CN105721577A - Software defined network-oriented server load balancing method - Google Patents

Abstract

The invention discloses a server load balancing method facing a Software Defined Network (SDN), which is realized by using a modularization idea in a Software Defined Network (SDN) environment and sequentially comprises the following steps: the port detection module periodically counts the real-time flow of the SDN switch port connected with each server, and calculates the flow rate and the port predicted flow rate of the next test time point; the port flow velocity analysis module analyzes the flow rate of each port obtained from the port detection module and judges whether load balancing is needed or not; and the SDN controller issues a flow table to distribute new service flow to the specified server. The method comprehensively considers the real-time flow rate and the predicted flow rate and combines the structural advantages of SDN control and forwarding separation, and has better load balancing capability and efficiency.

Description

A kind of server load balancing method of software-oriented definition network

Technical field

The invention belongs to technical field of the computer network, be specifically related to the server load balancing method of a kind of software-oriented definition network.

Background technology

In traditional TCP/IP network struture system, network equipment logic control and data forward close coupling, lack motility, expand rapidly along with network size and expose variety of problems and challenge.People are only by being continuously increased new procotol or being added in the stub network equipment such as switch and router by various sophisticated functions.These method for repairing and mending adds additional the complexity of basic unit's equipment, increases the redundancy of network transaction data and the management difficulty of network.The appearance of SDN concept and correlation technique is precisely in order to overcome disadvantage mentioned above.

SDN is the new network framework of a kind of numerical control separation, Network Programmable.It is divided into application layer, key-course, forwarding from top to bottom.Application layer mainly provides a user with various service；Key-course is mainly made up of several SDN controllers, is used for being responsible for processing the data resource collected, safeguards network topology, information state etc.；Forwarding primary responsibility processes based on the data of stream table, forwards and state collection.In this separation architecture, technical staff can utilize the SDN controller of key-course to formulate efficiently, neatly by the mode programmed to meet the forwarding strategy of business demand or test new procotol.

Load balancing is primarily referred to as in network and in the way of concurrent or independent access, mass data stream is distributed to multiple stage node device so that it is process information simultaneously.The wait corresponding time of user is so greatly reduced, thus improve the disposal ability of system.Its essence is through carrying out packet data dynamically adding up and analyzing in real time, according to analyzing result, data stream is reasonably assigned on multiple stage node device.Load balancing strengthens the disposal ability of data in network, improves the performance of server cluster.

Z-score is also standard score, and calculating process is that the number difference with average is again divided by standard deviation.Z-score can reflect a given mark range averaging number and have how many standard deviations, represents mark relative position amount number in group in units of standard deviation.

Time series forecasting is a kind of to consider variable development and change rule in time and make the Forecasting Methodology extrapolated with the conventional statistics founding mathematical models of this variable.

Single Exponential Smoothing is the one of time series forecasting, is used as the prediction of middle or short term economic trend in statistics, the flow in network can be carried out short-term forecast in network communication field.

Summary of the invention

Goal of the invention: it is an object of the invention to solve the deficiencies in the prior art, the server load balancing method of a kind of software-oriented definition network is provided, by each port flow rate information of the switch of Connection Service device cluster is judged, calculate the designated port of assignment of traffic, thus reaching server cluster is realized the purpose of load balancing, efficient port flow rate detection and prediction and the shunting to server cluster can be realized by the present invention.

Technical scheme: the server load balancing method of a kind of software-oriented definition network of the present invention, comprises the following steps successively:

(1) port detecting module periodic statistical goes out the SDN switch port real-time traffic that each server connects, and calculates flow rate and the port prediction flow velocity of point of next testing time；

(2) each port flow speed that the analysis of port flow velocity analysis module obtains from port detecting module, it may be judged whether need to carry out load dispatch；

(3) load balancing module is responsible for the access request of new user and the load dispatch to the unbalance server cluster of flow velocity, and SDN controller adds stream table and sets up new session with the server specified or revise the server that original session is deployed to specify by stream table.

