CN103067297A - Dynamic load balancing method and device based on resource consumption prediction - Google Patents

Abstract

The invention relates to a dynamic load balancing method and device based on resource consumption prediction. The dynamic load balancing method comprises the following steps of: calculating average resource consumption, wherein the average resource consumption is resource consumption of each service node aiming at single service of each service type; predicting the resource consumption of each service node after receiving a request according to the service type and the average resource consumption of the request to obtain predicted resource consumption; calculating the priority of each service node to the request according to the predicted resource consumption; and issuing the request to the service node with highest priority. The dynamic load balancing method and device can be used for predicting the real-time resource consumption of the service nodes through acquiring the resource consumption and online service quantity of the service nodes, improving the non-real-time property of resource consumption data and optimizing request issuing by taking full account of the difference of different types of requests to each resource consumption in the service nodes.

Description

A kind of dynamic load balancing method and device based on the resource consumption prediction

Technical field

The present invention relates to the communications field, specially refer to a kind of dynamic load balancing method and device based on the resource consumption prediction.

Background technology

According to the real load situation of whether considering the back-end services node, load-balancing method can be divided into static load balancing method and dynamic load balancing method.Comparatively speaking, dynamic load balancing method is more more reasonable on request distribution aspect than static load balancing method.

On the one hand, existing dynamic load balancing method need to gather every resource consumption of back-end services node, gathers a CPU usage that comprises the back-end services node, memory usage etc.Then, draw the load metric of back-end services node according to the index that gathers, thereby decide the request distribution with this load metric.Yet in the execution of dynamic load leveling, every certain interval of time will gather the resource consumption of a back-end services node, and the resource consumption data that obtain like this are not real-time.Existing dynamic load balancing method can not well address this problem, thereby causes the unbalanced of request distribution.

On the other hand, have a large amount of dissimilar services in present service system, wherein, the multiclass service can be provided by a service node.The resource consumption situation of in fact, dissimilar services is different.For example, the request of static page is mainly consumed magnetic disc i/o and bandwidth, the request of dynamic page is mainly consumed CPU and internal memory.Existing load-balancing method does not fully take into account this point, thereby can't realize optimum request distribution.

This shows, existing dynamic load balancing method exist the resource consumption data non real-time, do not take into full account dissimilar services to the otherness of resource consumption, thereby cause request distribution unreasonable.

Summary of the invention

The purpose of this invention is to provide a kind of dynamic load balancing method and device based on resources that can overcome defects.

In a first aspect of the present invention, a kind of dynamic load balancing method based on the resource consumption prediction is provided, comprising: calculate average resource consumption, described average resource consumption is that each service node is for the resource consumption of the single service of every kind of COS; According to COS and the described average resource consumption of request, dope the resource consumption of each service node after receiving described request, obtain predicting resource consumption; According to described prediction resource consumption, calculate each service node for the priority of described request; And described request is distributed to the highest service node of priority.

In a second aspect of the present invention, a kind of dynamic load leveling device based on the resource consumption prediction is provided, comprise: average resource consumption calculations module, be used for calculating average resource consumption, described average resource consumption is that each service node is for the resource consumption of the single service of every kind of COS; Prediction resource consumption computing module is used for COS and described average resource consumption according to request, calculates the resource consumption of each service node after receiving described request, obtains predicting resource consumption; The priority computing module is used for according to described prediction resource consumption, calculates each service node for the priority of described request; And the request distribution module, be used for described request is distributed to the highest service node of priority.

The present invention has improved the non real-time property of resource consumption data by the resource consumption that gathers service node and the real-time resource consumption that online service quantity is predicted service node; The present invention also fully takes into account dissimilar requests for the otherness of every resource consumption in the service node, has optimized the request distribution.

Description of drawings

Fig. 1 is the schematic flow sheet based on the dynamic load leveling of resource consumption prediction according to the embodiment of the invention.

Fig. 2 is the schematic diagram based on the dynamic load leveling device of resource consumption prediction according to the embodiment of the invention.

Embodiment

Below by drawings and Examples, technical scheme of the present invention is described in further detail.

