CN102664784A - Virtual network mapping method capable of realizing adaptive equalization of weight of node link pressure - Google Patents

Abstract

The invention provides a virtual network mapping method capable of realizing adaptive equalization of weight of node link pressure. By adopting the method, the optimization objective mapped by a virtual network node is real-timely adjusted according to the pressure state of a current physical network node link, so that adaptive equalization between the pressure of the node in the optimization objective and the weight of the link pressure is realized, and a detecting mechanism is used to guarantee that the adaptive equalization of the weight is not diverged, therefore, the virtual network mapping result has the characteristic of comprehensive optimization of node link pressure. The method can enhance the virtual network operation stability, and has great significance in practical application of the network virtualization technology in network operation.

Description

A kind of virtual network mapping method of node link pressure weight adaptive equalization

Technical field

The network virtualization technology is one of important method that promotes the Internet architectural framework development; Its essence is on a public physical network, to run a plurality of virtual subnets through abstract, distribution, isolation mech isolation test independently; Each virtual subnet can use separate protocol architecture; And can carry out reasonable disposition to node and link circuit resource in the whole network according to the demand of user's dynamic change; Thereby strengthen network more flexible and diversity, realize the measurable and controllable property of network, the distribution of peak optimizating network resource and scheduling; Improve safety with service quality, reduce operation maintenance cost, in the hope of essence ground solve the Internet existing rigid, with patch be updated to main current situation.

The network virtualization technology can be used to the new network Research of structure basis of sharing the Physical Experiment network is provided; It can also be separated bottom physical facility provider and network service provider simultaneously; Allow the network of a plurality of operators to share same public bottom physical network architecture (link, switching node etc.); Each network all has the Internet resources share that can adjust flexibly again by other web influences therein; Heterogeneous networks operator can adopt the various network agreement, and the end-to-end service of innovation is provided, so network virtualization also gets a good chance of becoming a kind of main flow operation mode of future network.

Background technology

The virtual network mapping problems then is a requisite link in the network virtualization technology; Its major function is that user's virtual network request (Virtual Request) is reasonably mapped to the bottom physical network facility (Substrate Network) that operator provides; Mapping process not only will be realized the separation between the virtual network and be independent of each other; Thereby guarantee the service quality (QoS) of each virtual network user; Also to try one's best simultaneously and reasonably distribute the bottom physical network resource, improve resource utilization.

The virtual network mapping problems can further be subdivided into node mapping and link maps two parts, owing in the mapping process node cpu capacity and link bandwidth are had many restrictive conditions, so the virtual network mapping problems is a NP-hard problem.For guaranteeing that this way to solve the problem has engineering practice property, the researcher has mainly used the heuritic approach of suboptimum both at home and abroad, and has proposed the mapping model based on time window.The time window model is to be the empty net request of unit online treatment with the time window with mapping problems; The current all void net mapping request of statistics in each time window; And use mapping algorithm to shine upon, to shining upon successful request, corresponding renewal bottom physical network state; The lost request that loses of mapping is put into waiting list with request; Or directly refuse this request after satisfying certain condition.This process is as shown in Figure 1.

In void net mapping problems, node mapping and link maps are generally carried out respectively according to precedence, and complexity is reduced, but owing to coupling between two steps is not enough, the reduction that also can bring performance.The main stream approach of tackling this problem is through in the node mapping step, adjusting optimization aim, look after the characteristics of follow-up link maps, thereby realizes two mapping step ground couplings effectively.This research also will be adopted this starting point, through flexible adjustment node mapping optimization target to realize this specific objective of node link isostasy of virtual network.Load balancing in the network virtualization mapping mainly is meant the maximum pressure that as far as possible reduces empty net mapping posterior nodal point and link, thereby reduces the variance of network pressure, and the pressure that makes the network various piece is more near average pressure.In research to this problem; The method of difference equalizing section point pressure and link pressure has at first been considered in existing work: in only considering the balanced research of node pressure; Because the preferential little node of selection pressure tends to cause euclidean distance between node pair far away; Therefore a virtual link just needs many physical links to realize, so the link pressure build-up is bigger, and mapping efficient and success rate are also lower; In the research of only considering link pressure, same because to not considering node pressure, possibly cause node pressure excessive or to be mapped to power low.Therefore; In order to solve this contradiction, should take all factors into consideration two mapping optimization targets of node link pressure, comprise based on the result of study of this thinking: the one, directly make up the formula of two optimization aim with the mode of summation or product; Two factors all are taken into account in optimizing process like this; But both equilibrium relations are also indeterminate, possibly cause one of them optimization aim dominate all the time, and the effect of another optimization aim are difficult to manifest; The 2nd, dynamically select to use node or link pressure as optimization aim according to the current network pressure state, the problem of doing like this is the state that maintains single optimization aim all the time, so the complex optimum effect is difficult to ensure.