Further, described step (1) method particularly includes:

(1.1) SDN controller regularly sends port statistics request message to all SDN switch in network by port detecting module and obtains port information, port statistics response message is sent to SDN controller by safe lane by SDN switch, controller is by the real-time traffic of the response message collection of each switch ports themselves of Connection Service device to each port the flow velocity portspeed (i calculating each port, t), wherein i represents the i-th port of SDN switch；

(1.2) according to the flow velocity portspeed of each server connectivity port, (i t), uses the Single Exponential Smoothing in statistics to calculate the port prediction flow rate F (i, t+1) of point of next testing time.

Further, described step (2) method particularly includes:

(2.1) flow rate that the analysis of port flow velocity analysis module obtains from port detecting module, calculates the meansigma methods of each port flow velocityVariance VarianceSpeed (t), standard deviation sigma_portspeed(t)；By the flow speed value portspeed of relatively each port, (i t) calculates Peak Flow Rate MaxPortspeed (t) that each port takies；Meansigma methods by Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, calculate flow velocity imbalance values P (t) according to the z-score method in statistics；According to current flow value and flow velocity predictor calculation go out mixture velocity value Q (i, t)；

(2.2) P (t) threshold value PThreshold and bandwidth threshold BwThreshold is set according to the actual demand of network, when P (t) value reaches P (t) threshold value PThreshold or Peak Flow Rate MaxPortspeed (t) reaches bandwidth threshold BwThreshold, server cluster needs to carry out load dispatch.

Further, described step (3) method particularly includes:

(3.1) when there being new main frame to access server cluster, load balancing module selects mixture velocity value Q, and (i, t) minimum port is as designated port；Load balancing module judges according to the object information of port flow velocity analysis module every 15s, if server cluster needs to carry out load balancing, (i, port t) is set to designated port will to have minimum mixture velocity value Q；

(3.2) SDN controller issues stream table to the SDN switch port specified, and sets up session or by original session synchronization to given server, it is achieved the load balancing to server cluster between main frame and given server.

Further, each port flow velocity portspeed in described step (1.1) (i, defining method t) is:

(1.1.1) determine flow velocity portspeed (i, t):

Portspeed (i, t)=(portstraffic (i, t)-portstraffic (i, t-1))/intervaltime,

Wherein (i, t) represents the i port flow value of t to portstraffic, and portstraffic (i, t-1) represents the i port flow value in t-1 moment, and intervaltime represents the interval of two testing time points.

Further, in described step (1.2), the defining method of the port prediction flow rate F (i, t+1) of next testing time point is:

(1.2.1) Single Exponential Smoothing in statistics is used to determine some port flow velocity predictive value F of the next testing time (i, t+1):

$\left\{\begin{array}{c}F(i,t+1)=\mathrm{\α}*portspeed(i,t)+(1-\mathrm{\α})*F(i,t)\\ F(i,2)=portspeed(i,2)\end{array}\right.,$

Wherein portspeed (i, t) for the flow speed value of t, F (i, t) for the flow velocity predictive value in t-1 moment, wherein the flow velocity predictive value of the 1st time period and the flow speed value of the 1st time period are equal, and α is smoothing constant, and span is [0,1], α value is set according to real network flow condition, generally in order to make flow velocity predictive value reflect up-to-date changing value sensitively, bigger α value should be taken；If reflecting this seasonal effect in time series long-term forecast value, less α value should be taken.

Further, in described step (2.1), the defining method of P (t) value is:

(2.1.1) by the flow speed value portspeed of each port, (i t) determines the meansigma methods of each port flow velocityPeak Flow Rate MaxPortspeed (t):

$\stackrel{\‾}{Portspeed\left(t\right)}=\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}portspeed\right(i,t\left)\right)/n,$

$MaxPortspeed\left(t\right)=\underset{i=1}{\overset{n}{Max}}\left(portspeed\right(i,t\left)\right),$

WhereinFor all of the port flow velocity and, it is assumed that have n port；

(2.1.2) by the flow speed value portspeed of each port, (i, t) with the meansigma methods of port flow velocityDetermine the variance VarianceSpeed (t) of each port flow velocity, standard deviation sigma_portspeed(t):

