Σ}}d\left(e\right)-\underset{e\∈p(s,v)}{\mathrm{\Σ}}d\left(e\right)\left|\right\}\≤\mathrm{\δ}$

● the cost constraint: in the all-multicast tree of satisfy condition (1), (2), $\cos t\left(T\right)=\underset{e\∈T}{\mathrm{\Σ}}\cos t\left(e\right)$ Minimum.

Wherein, Δ _T, δ _T, cost (T) represents time delay, delay variation and the cost of multicast tree T respectively.

Suppose that each node has to the k bar of each destination node and prolong path and its path delay of time the most in short-term.These information stores are represented with Route in the local routing table of each node, and are obtained by the distributed kBF method calculation delay metric of operation.In Route, each node v ∈ V contain k * | M| clauses and subclauses, the k bar shortest path of the corresponding destination node of each clauses and subclauses.Clauses and subclauses Route[i] [m] have two field: Route[i] [m] .n and Route[i] [m] .d, represent that respectively v prolongs the next adjacent node on the path and the time delay size of this paths the most in short-term to the i bar of m.In addition, for every multicast tree T in the optimal solution construction process _Now, suppose that each node stores its previous adjacent node (along the source node direction) to the reverse routing iinformation of each destination node, represent this routing iinformation with RevRoute.Each node v has | M| clauses and subclauses, wherein, clauses and subclauses RevRoute[m] a field RevRoute[m arranged] .p represents T _NowIn the previous adjacent node of s v in the destination node m path.

The data structure of control messages is with a tlv triple＜type, kth, dest〉expression, wherein type represents that type of message, kth represent that alternative path concentrates the k bar to prolong the destination node that current selection is represented in path, dest the most in short-term.The present invention has used 9 kinds of main type of messages, and is as shown in table 1:

Table 1 type of message of the present invention

Message name	Message semantic definition
		open	Open multicast and connect, obtain the alternative path collection
Ini_start	Begin to construct initial multicast tree
		Ini_add	Add an alternative path of source node to a destination node to initial multicast tree
Ini_notify	Destination node of notification source node is added in the tree
		Ini_finish	Finish the structure of initial multicast tree
Gn_start	The neighborhood that generates current multicast tree is separated
		Gn_add	Add an alternative path to current subtree
Gn_delete	From present tree, delete the path of specifying destination node
		Gn_finish	Finish the generation that neighborhood is separated

Each node in the system is carried out identical method for routing, is initially foundation requests to be connected such as dummy status.In the initial solution construction process, receive when node and to open multicast connection requests (open message) that this request has following parameter: destination node collection M, time delay upper limit Δ and delay variation upper limit δ.This node utilizes distributed kBF method calculating s to prolong the path the most in short-term to the preceding k bar of each destination node m ∈ M.If P _mThe preceding k bar of expression destination node m prolongs the path the most in short-term, i.e. P _m={ p ₁(s, m), p ₂(s, m) ..., p _k(s, m) },

Wherein, d (p ₁(s, m))≤d (p ₂(s, m))≤...≤d (p _k(s, m)).

After source node s receives the ini_start request, initial multicast tree T ₀Be initialized to an empty tree, and select a destination node m at random.If maxd and mind represent respectively from source node s to current T ₀In the maximum delay and the minimal time delay in path in the path of the destination node that comprises, initial value is d (p ₁(s, m)) (be Route[1] [m] .d).Ini_add connection message＜ini_add, 1, m〉be sent to along path p ₁(s, next adjacent node v m), and link (s v) is added to multicast tree T ₀On.

Be delivered to when ini_add message arrive an intermediate node u in the process of specifying destination node m after, node u is again with ini_add message＜ini_add, kth, m〉pass to it next adjacent node u ' (u '=Route[kth] [m] .n), link (u, u ') is added to multicast tree T subsequently ₀On.

