CN107566535B - Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service - Google Patents
Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service Download PDFInfo
- Publication number
- CN107566535B CN107566535B CN201711024188.3A CN201711024188A CN107566535B CN 107566535 B CN107566535 B CN 107566535B CN 201711024188 A CN201711024188 A CN 201711024188A CN 107566535 B CN107566535 B CN 107566535B
- Authority
- CN
- China
- Prior art keywords
- load
- wmsp
- service
- rate
- period
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Computer And Data Communications (AREA)
Abstract
The invention discloses a self-adaptive load balancing strategy based on a user concurrent access time sequence rule in a Web map service, which comprises the following steps: s1, WMSP access arrival rate time sequence analysis and parameter acquisition; s2, variable load feedback period strategy; s3, calculating load expectation of the WMSP server cluster nodes in a feedback period; and S4, based on the load expectation of the WMSP server cluster nodes, carrying out task scheduling and load distribution reasonably. The invention improves the service throughput of WMSP; the problem that the real-time service processing capacity of the server cannot be rapidly and effectively acquired is solved, the CPU utilization rate of the WMSP is obviously improved, and meanwhile, the load imbalance degree is small; the method highly accords with the time sequence distribution characteristics of concurrent access of users in the WMSP, and the proposed variable feedback period strategy effectively coordinates and balances the contradiction relationship between WMSP service resource consumption and load state acquisition, ensures the optimization of service efficiency and load balance, and effectively improves the service performance and service resource utilization efficiency of the WMSP.
Description
Technical Field
The invention relates to the technical field of network space information service, in particular to a self-adaptive load balancing strategy based on concurrent access timing sequence rules of Web map service.
Background
With the rapid development of the Internet, the continuous popularization of mobile networks and the rapid development of geospatial information systems, users of Web Map Service Platforms (WMSP) such as Google Earth and Baidu maps are increasing day by day, and the real-time online operation reaches millions. The server service overload and access delay performance problem is brought by high-strength, strong aggregation and high concurrency access while brand-new experience is brought to the user. The cloud cluster service system at the back end of the Web map service has distributivity, heterogeneity and dynamics, and the load balancing capability of the cloud cluster service system is a key for ensuring that good service capability can still be maintained during a peak period of user access, such as high throughput, low response time, high service concurrency number, server effectiveness and the like. The periodicity and the burstiness of the concurrent space-time access behaviors of the users directly influence the space-time demand and the load distribution of the Web map service on cloud resources. Meanwhile, differences of access time and access content of a user and differences of cloud cluster server service capabilities of the WMSP require that a load balancing strategy of the cloud end has real-time performance and self-adaptability, and computing resources of the cloud service are consumed to the minimum extent. Therefore, how to balance the resource consumption of the timing sequence unbalance of the user access on the WMSP and the load balancing strategy and improve the service quality of the WMSP is one of the problems to be solved urgently.
Most of the traditional load balancing algorithms are based on task static allocation, such as: the method comprises a rotation method, a weighted rotation method, a target address hash scheduling algorithm, a source address hash scheduling algorithm and the like, and the algorithm is suitable for small-scale isomorphic cluster systems with single configuration. On the basis, scholars propose load balancing algorithms such as a minimum connection method and a minimum load for task distribution based on the current state of a server. And if Xmitbyte, calculating a load weight according to the average output flow to carry out load distribution. The algorithm can obviously improve the throughput and the service efficiency of the system; but the load cannot be accurately reflected in real time, and the method is difficult to adapt to large-scale high-concurrency task allocation. Many scholars propose load balancing methods with dynamic feedback. Stankovic et al have proposed a load balancing method based on dynamic feedback for the first time, and have collected load information of computing nodes through a feedback mechanism to perform effective scheduling and allocation of tasks. The method is suitable for network application systems with stable load change and small fluctuation. Then, the scholars propose a load balancing method based on adaptive load change, for example, Kencl et al propose a feedback control mechanism by expanding a Highest Random Weight (HRW) algorithm, dynamically modify the set weight vector, and realize adaptive load balancing. Alam M et al propose a quality of service Qos based load balancing algorithm. Zhu H et al propose a DDSD (Demand-drive Service Differentiation) Differentiation strategy based on queuing theory. The load balancing methods mentioned above all have good real-time adaptability of load. But the change of load is changed along with the change of the access behavior of the user, and the research cannot simply adapt to the change, thereby bringing about certain extra consumption of computing resources or cache resources.