(2.1.3) by the meansigma methods of port Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, determine P (t) value according to the z-score method in statistics:

Z-score is for calculating given how many standard deviations of mark range averaging number in statistics, mark relative position amount number of present position in group is represented in units of standard deviation, use z-score to calculate port Peak Flow Rate value relative position amount number in all of the port flow velocity, reflect whether overall flow velocity is in the state of equilibrium to a certain extent；

(2.1.4) utilize port flow speed value portspeed (i, t) and port flow velocity predictive value F (i, t+1) determine each port mixture velocity value Q (i, t):

Q (i, t)=β * portspeed (i, t)+(1-β) * F (i, t+1), 1 < i < n

Wherein (i, t) for each port flow speed value, F (i, t+1) is each port flow velocity predictive value at next testing time point to portspeed.The selection of β is determined as the case may be, and β is more big, and (i, value t) is more subject to the impact of current flow to Q；β is more little, and (i, value t) is more subject to the impact of flow velocity predictive value to Q.

Beneficial effect: the present invention utilizes knowledge of statistics, mixture velocity information according to each port of SDN switch being connected with server, dynamically by server extremely suitable to user's request and conversation dispatching, achieve the load balancing of server, real-time flow rate and predicted velocity are considered and combines SDN and control and forward the framework advantage being separated, there is good load balance ability and efficiency.Compared with prior art, the invention have the advantages that

(1) present invention adopts modularization programming, extract the switch ports themselves flow rate information under SDN framework, use the calculated flow velocity imbalance values of z-score, port Peak Flow Rate jointly to judge that server cluster is the need of carrying out load balancing, server cluster flow velocity overload is had flexibly and reacts timely.

(2) the single exponential smoothing predicted method in statistics is applied in the selection calculating of port by the present invention, compares traditional dynamic load balancing method, it is possible to network flow velocity carries out certain anticipation and valuation, thus improve the effect of load balancing.

Accompanying drawing explanation

Fig. 1 is the load-balancing method flow chart in the present invention；

Fig. 2 is that the load-balancing method in the present invention implements network diagram.

Detailed description of the invention

Below technical solution of the present invention is described in detail, but protection scope of the present invention is not limited to described embodiment.

The present invention is applicable to software defined network (SoftwareDefinedNetwork, it is abbreviated as SDN) in environment, use each port flow velocity of switch obtaining Connection Service device cluster on SDN controller and adopt the z-score in statistics and Single Exponential Smoothing to achieve the server load balancing under SDN environment.

As it is shown in figure 1, the present invention newly added three modules in SDN controller, it is port detecting module, port flow velocity analysis module, load balancing module respectively.OpenFlow agreement is most important south orientation agreement in SDN framework, and port statistics request message and port statistics response message are the important component parts in OpenFlow agreement.

Port detecting module in the present invention regularly sends port statistics request message to each switch in network, port statistics response message is sent to controller by safe lane by SDN switch, and controller obtains the flow of each port by the response message of each switch ports themselves of Connection Service device and calculates real-time flow rate value and flow velocity predictive value.Port flow velocity analysis module calculates maximum port flow speed value and flow velocity imbalance values by the real-time flow rate value of each port, and goes out server cluster the need of carrying out load balancing by the Analyzing on Size of the two parameter value.If needing load balancing, each port real-time flow rate value that load balancing module provides according to port detecting module and flow velocity predictor calculation go out the mixture velocity value of each port, will have the switch ports themselves that port label is flow to be allocated of minimum mixture velocity value；When there being new main frame to access server cluster, load balancing module selects the minimum port of mixture velocity value as the switch ports themselves specified.SDN controller obtains the server specified by the switch ports themselves specified, and adds stream table and sets up session or amendment stream table by original session synchronization to the server specified, complete the load balancing to server cluster between main frame and given server.

The specific works flow process of the present invention is:

(1) port detecting module periodic statistical goes out the SDN switch port real-time traffic that each server connects, and calculates flow rate and the port prediction flow velocity of point of next testing time.