Arrive the destination node of appointment when ini_add message after, ini_notify message is sent to source node s, informs that a destination node is added on the multicast tree.After receiving ini_notify message, select the outer destination node of a tree, for example m '.Then, concentrate from alternative path and select time delay at article one alternative path between [max{0, maxd-δ }, min{mind+ δ, Δ }].Suppose that this alternative path is that the i bar prolongs the path the most in short-term, so＜ini_add, i, m '〉message is sent to along this path to the next adjacent node v ' of m ' direction.After the value of upgrading maxd and mind, link (s, v ') is added to multicast tree T ₀On.

Above-mentioned several steps repeats between destination node, and all destination nodes in M all are included in multicast tree T ₀In.Last destination node in ini_add request message arrival M, this node sends ini_finish message to source node s.Like this, structure satisfies the initial multicast tree process of time delay and delay variation to be finished, and method enters the optimal solution construction process, and with multicast tree T ₀Initial solution as simulated annealing.

The optimal solution construction process calls the continuous iteration of simulated annealing method and reduces the cost of initial multicast tree, and does not violate time delay and delay variation constraint, finally obtains the multicast router tree of minimum cost.Wherein, the simulated annealing parameter is provided with as follows:

● the encoding and decoding of multicast tree: $T=p(s,m_1)\∪p(s,m_2)\∪...\∪p(s,m_{\left|M\right|})=\underset{i=1}{\overset{\left|M\right|}{\∪}}p(s,m_i),$

 m wherein _i∈ M, 1≤i≤| M|, p (s, m _i) represent that source node s is to destination node m among the multicast tree T _iThe path of ∈ M.

● initial temperature: t ₀=100;

● evaluation function: f (x)=cost (x);

● move back temperature strategy: t _I+1=λ t _i, i 〉=0 wherein, λ=0.9, λ is called and moves back warm speed;

● state is accepted function: $A_{\mathrm{ij}}\left(t\right)=\left[\begin{array}{cc}1,& \cos t\left(T_i\right)\&GreaterEqual;\cos t\left(T_j\right)\\ \exp (-{\mathrm{\&Delta;c}}_{\mathrm{ij}}/t),& \cos t\left(T_i\right),<\cos t\left(T_j\right)\end{array}\right]$

Wherein, A _Ij(t) expression is from the current T that separates _iNeighborhood N (T _i) in picked at random separate T _jAfter, accept T _jProbability; T is a Current Temperatures; Δ c _IjBe state T _iWith T _jGoal discrepancy, i.e. Δ c _Ij=cost (T _i)-cost (T _j).

● interior loop termination criterion:, then move back temperature as long as continuous 20 iteration of optimal solution are not improved;

● termination rules: it is 100 that maximum iteration time is set.

The key issue of optimal solution construction process is the generation of neighborhood disaggregation, the present invention constructs the neighborhood disaggregation by switching path in the feasible solution scope, promptly in the feasible solution scope, construct the neighborhood disaggregation by switching path, adopt shortcomings such as region of search expansions that exists in the penalty function strategy, increase computing time thereby overcome existing method, and guarantee that final optimal tree satisfies time delay and delay variation retrains.

In conjunction with Fig. 2, the step that described neighborhood is separated the structure process is as follows:

At structure present tree T _NowNeighborhood separate in the process, after a selecteed leaf destination node m receives gn_start message, can be with＜gn_delete ,/, m〉message along s to the path of m to source node s direction by the node transmission, be sent to source node or other destination node up to this message.Every link that the message transmission is passed through is all deleted, T _NowDelete formed subtree T behind these links _SubExpression, wherein T _SubBe initialized to T _NowSuppose T _NowIn path p (s, m) the previous adjacent node of going up destination node m is v, v can obtain (v=RevRoute[m] .p) by reverse routing table RevRoute.Then＜gn_delete ,/, m〉message is sent to node v, and link (m is v) from T _SubMiddle deletion.