Fig. 1 is the schematic flow sheet based on the dynamic load leveling of resource consumption prediction according to the embodiment of the invention.

In step 101, provide single service required resource consumption to each service node for every kind of service

Carry out initialization and calculate, wherein, i is the service node sign, and j is the COS sign.By realizing each service node of test for the maximum load capability of every kind of COS, can obtain the maximum service quantity of resource consumption that each service node connects for the maximum service of every kind of COS and corresponding every kind of COS thereof.

For example, service node Serve _RiProvide the resource consumption of the maximum service connection of j type to be expressed in matrix as:

$L_i^{j\max }=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}\right],$ Wherein $\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}$ Be respectively CPU usage, memory usage, magnetic disc i/o utilization rate, the network bandwidth utilization rate of this service node, the normalized value of route time.

Described The calculation of initial value method be:

Wherein

The maximum service linking number of the j kind service that provides for i service node.

In step 102, regularly gather the resource consumption of each service node.

The resource consumption of service node includes but not limited to the normalized value of CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate, route time (other internodal network condition on reflection and the network).Service node Server _iResource consumption can be expressed in matrix as:

$L_i=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}\right],$ Wherein $\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}$ Be respectively CPU usage, memory usage, magnetic disc i/o utilization rate, the network bandwidth utilization rate of service node, the normalized value of route time, as shown in table 1.

Table 1 shows the resource consumption of each service node.

Table 1

In step 103, the quantity of every type of service of the online service of each service node of real time record.When the request of sending to the service node distribution customer terminal, according to the COS of this request, increase the quantity of service of this COS of this service node; When service node is finished the service of this request, reduce the quantity of service of this COS of this service node.Like this, can real time record and every type quantity of service providing of update service node, as shown in table 2.

Table 2 shows the quantity of all kinds of services that each service node provides.

In step 104, the variation of every type the quantity of service that provides according to variation and each service node of the resource consumption of each service node is right

Upgrade.

By gathering service node Server _iThe up-to-date resource consumption that constantly counts at t is:

$L_i\left[t\right]=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}\left[t\right]& L_i^{\mathrm{mem}}\left[t\right]& L_i^{\mathrm{io}}\left[t\right]& L_i^{\mathrm{net}}\left[t\right]& L_i^{\mathrm{route}}\left[t\right]\end{array}\right]$

$\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}\left[t\right]& L_i^{\mathrm{mem}}\left[t\right]& L_i^{\mathrm{io}}\left[t\right]& L_i^{\mathrm{net}}\left[t\right]& L_i^{\mathrm{route}}\left[t\right]\end{array}$ Be respectively service node in the consumption of t every resource of the moment, i.e. CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate and route time.

Service node Server _iQuantity of service/number of request of every type in the constantly service that provides of t is:

N is the quantity of COS

Suppose to gather every the Δ T time resource consumption of a service node, so calculation services node Server _iOn the single Service Source consumption of every type of service is provided Method may further comprise the steps:

At step 104-1, according to the up-to-date resource consumption L that collects _i[t] and the last resource consumption L that collects _i[t-Δ T] obtains service node Server within this collection period _iThe changes delta L of resource consumption _i[t]:

ΔL _i[t]=L _i[t]-L _i[t-ΔT]

At step 104-2, the quantity of service during according to up-to-date the collection

Quantity of service when gathering with the last time Obtain within this collection period service node Server _iTo Service _jThe variation of the quantity of service/number of request of service

${\mathrm{\ΔCon}}_i^j={\mathrm{Con}}_i^j\left[t\right]-{\mathrm{Con}}_i^j[t-\mathrm{\ΔT}];$

At step 104-3, according to the as a result calculation services node Server of step 104-1 and step 104-2 _iTo COS Service _jThe average resource consumption of each request

Can calculate by the following equation group:

$\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{AvgL}}_i^j*{\mathrm{\ΔCon}}_i^j={\mathrm{\ΔL}}_i^j,$ N is the type service number;

At step 104-4, if above-mentioned equation group is not enough to solve Then can utilize in the last collection period

With Construct according to the method described above new equation, until can solve each Till.In other words, by obtaining the difference of more resource consumption and quantity of service, make up more equation and find the solution.Generally speaking, if the different types of service of N kind is arranged, need at least N equation.