In recent years, many scholars think that the social laws and the interaction modes expressed when large-scale users roam in a geographic information service system are mined, and very important guiding functions are provided for optimizing the performance of the system and improving the roaming smooth experience of the users. Xia et al propose that the load brought to the server by the user when accessing the public map service platform has a "tide" characteristic, and Li et al propose that the access of geographic data has spatiotemporal locality. These studies demonstrate that the user's access requests are also subject to spatiotemporal aggregations due to the spatiotemporal attributes of the geographic data. The access characteristic has an important influence on the service performance of the Web map service, such as traffic fluctuation, service queue length and the like, and simultaneously determines the space-time unbalanced requirement of the Web map service on computing resources. Meanwhile, the access strength of the user in the Web map service is changed in time series. The basic idea of the dynamic load balancing strategy is to periodically collect real-time loads of all cluster nodes, and further maintain the balance of the load of the whole cluster system through a corresponding task allocation strategy. The load feedback cycle is too long to reflect the load state in real time, and too short results in additional system overhead caused by frequent collection of load information. Therefore, how to effectively capture the time sequence change of the user access aggregation and the access strength thereof, perform accurate and rapid service load prediction and task allocation, reduce the service resource consumption brought by information acquisition, and solve the problems of service resource demand and server performance brought by the centralization and the outburst of the user access is difficult and fundamental.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a self-adaptive load balancing strategy based on a Web map service concurrent access timing sequence rule aiming at the defect that load of each node of a server in a server cluster is unbalanced due to user timing sequence unbalanced concurrent access behaviors in WMSP.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the invention provides a self-adaptive load balancing strategy based on a concurrent access timing sequence rule of a Web map service, which comprises the following steps:
s1, acquiring an access log generated in the long-term operation of the WMSP server cluster, performing access to the access log based on different time granularities of a long period and a short period to achieve time sequence analysis, and calculating a variable load feedback cycle arrival rate threshold according to the maximum concurrency number which can be contained by the WMSP server cluster per second and the access log access to achieve time sequence analysis results;
s2, calculating a variable load feedback period according to the variable load feedback period arrival rate threshold;
s3, calculating the service rate of each node load of the WMSP server cluster, and establishing an association table of the cluster node load and the service rate; calculating load expectation of the WMSP server cluster nodes in a feedback period;
s4, generating a forwarding probability space of the cluster nodes based on load expectation, and obtaining a larger probability space by the cluster nodes with larger load expectation according to the association table of the cluster node load and the service rate; and when the access request arrives at the WMSP server cluster, a random number is generated in the [0,1] space, a target cluster node is determined according to the drop point of the random number in the probability space, and the access request is distributed to the target cluster node.
Further, in step S1 of the present invention, the method for calculating the variable load feedback cycle arrival rate threshold includes:
counting WMSP user access logs based on different time granularities of a long period and a short period, wherein the long period counting of the access logs is used for extracting period characteristics of user access, and the short period counting of the logs is used for extracting variable strength characteristics of the user access;
defining the access arrival rate sequence of the user as a time sequence:
H(R,t)＝{R(t1),R(t2)…R(tn)}
wherein, R and t are the arrival rate factors of the user arrival time sequence H (R, t) respectivelyAnd a time factor; r (t)i) Represents tiThe number of service requests arriving in a time period, i.e. the user access strength;
based on the maximum concurrency number M which can be accommodated by WMSP per second and the WMSP long-period access log time sequence analysis, the user arrival rate threshold value which meets the load feedback period change is taken as M0β M, where β is an empirical value.
Further, the method for calculating the variable load feedback period in step S2 of the present invention is:
s21, if the current arrival rate lambda (t)i)＜M0Then with T0As a duty feedback period, where λ (t)i) Is tiArrival rate of WMSP in time slot, T0A fixed value of the duty feedback period;
s22, if the current arrival rate lambda (t)i)≥M0Calculating the current period TkAverage rate of change of inter-arrival rate
Wherein T iskRepresenting the kth load feedback cycle; suppose period TkStarting at time tiThe period is n time intervals, tiThe arrival rate change rate at the time is
Wherein, λ (t)i-1) Is tiLast moment ti-1User access arrival rate; thus obtaining
Further calculating the load feedback period
Wherein, the value L is a correlation coefficient of the feedback period and the load change rate; then, a load feedback period prediction method based on one-time exponential smoothing is used for predicting the subsequent load of the WMSP and calculating a feedback period based on the load
Wherein Fk-1For the load feedback period Tk-1The average rate of change of the arrival rate of (a) is an exponential smoothing prediction value, and α is a weight coefficient and is an empirical value.
Further, in step S22 of the present invention, the load feedback prediction method based on the first exponential smoothing specifically includes:
predicting the average change rate of the user access arrival rate on the basis of the prior data as follows:
by calculating an exponential smoothing value, giving a larger influence weight to the user access arrival rate change actual value close to the time on the user access arrival rate change predicted value in the next period, and predicting the average change rate of the user access arrival rate in the next period; wherein, FkIs a period TkIs an exponentially smooth predicted value of the average rate of change of the arrival rate, Fk-1Is a period Tk-1The index smoothing trend prediction value of (1);
period Tk-1α is a weight coefficient, is an empirical value, and can obtain a load feedback period matching the current load state
With the proviso that λ (t)i)≥M0And the L value is a correlation coefficient of the feedback period and the load change rate.