After an intermediate node is received gn_delete message, immediately this message is delivered on its next adjacent node.Otherwise,, mean the EO of deleting the path if source node or another destination node are received gn_delete message.Next, select alternative path to concentrate other alternative path to replace deleted path, form T to destination node m _NowNew neighborhood separate.On source node or another destination node, at first calculate and lead to T _SubThe minimum and maximum time delay in the path of all destination nodes that comprise in the subtree (representing with maxd and mind respectively) sends gn_add message＜gn_add, 1, m to source node subsequently〉begin to add operation.

If T represents T _NowA new neighborhood multicast tree, comprise T when initial _SubAll links.When an intermediate node v receives gn_add message, it is delivered to next node v ' with this message, and wherein, v ' is at the adjacent node that leads to v on the k bar alternative path of destination node m.Subsequently, link (v, v ') joins among the multicast tree T.

After source node s received gn_add message, the destination node m of appointment and the value of kth extracted from message and are recorded in the source node.Source node is selected the alternative path of a s to m repeatedly, up to the time delay in this path between [max{0, maxd-δ }, min{mind+ δ, Δ }].If all accessed mistake of all k bar alternative paths then sends gn_finish message to source node s, inform that neighborhood separates structure and finish.Otherwise, to the next adjacent node u transmission＜gn_add of source node s, kth, m〉and message, and with (s, u) link joins among the multicast tree T.

Specify destination node when gn_add message arrives, show successfully to have generated a new neighborhood multicast tree T.If T is acyclic, then be inserted into neighborhood disaggregation N (T _Now) in as T _NowA neighborhood separate.Check the value of kth subsequently, if the kth value greater than k, then sends gn_finish message to source node s.Otherwise the kth value adds 1, and general＜gn_add, kth, m〉message is sent to source node s to check next bar alternative path.

Add operation and add every alternative path that leads to the leaf destination node of appointment to subtree T _SubOn, prolong all accessed mistake in path the most in short-term up to all k bars.After gn_finish message is sent to source node, promptly finished the structure that neighborhood is separated.

When finish a minimum cost multicast tree that satisfies the constraint of time delay and delay variation by initial solution construction process of the present invention and optimal solution construction process after, the situation that the multicast member dynamically adds or leave multicast conversation can occur, at this moment need the multicast tree destination node of dynamically recombinating.At present, (Internet Group Management Protocol IGMP) is used to handle the problem that the multicast group dynamically changes at the multicast igmpinternet that has occurred standard on the internet.But because there is not the logic connection in the characteristic of internet itself between the multicast group membership, so IGMP does not consider the network cost, thereby directly IGMP to be applied in the multi-medium multi-packet broadcasting session be inappropriate.Simultaneously,,, will block ongoing multicast conversation, just and this session can restart after can only finishing in new multicast tree foundation if re-construct multicast tree according to new destination node collection in order to support the dynamic reorganization of multicast group.Therefore the method for this dynamic reorganization also is worthless.For this reason, the invention provides the dynamic reconfiguration method of following multicast tree:

When destination node member m ∈ M request " leaving " multicast group, in two kinds of situation: (1) if destination node m is not a leaf node, this situation is not made any change to multicast tree, as long as allow the user forwarding packet data of this node stop to oneself; (2) if destination node m is a leaf node, it sends leave request message to the source node direction of multicast tree so, and a node connects a node, arrives source node or another destination node up to this message; Message the link of process with deleted;

When new node v  M request " adding " multicast group, it sends join request message to source node so, divide three kinds of situations: (1) is not if v is the node in the tree, the preceding k bar that then obtains from source node to node v prolongs the path the most in short-term, and select article one time delay at [max{0, maxd-δ }, min{mind+ δ, Δ] between the path join in the current multicast tree; If can't find such path, then abandon joining request of node v; (2) if v be the tree in node, and new multicast group M ∪ { v} satisfies delay variation upper limit δ, and then this multicast tree itself is exactly a feasible tree; (3) if v is a tree node, and new multicast group M ∪ { v} does not satisfy delay variation upper limit δ, and the intermediate node that node v must be a source node to one or more destination nodes path is described; In this case, delete the path that this comprises v and these destination nodes, again node v and these destination nodes are joined in the routing tree successively.