In step 105, receive the request that sends by network from client.

In step 106, suppose that each service node asks to provide service for this reason, utilize each service node that the single Service Source consumption of this type service is provided, predict the resource consumption of each service node, that is: service node Server _iIf accept Service _jResource consumption after the request is expressed as:

In step 107, the calculation services node provides the priority of request.

Each service node has the resource consumption alarm threshold that sets in advance

$\mathrm{TotalL}=\begin{array}{ccccc}{L}^{\mathrm{cpu}}& L^{\mathrm{mem}}& L^{\mathrm{io}}& L^{\mathrm{net}}& L^{\mathrm{route}},\end{array}$ The every resource consumption that is each service node all is provided with alarm threshold, i.e. CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate, route time.By with NextL _iWith TotalL ^jIn every resource threshold compare respectively, the more a plurality of differences that relatively obtain are carried out normalization, then therefrom find out minimum normalization difference and with it as the priority of this service node for this request.

For example, to Service _jThe request of COS, service node Server _iPriority for this request can be expressed as: $P_i=\min (1-\frac{L_i+{\mathrm{AvgL}}_i^j}{\mathrm{TotalL}}).$

In step 108, select the service node of priority maximum to ask to provide service for this reason.

Relatively all service nodes are therefrom selected the service node of priority maximum for the priority of this request, and this request is transmitted to this service node.

Fig. 2 is the schematic diagram based on the dynamic load leveling device of resource consumption prediction according to the embodiment of the invention.

Start initially at device, the average resource initialization module provides single service required resource consumption to each service node for every kind of service

Carry out initialization and calculate, wherein, i is the service node sign, and j is the COS sign.By realizing each service node of test for the maximum load capability of every kind of COS, can obtain the maximum service quantity of resource consumption that each service node connects for the maximum service of every kind of COS and corresponding every kind of COS thereof.

For example, service node Server _iProvide the resource consumption of the maximum service connection of j type to be expressed in matrix as:

$L_i^{j\max }=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}\right],$ Wherein $\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}$ Be respectively CPU usage, memory usage, magnetic disc i/o utilization rate, the network bandwidth utilization rate of this service node, the normalized value of route time.

Described

The calculation of initial value method be:

Wherein The maximum service linking number of the j kind service that provides for i service node.

After device started, acquisition module regularly gathered the resource consumption of each service node.

The resource consumption of service node includes but not limited to the normalized value of CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate, route time (other internodal network condition on reflection and the network).Service node Server _iResource consumption can be expressed in matrix as:

$L_i=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}\right],$ Wherein $\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}& L_i^{\mathrm{mem}}& L_i^{\mathrm{io}}& L_i^{\mathrm{net}}& L_i^{\mathrm{route}}\end{array}$ Be respectively CPU usage, memory usage, magnetic disc i/o utilization rate, the network bandwidth utilization rate of service node, the normalized value of route time, as shown in table 1.

Table 1 shows the resource consumption of each service node.

Table 1

The quantity of every type of service of the online service of each service node of logging modle real time record.When the request of sending to the service node distribution customer terminal, according to the COS of this request, increase the quantity of service of this COS of this service node; When service node is finished the service of this request, reduce the quantity of service of this COS of this service node.Like this, can real time record and every type quantity of service providing of update service node, as shown in table 2.

Table 2 shows the quantity of all kinds of services that each service node provides.

Based on gathering and the information of record, average resource consumption calculations module is right according to the variation of every type the quantity of service that variation and each service node of the resource consumption of each service node provides

Upgrade.

By gathering service node Server _iThe up-to-date resource consumption that constantly counts at t is:

$L_i\left[t\right]=\left[\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}\left[t\right]& L_i^{\mathrm{mem}}\left[t\right]& L_i^{\mathrm{io}}\left[t\right]& L_i^{\mathrm{net}}\left[t\right]& L_i^{\mathrm{route}}\left[t\right]\end{array}\right]$

$\begin{array}{ccccc}{L}_i^{\mathrm{cpu}}\left[t\right]& L_i^{\mathrm{mem}}\left[t\right]& L_i^{\mathrm{io}}\left[t\right]& L_i^{\mathrm{net}}\left[t\right]& L_i^{\mathrm{route}}\left[t\right]\end{array}$ Be respectively service node in the consumption of t every resource of the moment, i.e. CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate and route time.