(1.1) SDN controller regularly sends port statistics request message to all SDN switch in network by port detecting module and obtains port information, and port statistics response message is sent to controller by safe lane by SDN switch.Controller is by the real-time traffic of the response message collection of each switch ports themselves of Connection Service device to each port and calculate the flow velocity portspeed of each port (i, t), wherein i represents the i-th port of SDN switch.

(1.2) according to the flow velocity portspeed of each server connectivity port, (i t), uses the Single Exponential Smoothing in statistics to calculate the port prediction flow rate F (i, t+1) of point of next testing time.

(2) each port flow speed that the analysis of port flow velocity analysis module obtains from port detecting module, it may be judged whether need to carry out load balancing.

(2.1) flow rate that the analysis of port flow velocity analysis module obtains from port detecting module, calculates the meansigma methods of each port flow velocityVariance VarianceSpeed (t), standard deviation sigma_portspeed(t)；By the flow speed value portspeed of relatively each port, (i t) calculates Peak Flow Rate MaxPortspeed (t) that each port takies；Meansigma methods by Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, calculate flow velocity imbalance values P (t) according to the z-score method in statistics；According to current flow value and flow velocity predictor calculation go out mixture velocity value Q (i, t).

(2.2) P (t) threshold value PThreshold and bandwidth threshold BwThreshold is set according to the actual demand of network, when P (t) value reaches P (t) threshold value PThreshold or Peak Flow Rate MaxPortspeed (t) reaches bandwidth threshold BwThreshold, server cluster needs to carry out load dispatch.

(3) load balancing module is responsible for the access request of new user and the load dispatch to the unbalance server cluster of flow velocity, and SDN controller adds stream table and sets up new session with the server specified or revise the server that original session is deployed to specify by stream table.

(3.1) when there being new main frame to access server cluster, load balancing module selects mixture velocity value Q, and (i, t) minimum port is as designated port；Load balancing module judges according to the object information of port flow velocity analysis module every 15s, if server cluster needs to carry out load balancing, (i, port t) is set to designated port will to have minimum mixture velocity value Q；

(3.2) SDN controller issues stream table to the SDN switch port specified, and sets up session or by original session synchronization to given server, it is achieved the load balancing to server cluster between main frame and given server.

In the present invention, each port flow velocity portspeed in step (1.1) (i, defining method t) is:

(1.1.1) determine flow velocity portspeed (i, t):

Portspeed (i, t)=(portstraffic (i, t)-portstraffic (i, t-1))/intervaltime,

Wherein (i, t) represents the i port flow value of t to portstraffic, and portstraffic (i, t-1) represents the i port flow value in t-1 moment, and intervaltime represents the interval of two testing time points.

In the present invention, in step (1.2), the defining method of the port flow rate F (i, t+1) of next testing time point is:

(1.2.1) Single Exponential Smoothing in statistics is used to determine some port flow velocity predictive value F of the next testing time (i, t+1):

$\left\{\begin{array}{c}F(i,t+1)=\mathrm{\&alpha;}*portspeed(i,t)+(1-\mathrm{\&alpha;})*F(i,t)\\ F(i,2)=portspeed(i,2)\end{array}\right.,$

Wherein (i, t) for the flow speed value of t, F, (i, t) for the flow velocity predictive value in t-1 moment for portspeed.Wherein the flow velocity predictive value of the 1st time period and the flow speed value of the 1st time period are equal.α is smoothing constant, and span is [0,1], arranges α value according to real network flow condition in the present invention.General in order to make flow velocity predictive value reflect up-to-date changing value sensitively, bigger α value should be obtained；If reflecting this seasonal effect in time series long-term forecast value, less α value should be taken.

In the present invention, in step (2.1), the defining method of P (t) value is:

(2.1.1) by the flow speed value portspeed of each port, (i t) determines the meansigma methods of each port flow velocityPeak Flow Rate MaxPortspeed (t):

$\stackrel{\&OverBar;}{Portspeed\left(t\right)}=\left(\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}portspeed\right(i,t\left)\right)/n,$

$MaxPortspeed\left(t\right)=\underset{i=1}{\overset{n}{Max}}\left(portspeed\right(i,t\left)\right),$

WhereinFor all of the port flow velocity and, it is assumed that have n port.