The process of work of the present invention is described below by an embodiment.

Network topology structure as shown in Figure 3, s={F} wherein, M={B, D, E, H}, solid circles is represented source node among the figure, the thick line circle is represented destination node, and mark is the cost and the time delay value of this link in the bracket of every link, and given delay constraint Δ and delay variation constraint δ are respectively 10,3.As shown in table 2 by the alternative path collection that distributed kBF method obtains, k=5 wherein.

The alternative path collection of table 2 example network

Destination node	B		D		E		H
									kth	The path	Time delay	The path	Time delay	The path	Time delay	The path	Time delay
1	F，D，C，G，B	5	F，D	2	F，D，E	3	F，H	2
									2	F，H，G，B	6	F，H，E，D	4	F，H，E	3	F，D，E，H	4
3	F，H，E，D，C，G，B	7	F，A，E，D	5	F，A，E	4	F,A，E，H	5
									4	F，A，E，D，C，G，B	8	F，H，G，C，D	7	F，H，G，C，D，E	8	F，D，C，G，H	7
5	F，D，C，B	8	F，A，E，H，G，C，D	10	F，D，C，G，H，E	8	F，A，E，D，C，G，H	10

Fig. 4 has shown the initial solution construction process of utilization example.At first select destination node D, and F (F D) is added on the empty tree, shown in Fig. 4 (a) to the minimal time delay path of D.Then, E becomes the next destination node that adds.Because maxd and mind are 2, therefore select to lead to the alternative path of article one time delay of E in [0,5].Time delay is 3 path (F, D, E) adding T as a result ₀In, shown in Fig. 4 (b).Subsequently, by same way as add the path (F, D, C, G, B), destination node B is also contained in T ₀In, shown in Fig. 4 (c).At this moment, maxd and mind are updated to 5 and 2 respectively.Consider last destination node H, article one from F to H and the alternative path of time delay distribution in [max{0,5-3}, min{2+3,10}] (i.e.[2,5]) be { F, H}, so { F, H} is added into T in the path ₀In, shown in Fig. 4 (d).So far, the initial solution construction process finishes.

For the construction process that neighborhood is separated, suppose that current separating is the T shown in Fig. 4 (d) ₀Select destination node E at random, and the path F, D, E} from the tree T ₀Middle deletion.Notice that because D is a destination node, (D, E) deleted, the subtree after the deletion is shown in Fig. 5 (a) so have only link.At this moment, maxd and mind are respectively 5 and 2.As known from Table 2, except deleted path F, D, outside the E}, the time delay distribution of having only two paths between [2,5], that is: (F, H, E) and (F, A, E).Therefore, this two paths is added respectively in the subtree, just formed T ₀Two neighborhoods separate, as Fig. 5 (b) with (c).Fig. 5 (d) has provided the optimum multicast tree of the example network that utilization the present invention obtains.Initial solution T ₀, T ₀Two neighborhoods separate and the coding of optimal solution as shown in table 3.

The coding situation that table 3 is separated

	Coding	cost(T)	Δ T	δ T
					Initial solution T 0	(F，D，C，G，B)∪(F，D)∪(F，D，E)∪(F，H)	45	5	3
T 0Neighborhood separate	(F，D，C，G，B)∪(F，D)∪(F，H，E)∪(F，H)	38	5	3
					T 0Neighborhood separate	(F，D，C，G，B)∪(F，D)∪(F，A，E)∪(F，H)	39	5	3
Finally separate	(F，D，C，G，B)∪(F，D)∪(F，A，E)∪(F，A，E，H)	31	5	3