Service node Server _iQuantity of service/number of request of every type in the constantly service that provides of t is:

N is the quantity of COS

Suppose to gather every the Δ T time resource consumption of a service node, so calculation services node Server _iOn the single Service Source consumption of every type of service is provided

Method may further comprise the steps:

At first, according to the up-to-date resource consumption L that collects _i[t] and the last resource consumption L that collects _i[t-Δ T] obtains service node Server within this collection period _iThe changes delta L of resource consumption _i[t]:

ΔL _i[t]=L _i[t]-L _i[t-ΔT]

Quantity of service during next, according to up-to-date the collection

Quantity of service when gathering with the last time

Obtain within this collection period service node Server _iTo Service _jThe variation of the quantity of service/number of request of service

${\mathrm{\ΔCon}}_i^j={\mathrm{Con}}_i^j\left[t\right]-{\mathrm{Con}}_i^j[t-\mathrm{\ΔT}];$

Then, according to above-mentioned result of calculation calculation services node Server _iTo COS Service _jThe average resource consumption of each request Can calculate by the following equation group:

$\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{AvgL}}_i^j*{\mathrm{\ΔCon}}_i^j={\mathrm{\ΔL}}_i^j,$ N is the type service number.

In addition, if above-mentioned equation group is not enough to solve

Then can utilize in the last collection period

With

Construct according to the method described above new equation, until can solve each

Till.In other words, by obtaining the difference of more resource consumption and quantity of service, make up more equation and find the solution.Generally speaking, if the different types of service of N kind is arranged, need at least N equation.

The dynamic load leveling device based on resources according to the embodiment of the invention can be contained in the server, and this server is connected with service node with client by network.Therefore, can be by the service request of network reception from client according to the dynamic load leveling device based on resources of the embodiment of the invention.

Suppose that each service node asks to provide service for this reason, prediction resource consumption module utilizes each service node that the single Service Source consumption of this type service is provided, and predicts the resource consumption of each service node, that is: service node Server _iIf accept Service _jResource consumption after the request is expressed as:

Then, priority computing module calculation services node provides the priority of request.Priority block comprises comparison module, normalization module, priority selection module.

Each service node has the resource consumption alarm threshold TotalL that sets in advance ^j, $\mathrm{TotalL}=\begin{array}{ccccc}{L}^{\mathrm{cpu}}& L^{\mathrm{mem}}& L^{\mathrm{io}}& L^{\mathrm{net}}& L^{\mathrm{route}},\end{array}$ The every resource consumption that is each service node all is provided with alarm threshold, i.e. CPU usage, memory usage, magnetic disc i/o utilization rate, network bandwidth utilization rate, route time.Comparison module passes through NextL _iWith TotalL ^jIn every resource threshold compare respectively, the normalization module is carried out normalization with a plurality of differences that relatively obtain again, then priority select module therefrom find out minimum normalization difference and with it as the priority of this service node for this request.

For example, to Service _jThe request of COS, service node Server _iPriority for this request can be expressed as: $P_i=\min (1-\frac{L_i+{\mathrm{AvgL}}_i^j}{\mathrm{TotalL}}).$

At last, the request distribution module selects the service node of priority maximum to ask to provide service for this reason.

Request distribution module relatively all service nodes is therefrom selected the service node of priority maximum for the priority of this request, and this request is distributed to this service node.

The professional should further recognize, unit and the algorithm steps of each example of describing in conjunction with embodiment disclosed herein, can realize with electronic hardware, computer software or the combination of the two, for the interchangeability of hardware and software clearly is described, composition and the step of each example described in general manner according to function in the above description.These functions are carried out with hardware or software mode actually, depend on application-specific and the design constraint of technical scheme.The professional and technical personnel can specifically should be used for realizing described function with distinct methods to each, but this realization should not thought and exceeds scope of the present invention.