In sum, in the research that realizes good coupling between the node link pressure optimized target, still have a lot of problem values to inquire into, especially need the adjustable optimization aim equilibrium strategy of a kind of efficient stable, to realize the equilibrium of whole network node link pressure.

Summary of the invention

It is the complex optimum situation of optimization aim with node and link pressure simultaneously that the present invention has analyzed in the empty net mapping process, finds to exist trade-off relation between the two, promptly is reduced to optimization aim with node pressure and will causes link pressure to raise, and vice versa.Therefore, in order to realize two equilibrium relations between the optimization aim, need the adjustable complex optimum target function of a kind of weight of design; Simultaneously, in order to adapt to the topological diversity of request and the fluctuation of request arriving rate, being provided with of this parameter needs the dynamic adjustable joint; At last, in order to guarantee the stability of system, need certain measure to guarantee the convergence of this dynamic adjustable joint parameter.

The present invention is according to this starting point; Designed the balanced virtual network mapping method of a kind of node link pressure adaptive; This method has been used the adjustable complex optimum target of balance parameters, not only can realize the complex optimum to node pressure, link pressure, can also regulate weight relationship between the two according to mapping result; Two optimization aim fully are coupled, thereby realize the aggregative equilibrium of the whole network pressure; In addition, we have designed a judgment mechanism to adjustable parameter convergence direction, disperse to prevent adjustable parameter, realize convergence fast, thus the stability of the system of assurance.

The definition that the present invention relates to:

1) node link pressure

is meant the pressure of bottom physical network nodes i, by total CPU capacity of this node and the decision of residue CPU capacity;

is meant the pressure of bottom physical network links j, determined by this link total bandwidth capacity and remaining bandwidth capacity:

$S_i^n=1-R_i^n/C_i^n---\left(1\right)$

$S_j^l=1-R_j^l/C_j^l---\left(2\right)$

Wherein,

and

refers to the CPU capacity and total CPU capacity of this physics net node i current residual respectively, and

and

refers to the bandwidth capacity and the total bandwidth capacity of this physics network chain road j current residual respectively.

2) average pressure and maximum pressure

With reference to the definition of upper node link pressure, obtain being defined as of average pressure and maximum pressure:

$\stackrel{\‾}{S^n}=\frac{4}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{S}_i^n---\left(3\right)$

$\stackrel{\‾}{S^l}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\Σ}}}{S}_j^l---\left(4\right)$

$S_{\max }^n=\max \left\{S_i^n\right\}$ (i∈{1，2，...，N}) (5)

$S_{\max }^l=\max \left\{S_j^l\right\}$ (j∈{1，2，...，N}) (6)

Wherein, N, L refer to physics net node link quantity respectively.

3) empty network planning mould

Empty network planning mould has mainly been described the size of virtual network (VN), is determined by its node cpu capacity and link bandwidth:

$\mathrm{Scale}\left({\mathrm{VN}}_k\right)=\underset{i\∈{\mathrm{VN}}_k}{\mathrm{\Σ}}{\mathrm{CPU}}_i^n+\underset{j\∈{\mathrm{VN}}_k}{\mathrm{\Σ}}{\mathrm{BW}}_j^l---\left(7\right)$

Wherein

is meant the CPU capacity of node i, and

is meant the bandwidth of link j.

3) node scale (H ⁿ(i))

The node scale has mainly been described the significance level of node in network, by the CPU capacity of this node be connected the bandwidth decision:

$H^n\left(i\right)={\mathrm{CPU}}_i^n\underset{j\∈L\left(i\right)}{\mathrm{\Σ}}{\mathrm{BW}}_j^l---\left(8\right)$

Wherein

is meant the CPU capacity of node i;

is meant the bandwidth of link j, the link set that L (i) expression and node i directly link to each other.

According to above-mentioned definition, the present invention one has proposed the balanced virtual network mapping target of node link pressure adaptive, and this target can realize making overall plans and coordinate of node optimization and link optimization, thereby reaches the effect of overall pressure optimized; The 2nd, the adjustable parameter in the target function is designed, a kind of adaptive parameter prediction pattern has been proposed, weight parameter is optimized and revised according to network state variation and two factors of self evolution gradually; The 3rd, in order to guarantee the stability of this self adaptation adjustable parameter; A kind of mechanism that adjustable parameter convergence direction is judged has been proposed; This mechanism can judge whether system gets into divergent state through the time that the statistical weight parameter stops on the span border, and adopts corresponding strategy to make system recover balanced again.