At present, we by experiment simplation verification the present invention have good performance.In evaluation procedure, the employing random network topological method generation node number of degrees are 4 topological network figure, and the rectangular area size of network node random distribution is set at 4000km * 2400km, and the capacity of every link is 155Mbps.Whether there is link probability function P (u between any two node u and the v, v)=β exp (l (u, v)/and α L) determine that wherein (u v) is u to l, Euler's distance between the v, L is any two internodal ultimate ranges, and less α will increase the density on short chain road, and bigger β will cause higher chain direction density, we get α=0.15, β=2.2.The chain-circuit time delay of network is defined as the propagation delay time of link, ignores queuing delay and transmission delay, and transmission speed is 2/3 of the light velocity, and link cost is random distribution between [10,120] Mbps.The experiment number of each simulation points is 200, produces network topology and source/destination node, assembly average then at random at every turn.

We choose existing DVMA, SPT (distributed Bellman-Ford shortest path tree method, Bellman RE, Dynamic programming, Princeton University, Princeton, NJ, 1957.), DDVBM and DDVMBRA method and the method for the invention compare.The cost performance that has at first compared multicast tree that this several method obtains, wherein, Fig. 6 is that fixedly the destination node number is 5, Δ=35ms, under δ=20ms situation, the relation of number of network node and cost.Fig. 7 is that fixed network node number is 50, Δ=35ms, and under δ=20ms situation, the relation of destination node number and cost.Fig. 6 and Fig. 7 two width of cloth figure show, compare with DVMA, SPT, DDVBM, DDVMBRA method, and the multicast tree cost that this method obtained is minimum, and the cost performance is best.As can be seen from Figure 6, the cost of all methods multicast tree of trying to achieve increases along with the increase of number of network node.This is that the path that source node arrives each destination node increases thereupon owing to the increase along with network size, thereby makes the cost of tree increase.As can be seen from Figure 7, the cost of all methods multicast tree of trying to achieve increases along with the increase of destination node number.This is because along with the increase of destination node number, the scale of multicast tree will become greatly, thereby takies more resources, causes corresponding cost to increase.

Similar above-mentioned experimental technique, performances such as the time delay of multicast tree that these multi-broadcast routing methods obtain, delay variation, route success rate and convergence that we have gone back simplation verification, all experiments show, this method has all shown good performance at aspects such as time delay, delay variation and route success rates, and this method has convergence rate faster, satisfies the requirement of real time communication.

Claims (3)

1, a kind of dynamically distributed multicast routing method based on simulated annealing is characterized in that comprising initial solution construction process and optimal solution construction process, promptly

1.1 the step of described initial solution construction process is as follows:

1.1.1 the preceding k bar that uses distributed k-Bellman-Ford k bar shortest-path method to calculate from source node s to each destination node m ∈ M prolongs the path the most in short-term, as the alternative path collection, wherein M is the destination node collection, claims the multicast group again;

1.1.2 select a destination node at random, and add alternative path concentrated source node s to initial multicast tree T to the path of prolonging the most in short-term of this destination node ₀In, T wherein ₀Initial condition be an empty tree;

1.1.3 concentrate to select to satisfy formula max{0, maxd-δ from source node s destination node m outside tree at alternative path }≤d (p (s, m))≤min{mind+ δ, Δ } article one alternative path, this m ∈ M, m  T ₀, and add it to T ₀In, wherein (p (s, m)) represents the path delay of time from source node s to destination node m to d; Maxd and mind represent respectively from source node s to T ₀In the maximum delay and the minimal time delay in path in the path of the destination node that comprises, Δ and δ are respectively the time delay upper limit and the delay variation upper limit of multicast conversation;

1.1.4 repeating step 1.1.3, all destination nodes in M all are included in T ₀In, so far the initial solution construction process finishes, the initial multicast tree T of this procedure construction ₀Will be as the initial solution of optimal solution construction process;

1.2 the step of described optimal solution construction process is as follows:

1.2.1 T is set _Best=T _Now=T ₀, T wherein _BestRepresent the multicast tree of cost minimum up to the present, i.e. optimal solution, T _NowRepresent current multicast tree, promptly current separating;

1.2.2 set initial temperature t=t ₀

1.2.3, then enter step 1.2.8 as if loop termination criterion in this temperature reaches; Otherwise enter next step 1.2.4;

Produce the current T of separating 1.2.4 separate the structure process with neighborhood _NowNeighborhood disaggregation N (T _Now), and therefrom select a multicast tree T at random _Next, calculate Δ c=cost (T _Next)-cost (T _Now), cost (T wherein _Next) and cost (T _Now) represent multicast tree T respectively _NextWith the current T that separates _NowCost;

If 1.2.5 Δ c≤0 or exp (Δ c/t)＞random (0,1), T _Now=T _Next

If 1.2.6 cost (T _Now)＜cost (T _Best), cost (T wherein _Best) expression optimal solution cost, T _Best=T _Now

1.2.7 repeating step 1.2.3～step 1.2.6;

1.2.8 move back temperature, t=λ t, wherein λ is for moving back warm speed; If satisfy termination rules, stop calculating, otherwise get back to step 1.2.3; So far the optimal solution construction process finishes, and the output of this process is exactly a minimum cost multicast tree that finally satisfies time delay and delay variation constraint.

2, the dynamically distributed multicast routing method based on simulated annealing according to claim 1 is characterized in that: neighborhood is separated the structure process and is constructed the neighborhood disaggregation by switching path in the feasible solution scope, and its step is as follows:

Step 1: at current multicast tree T _NowIn select a leaf destination node m at random;

Step 2:, delete current multicast tree T from destination node m _NowIn lead to the link of source node s, up to running into another destination node or source node, form stalk tree T _Sub

Step 3: calculate current subtree T _SubIn maximum delay maxd and minimal time delay mind from source node s to each destination node path;

Step 4: concentrate at alternative path, with time delay between [max{0, maxd-δ }, min{mind+ δ, Δ }] and the alternative path from source node s to destination node m join current subtree T _SubIn; If the new tree that forms is acyclic, then become current multicast tree T _NowA neighborhood separate;

Step 5: repeating step 4, all checked up to all k bar alternative paths.

3, the dynamically distributed multicast routing method based on simulated annealing according to claim 1, it is characterized in that: satisfy a minimum cost multicast tree of time delay and delay variation constraint at structure after, the situation that the multicast member dynamically adds or leave multicast conversation can appear, at this moment need the multicast tree destination node of dynamically recombinating, and the multicast tree after guaranteeing to recombinate still satisfies time delay and delay variation retrains, and described dynamic reconfiguration method is as follows:

When destination node member m ∈ M request " leaving " multicast group, in two kinds of situation: (1) if destination node m is not a leaf node, this situation is not made any change to multicast tree, as long as allow the user forwarding packet data of this node stop to oneself; (2) if destination node m is a leaf node, it sends leave request message to the source node direction of multicast tree so, and a node connects a node, arrives source node or another destination node up to this message; Message the link of process with deleted;

When new node v  M request " adding " multicast group, it sends join request message to source node so, divide three kinds of situations: (1) is not if v is the node in the tree, the preceding k bar that then obtains from source node to node v prolongs the path the most in short-term, and select article one time delay at [max{0, maxd-δ }, min{mind+ δ, Δ] between the path join in the current multicast tree; If can't find such path, then abandon joining request of node v; (2) if v be the tree in node, and new multicast group M ∪ { v} satisfies delay variation upper limit δ, and then this multicast tree itself is exactly a feasible tree; (3) if v is a tree node, and new multicast group M ∪ { v} does not satisfy delay variation upper limit δ, and the intermediate node that node v must be a source node to one or more destination nodes path is described; In this case, delete the path that this comprises v and these destination nodes, again node v and these destination nodes are joined in the routing tree successively.

Images