(2.1.2) by the flow speed value portspeed of each port, (i, t) with the meansigma methods of port flow velocityDetermine the variance VarianceSpeed (t) of each port flow velocity, standard deviation sigma_portspeed(t):

$VarianceSpeed\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(portspeed\right(i,t)-\stackrel{\&OverBar;}{Portspeed\left(t\right)})/n,$

${\mathrm{\&sigma;}}_{portspeed\left(t\right)}=\sqrt{VarianceSpeed\left(t\right)}.$

(2.1.3) by the meansigma methods of port Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, determine P (t) value according to the z-score method in statistics:

$P\left(t\right)=\left(MaxPortspeed\right(t)-\stackrel{\&OverBar;}{Portspeed\left(t\right)})/{\mathrm{\&sigma;}}_{portspeed\left(t\right)}.$

In the present invention, mixture velocity value Q in step (2.1) (i, defining method t) is:

(2.1.4) utilize port flow speed value portspeed (i, t) and port flow velocity predictive value F (i, t+1) determine each port mixture velocity value Q (i, t):

Q (i, t)=β * portspeed (i, t)+(1-β) * F (i, t+1), 1 < i < n

Wherein (i, t) for each port flow speed value, F (i, t+1) is each port flow velocity predictive value at next testing time point to portspeed.The selection of β is determined as the case may be, and β is more big, and (i, value t) is more subject to the impact of current flow to Q；β is more little, and (i, value t) is more subject to the impact of flow velocity predictive value to Q；

Embodiment 1:

In the present embodiment, the server load balancing method of software-oriented definition network, adopts Topology connection relevant device as shown in Figure 2, and four servers constitute a Web server cluster.Controller is responsible for disposing dynamic strategy and is carried out the access burden of balanced each Web server.Server cluster externally provides a virtual ip address, when new host subscriber accesses the virtual IP address of Web server cluster or server cluster needs to carry out load dispatch, according to the load-balancing method in the present invention, SDN controller distributes a suitable server to the main frame of user in the way of adding stream table or amendment stream table, and its concrete grammar comprises the following steps:

Step I, each port of SDN switch is carried out flow rate detection by port detecting module:

First, port detecting module sends port statistics request message every 5s to the SDN switch in network, and port statistics response message is sent to SDN controller by safe lane by switch.Port detecting module extracts the flow value of each port from port statistics response message, calculates flow speed value.Assume flow speed value respectively 16Mb/s, 25Mb/s, 15Mb/s, 70Mb/s of four Web server Server1, Server2, Server3, Server4 connectivity ports；

Then, port detecting module calculates the predicted velocity after next 5s by Single Exponential Smoothing.Assume that α takes 0.7, flow velocity predictive value respectively 11Mb/s, 27Mb/s, 20Mb/s, 30Mb/s to current time before a upper 5s.Being computed, after next 5s, predicted velocity value is 14.5Mb/s, 25.6Mb/s, 16.5Mb/s, 58Mb/s.

Step II, each port flow speed obtained from port detecting module is analyzed by port flow velocity analysis module, it may be judged whether need to carry out load dispatch:

First, port flow velocity analysis module obtains the flow rate of each port from port detecting module, calculates the connectivity port of Server4 and has Peak Flow Rate and maximum port flow speed value is 70Mb/s, and flow velocity imbalance values is 1.7.Assume that port bandwidth threshold value be 100Mbps, P threshold value is 1.5.Although now maximum port flow speed value is in the scope of port bandwidth threshold value, but flow velocity imbalance values is much larger than set P threshold value.The difference of the implication according to z-score, maximum port flow speed value and flow velocity meansigma methods has 1.7 standard deviations, and the balanced intensity of overall flow velocity is not within the scope set, and now server cluster needs load balance scheduling.

Then, port flow velocity analysis module obtains real-time flow rate value and the flow velocity predictive value of each port from port detecting module.Assume that β takes 0.6, calculated mixture velocity value respectively 15.4Mb/s, 25.2Mb/s, 15.6Mb/s, 65.2Mb/s of each port by Single Exponential Smoothing.