The method of describing in conjunction with embodiment disclosed herein or the step of algorithm can use the software module of hardware, processor execution, and perhaps the combination of the two is implemented.Software module can place the storage medium of any other form known in random asccess memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or the technical field.

Above-described embodiment; purpose of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the above only is the specific embodiment of the present invention; the protection range that is not intended to limit the present invention; within the spirit and principles in the present invention all, any modification of making, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1. dynamic load balancing method based on resource consumption prediction comprises:

The calculating average resource consumes, and described average resource consumption is that each service node is for the resource consumption of the single service of every kind of COS;

According to COS and the described average resource consumption of request, dope the resource consumption of each service node after receiving described request, obtain predicting resource consumption;

According to described prediction resource consumption, calculate each service node for the priority of described request; And

Described request is distributed to the highest service node of priority.

2. method according to claim 1, wherein, the step of described calculating average resource consumption comprises:

According to each service node the resource consumption in one or more moment and each service node in described one or more moment the quantity of service for every kind of COS, calculate described average resource consumption;

Wherein, described each service node obtains by each service node is carried out timing acquiring at the resource consumption in one or more moment, and described each service node is to obtain by the record that request distribution and request are finished in described one or more moment for the quantity of service of every kind of COS.

3. method according to claim 1 is wherein described according to described prediction resource consumption, calculates each service node and comprises for the step of the priority of described request:

Every resource consumption threshold value of more described prediction resource consumption and predefined each service node obtains a plurality of differences;

Described a plurality of differences are carried out normalization, obtain a plurality of normalization differences; And

Normalization difference minimum in a plurality of normalization differences is chosen as described priority.

4. method according to claim 1 also comprises:

According to the maximum resource consumption that obtains in advance and maximum service quantity initialization is carried out in described average resource consumption, wherein, described maximum resource consumption be each service node for the resource consumption of the maximum service quantity of every kind of COS, described maximum service quantity is that each service node is for the maximum service quantity of every kind of COS.

5. described method one of according to claim 1-4, wherein, described resource consumption comprises one or more in CPU usage, memory usage, magnetic disc i/o utilization rate, the network bandwidth and the route time of service node.

6. dynamic load leveling device based on resource consumption prediction comprises:

Average resource consumption calculations module is used for calculating average resource consumption, and described average resource consumption is that each service node is for the resource consumption of the single service of every kind of COS;

Prediction resource consumption computing module is used for COS and described average resource consumption according to request, calculates the resource consumption of each service node after receiving described request, obtains predicting resource consumption;

The priority computing module is used for according to described prediction resource consumption, calculates each service node for the priority of described request; And

The request distribution module is used for described request is distributed to the highest service node of priority.

7. device according to claim 6 also comprises:

Acquisition module is used for the resource consumption of each service node in one or more moment carried out timing acquiring; And

Logging modle is used for the record finished by to request distribution and request, obtain each service node in described one or more moment the quantity of service for every kind of COS;

Wherein, described average resource consumption calculations module calculates described average resource consumption in described one or more moment for the quantity of service of every kind of COS at the resource consumption in one or more moment and described each service node according to described each service node.

8. device according to claim 6 also comprises:

Average resource consumes initialization module, is used for according to the maximum resource consumption and the maximum service quantity that obtain in advance initialization being carried out in described average resource consumption;

Wherein, described maximum resource consumption be each service node for the resource consumption of the maximum service quantity of every kind of COS, described maximum service quantity is that each service node is for the maximum service quantity of every kind of COS.

9. device according to claim 6, wherein said priority computing module comprises:

Comparison module, the every resource consumption threshold value for more described prediction resource consumption and predefined each service node obtains a plurality of differences;

The normalization module is used for described a plurality of differences are carried out normalization, obtains a plurality of normalization differences; And

Priority is selected module, is used for the normalization difference that a plurality of normalization differences are minimum and is chosen as described priority.

10. server, comprise the described dynamic load leveling device based on the resource consumption prediction such as one of claim 6-9, described server is connected with described service node with client by network, is used for the request that client sends is distributed to described service node.

Images