(1) the void net of node link pressure weight adaptive equalization mapping target:

As indicated above; Existing is that the void net mapping algorithm of target has used summation or asked the simple combination optimization aim of product with the node link isostasy; Or the method for using two optimization aim to call in turn by demand; These methods are all comparatively simple, can't really embody the relation between the node link pressure, thereby make the optimization work of the whole network isostasy thorough inadequately.Therefore the present invention has at first proposed the void net mapping optimization target of a node link pressure weight adaptive equalization, promptly minimizes:

$H_{\mathrm{stress}}\left(i\right)=[\mathrm{\α}{S}_i^n+(1-\mathrm{\α})\underset{j\∈L\left(i\right)}{\mathrm{\Σ}}{S}_j^l][\underset{u\∈N_A}{\mathrm{\Π}}d(i,u)+1]---\left(9\right)$

In the following formula, the link set that L (i) expression and node i directly link to each other, like this, this optimization aim has just been taken all factors into consideration the pressure of node i, and the link pressure that is connected with node i, and α is a weight adjusting parameter, α ∈ (0,1).N _ARepresent to have shone upon the successful pairing physics net of void net node node set in the request of this void net; D (i, the u) distance (jumping figure) between 2 of expression i, the u, like this; Distance factor also becomes one of major influence factors of this optimization aim, and distance closely will reduce the difficulty of follow-up link maps.

(2) adaptive forecasting method of weight parameter α:

For the weight that makes node pressure and link pressure can reach balanced, and along with quantity, scale and the arrival rate of the request of void net are constantly adjusted optimization, the present invention proposes and a kind of weight parameter α is carried out the method for adaptive prediction, promptly

$\mathrm{\α}={\mathrm{\α}}^{-}[1+\mathrm{\β}[(S_{\max }^n-\stackrel{\‾}{S^n})-(S_{\max }^l-\stackrel{\‾}{S^l})\left]\right]---\left(10\right)$

In the following formula, α ^-Represent the α value of last time window, β representes the convergence rate of α, when void net solicited status changes when very fast, can increase β to accelerate the convergence rate of α.δ is the real number greater than 0; Be used to guarantee that denominator is non-vanishing;

and

then representes the difference of maximum node link pressure and average nodal link pressure respectively, and mean value is used to guarantee that the system pressure benchmark can constantly adjust along with the variation of void net request.If maximum node pressure is bigger with the gap of average link pressure than maximum link pressure with the gap of average nodal pressure; Explain that the node equilibrium is relatively poor; Then α increases; Cause the weight of node pressure in (7) formula bigger; The node that node pressure is little like this can preferentially obtain selecting, and has therefore reduced maximum node pressure, has accomplished the adaptive equalization of system; Vice versa.The initial value of α is set to [1-(1/D)], and wherein D representes the average connection degree of physical network nodes, and β is set to 0.1, to guarantee the stability of system.

(3) system mode is dispersed and is suppressed mechanism:

In last branch, the adjustment process of weight parameter α is an adaptive regression process, and when running well in system, the convergence of this parameter can be protected.But mapping problems is a complication system; No matter be physical network topology, state, or virtual network topology, state, all can be to shining upon it to influence; For fear of occurring failing in time to bring network pressure to regulate under special circumstances according to expection to the adjusting of α; Be the situation that system is in divergent state, the present invention has designed a system mode and has dispersed inhibition mechanism, promptly when the situation that detects α＜0 or α＞1 occurs; α is reset and is initial value, so just can guarantee the normal operation of system.