Step III, load balancing module is responsible for the access request of new user the load dispatch to the unbalance server cluster of flow velocity:

First, load balancing module obtains the mixture velocity value of each port from port flow velocity analysis module, and the switch ports themselves of Server1 server has minimum mixture velocity value.According to step II it can be seen that now server cluster mass flow discrepancy weighing apparatus, this partial session, to Server1 server, and is synchronized in Server1 server, it is achieved that the shunting to flow by the session stream table of load balancing module amendment part Server4 server.

Then, when the virtual IP address having new subscriber's main station to access server cluster, load balancing module sets up session by adding stream table between new subscriber's main station and the Server1 server with minimum mixture velocity value.

By above-described embodiment it can be seen that the present invention using flow velocity imbalance values, port Peak Flow Rate whether more than corresponding P threshold value, port bandwidth threshold value as the condition judging load balancing.Use the real-time bandwidth value of port, port flow velocity predictive value as parameter, calculate the port numbers of minimum mixture velocity value, service traffics distributed to the server of this port.

In sum, the present invention, based on novel SDN framework, carries out modularization programming in SDN controller, it is achieved that port detecting module, port flow velocity analysis module, load balancing module.The switch ports themselves of Connection Service device is carried out flow rate detection by port detecting module；Port flow velocity analysis module analyzes server cluster the need of carrying out load balancing by calculated load parameter；Load balancing module calculates the switch ports themselves distributing to new and old subscriber's main station and to add or to realize the load balancing to server cluster in the way of amendment stream table by minimum mixture velocity value.

Claims (7)

1. the server load balancing method of a software-oriented definition network, it is characterised in that: comprise the following steps successively:

(1) port detecting module periodic statistical goes out the SDN switch port real-time traffic that each server connects, and calculates flow rate and the port prediction flow velocity of point of next testing time；

(2) each port flow speed that the analysis of port flow velocity analysis module obtains from port detecting module, it may be judged whether need to carry out load dispatch；

(3) load balancing module is responsible for the access request of new user and the load dispatch to the unbalance server cluster of flow velocity, and SDN controller adds stream table and sets up new session with the server specified or revise the server that original session is deployed to specify by stream table.

2. the server load balancing method of software-oriented according to claim 1 definition network, it is characterised in that: described step (1) method particularly includes:

(1.1) SDN controller regularly sends port statistics request message to all SDN switch in network by port detecting module and obtains port information, port statistics response message is sent to SDN controller by safe lane by SDN switch, controller is by the real-time traffic of the response message collection of each switch ports themselves of Connection Service device to each port the flow velocity portspeed (i calculating each port, t), wherein i represents the i-th port of SDN switch；

(1.2) according to the flow velocity portspeed of each server connectivity port, (i t), uses the Single Exponential Smoothing in statistics to calculate the port prediction flow rate F (i, t+1) of point of next testing time.

3. the server load balancing method of software-oriented according to claim 1 definition network, it is characterised in that: described step (2) method particularly includes:

(2.1) flow rate that the analysis of port flow velocity analysis module obtains from port detecting module, calculates the meansigma methods of each port flow velocityVariance VarianceSpeed (t), standard deviation sigma_portspeed(t)；By the flow speed value portspeed of relatively each port, (i t) calculates Peak Flow Rate MaxPortspeed (t) that each port takies；Meansigma methods by Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, calculate flow velocity imbalance values P (t) according to the z-score method in statistics；According to current flow value and flow velocity predictor calculation go out mixture velocity value Q (i, t)；

(2.2) P (t) threshold value PThreshold and bandwidth threshold BwThreshold is set according to the actual demand of network, when P (t) value reaches P (t) threshold value PThreshold or Peak Flow Rate MaxPortspeed (t) reaches bandwidth threshold BwThreshold, server cluster needs to carry out load dispatch.

4. the server load balancing method of software-oriented according to claim 1 definition network, it is characterised in that: described step (3) method particularly includes:

(3.1) when there being new main frame to access server cluster, load balancing module selects mixture velocity value Q, and (i, t) minimum port is as designated port；Load balancing module judges according to the object information of port flow velocity analysis module every 15s, if server cluster needs to carry out load balancing, (i, port t) is set to designated port will to have minimum mixture velocity value Q；

(3.2) SDN controller issues stream table to the SDN switch port specified, and sets up session or by original session synchronization to given server, it is achieved the load balancing to server cluster between main frame and given server.

5. the server load balancing method of software-oriented according to claim 2 definition network, it is characterised in that: each port flow velocity portspeed in described step (1.1) (i, defining method t) is:

(1.1.1) determine flow velocity portspeed (i, t):

Portspeed (i, t)=(portstraffic (i, t)-portstraffic (i, t-1))/intervaltime,

Wherein (i, t) represents the i port flow value of t to portstraffic, and portstraffic (i, t-1) represents the i port flow value in t-1 moment, and intervaltime represents the interval of two testing time points.

6. the server load balancing method of software-oriented according to claim 2 definition network, it is characterised in that: in described step (1.2), the defining method of the port prediction flow rate F (i, t+1) of next testing time point is:

(1.2.1) Single Exponential Smoothing in statistics is used to determine some port flow velocity predictive value F of the next testing time (i, t+1):

$\left\{\begin{array}{c}F(i,t+1)=\mathrm{\&alpha;}*portspeed(i,t)+(1-\mathrm{\&alpha;})*F(i,t)\\ F(i,2)=portspeed(i,2)\end{array}\right.,$

Wherein portspeed (i, t) for the flow speed value of t, F, (i, t) for the flow velocity predictive value in t-1 moment, wherein the flow velocity predictive value of the 1st time period and the flow speed value of the 1st time period are equal, α is smoothing constant, and span is [0,1].

7. the server load balancing method of software-oriented according to claim 3 definition network, it is characterised in that: in described step (2.1), the defining method of P (t) value is:

(2.1.1) by the flow speed value portspeed of each port, (i t) determines the meansigma methods of each port flow velocityPeak Flow Rate MaxPortspeed (t):

$\stackrel{\&OverBar;}{Portspeed\left(t\right)}=\left(\underset{i=1}{\overset{n}{\&Sigma;}}portspeed\right(i,t\left)\right)/n,$

$MaxPortspeed\left(t\right)=\underset{i=1}{\overset{n}{Max}}\left(portspeed\right(i,t\left)\right),$

WhereinFor all of the port flow velocity and, it is assumed that have n port；

(2.1.2) by the flow speed value portspeed of each port, (i, t) with the meansigma methods of port flow velocityDetermine the variance VarianceSpeed (t) of each port flow velocity, standard deviation sigma_portspeed(t):

$VarianceSpeed\left(t\right)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(portspeed\right(i,t)-\stackrel{\&OverBar;}{Portspeed\left(t\right)})/n,$

${\mathrm{\&sigma;}}_{portspeed\left(t\right)}=\sqrt{VarianceSpeed\left(t\right)};$

(2.1.3) by the meansigma methods of port Peak Flow Rate MaxPortspeed (t) He each port flow velocityStandard deviation sigma_portspeed(t)As parameter, determine P (t) value according to the z-score method in statistics:

$P\left(t\right)=\left(MaxPortspeed\right(t)-\stackrel{\&OverBar;}{Portspeed\left(t\right)})/{\mathrm{\&sigma;}}_{portspeed\left(t\right)},$

Z-score is for calculating given how many standard deviations of mark range averaging number in statistics, represents mark relative position amount number of present position in group in units of standard deviation；

(2.1.4) utilize port flow speed value portspeed (i, t) and port flow velocity predictive value F (i, t+1) determine each port mixture velocity value Q (i, t):

Q (i, t)=β * portspeed (i, t)+(1-β) * F (i, t+1), 1 < i < n

Wherein portspeed (i, t) for each port flow speed value, F (i, t+1) the flow velocity predictive value put in the next testing time for each port, the selection of β is determined as the case may be, and β is more big, (i, value t) is more subject to the impact of current flow to Q；β is more little, and (i, value t) is more subject to the impact of flow velocity predictive value to Q.