Description of drawings

Virtual network mapping flow process under Fig. 1 time window pattern

The substep mapping method of Fig. 2 node link pressure weight adaptive equalization

Execution mode

Concrete operations flow process of the present invention is before each node mapping, and its optimization aim is made adaptive equalization, takes into full account follow-up link maps in the node mapping process thereby be implemented in, and reaches the mutual equilibrium of physical network nodes link pressure.Idiographic flow is as shown in Figure 2:

A. add up all empty net requests in this time window, be designated as set R ^v

B. if R ^vBe sky, get into step G.If R ^vBe not empty, choose the maximum void net request VN of current empty network planning mould _k, statistics VN _kIn all empty net nodes, be designated as set N ^v

C. if N ^vBe sky, then the node mapping finishes, and gets into step e; If N ^vBe not empty, then choose the largest void net node of node

From the bottom physical network, select residue CPU greater than node The physical network nodes set of CPU, be designated as N ^s

D. if N ^sBe sky, then should void net request mapping fail, get into step F; If N ^sBe not empty, the currency of regulating parameter alpha according to weight calculates N ^sIn the H of each physics net node i _Stress(i), choose H _Stress(i) the physics net node of minimum

And with the void net node of choosing among the step C

Map to this physics net node

On.Will

From N ^vStep C is returned in middle deletion.

E. use shortest path first to accomplish link maps, if the mapping failure is then sent this request into next time window or direct refusal; If shine upon successfully, then upgrade bottom physical network state, with VN _kFrom R ^vStep B is returned in middle deletion.

F. according to the α value of the next time window of the physical network state computation after upgrading, and use the judgment mechanism check and revise α value, the empty net of this time window shines upon end.

Claims (4)

1. the virtual network mapping method of a node link pressure weight adaptive equalization, carry out the virtual network mapping steps and comprise in a time window:

A. add up all empty net requests in this time window, be designated as set R ^v

B. if R ^vBe sky, get into step G.If R ^vBe not empty, choose the maximum void net request VN of current empty network planning mould _k, statistics VN _kIn all empty net nodes, be designated as set N ^v

C. if N ^vBe sky, then the node mapping finishes, and gets into step e; If N ^vBe not empty, then choose the largest void net node of node

From the bottom physical network, select residue CPU greater than node

The physical network nodes set of CPU, be designated as N ^s

D. if N ^sBe sky, then should void net request mapping fail, get into step F; If N ^sBe not empty, the currency of regulating parameter alpha according to weight calculates N ^sIn the H of each physics net node i _Stress(i), choose H _Stress(i) the physics net node of minimum

And with the void net node of choosing among the step C

Map to this physics net node

On.Will

From N ^vStep C is returned in middle deletion.

E. use shortest path first to accomplish link maps, if the mapping failure is then sent this request into next time window or direct refusal; If shine upon successfully, then upgrade bottom physical network state, with VN _kFrom R ^vStep B is returned in middle deletion.

F. according to the α value of the next time window of the physical network state computation after upgrading, and use the judgment mechanism check and revise α value, the empty net of this time window shines upon end.

2. the method for claim 1, wherein being defined as of node mapping optimization target:

$H_{\mathrm{stress}}\left(i\right)=[\mathrm{\α}{S}_i^n+(1-\mathrm{\α})\underset{j\∈L\left(i\right)}{\mathrm{\Σ}}{S}_j^l][\underset{u\∈N_A}{\mathrm{\Π}}d(i,u)+1]$

In the following formula, the link set that L (i) expression and node i directly link to each other, α is a weight adjusting parameter, α ∈ (0,1).D (i, u) distance (jumping figure) between 2 of expression i, the u.N _ARepresent to have shone upon the successful pairing physics net of void net node node set in the request of this void net. is meant the pressure of bottom physical network nodes i;

is meant the pressure of bottom physical network links j, defines as follows:

$S_i^n=1-R_i^n/C_i^n$

$S_j^l=1-R_j^l/C_j^l$

Wherein,

and

refers to the CPU capacity and total CPU capacity of this physics net node i current residual respectively, and

and

refers to the bandwidth capacity and the total bandwidth capacity of this physics network chain road j current residual respectively.

3. like claim 1 or the described method of claim 2, wherein being defined as of weight parameter α:

$\mathrm{\α}={\mathrm{\α}}^{-}[1+\mathrm{\β}[(S_{\max }^n-\stackrel{\‾}{S^n})-(S_{\max }^l-\stackrel{\‾}{S^l})\left]\right]$

In the following formula, α ^-Represent the α value of last time window, β representes the convergence rate of α.δ is the real number greater than 0; Be used to guarantee that denominator is non-vanishing; and then representes the difference of maximum node link pressure and average nodal link pressure respectively, and mean value is used to guarantee that the system pressure benchmark can constantly adjust along with the variation of void net request.The initial value of α is set to [1-(1/D)], and wherein D representes the average connection degree of physical network nodes, and β is set to 0.1, to guarantee the stability of system.

4. the method for claim 1, wherein the judgment mechanism of step G is meant: when the situation that detects α＜0 or α＞1 occurred, α was reset and is initial value.

Images