Further, in step S3, the method for calculating the load expectation of the WMSP server cluster node in a feedback cycle in the present invention includes:
s31, calculating the WMSP server cluster node service capacity, namely the service rate:
in order to quantify the service rate of cluster nodes of the WMSP server, an association table of the load and the service rate of each cluster node is established by a method based on equal interval division; for cluster node SiIs of workload capability Loadi(Loadq-1,Loadq) With a corresponding service rate of muiq；
S32, calculating load expectation of the WMSP server cluster nodes;
clustering nodes S for WMSP serveriBased on the Little law, queuing theory and the obtained service rate muiqAnd calculating to obtain the current cluster node SiLoad expected value E (λ)i) Wherein λ isiIs the user arrival rate.
Further, the method for calculating the service rate in step S31 of the present invention specifically includes:
for a group of cluster nodes S ═ { S ] in a WMSP server clusteriI is more than or equal to 1 and less than or equal to S, and the WMSP server cluster node S is madeiHas a workload capacity of Si(Loadmin,Loadmax) Dividing the workload range of the node into Q intervals at equal intervals:
Si([Loadmin,Load1],(Load1,Load2]…(Loadq-1,Loadq]…(LoadQ-2,LoadQ-1](LoadQ-1,Loadmax])
wherein (Load)q-1,Loadq]A load interval corresponding to the q-th interval; with corresponding service rate of muiqAnd the value is the average value of the service rates corresponding to the corresponding load intervals.
Further, the method for calculating the load expectation of the cluster node of the WMSP server in step S32 of the present invention specifically includes:
clustering nodes S for WMSP serveriWith a user arrival rate of λiCluster node service rate is muiMean latency between requests of cluster nodes is Di(ii) a The delay time of the request comprises the queuing time of the task and the task processing time;
then according to Little's law, get
The user arrival rate and the service rate on the cluster node determine the average queuing time and the average response time in the cluster node;
suppose WMSP expectsUser average response time of TeAnd calculating to obtain cluster node SiThe average arrival rate expected when this condition is satisfied is
Wherein muiRepresenting a cluster node SiA service capability;
when the load of the cluster node is larger, the residual processing capacity is smaller, so that the muiThe value size decreases as the cluster node load increases;
the obtained service rate muiqIntroduction of E (. lamda.)i) And calculating to obtain cluster node SiExpected average arrival rate of users under current load
I.e. cluster node SiIs desired.
Further, the method for generating the forwarding probability space of the cluster node in step S4 of the present invention is:
set of cluster nodes S ═ S in WMSP server clusteriI is more than or equal to 1 and less than or equal to S, and the corresponding forwarding probability is P ═ P based on the scheduling strategy expected by the load1,p2,...,ps}；
Wherein the cluster node SiHas a forwarding probability of
The forward probability space is
Wherein P is1+P2+…+Ps＝1。
The invention has the following beneficial effects: the invention discloses a self-adaptive load balancing strategy based on a Web map service concurrent access time sequence rule, and provides a method for realizing self-adaptive WMSP load change and reducing additional overhead brought by load acquisition by adopting a real-time variable load feedback period to match time sequence change of user concurrent access strength based on the aggregation characteristic and the time sequence characteristic when a user roams in WMSP; and then simply and effectively acquiring the current processing capacity of the server based on WMSP load expectation in a variable feedback period and the corresponding relation between different loads of each cluster node and the processing capacity, and realizing a self-adaptive load balancing strategy based on dynamic forwarding probability space distribution service requests.
1. The rules and the characteristics of user access behaviors in the WMSP are deeply mined, and the load feedback method is flexibly designed based on the rules of the periodicity of the user access intensity change and the time sequence correlation, so that the consumption of additional service resources is reduced, and the utilization of the WMSP service resources is maximized:
there is a timing imbalance and a strength imbalance in the access of users in the WMSP, which are key factors affecting the load imbalance of the WMSP. The method obtains the regular characteristics of the user access time sequence and the intensity change based on the log data statistical analysis method, scientifically sets the variable feedback period by modeling and predicting the user access intensity time sequence change, so that the WMSP can reasonably distribute service resources no matter in the peak or the valley of the access amount, the service resource utilization is maximized, and the overall performance of the WMSP is improved.
2. By predicting the change of the user access intensity, the variable load feedback period is accurately and flexibly acquired, the service resource consumption is reduced, and the efficiency and the reliability of the load feedback are improved:
the invention provides a variable load feedback period method based on the change rule of the user access arrival rate. The access concentration, load intensity and change rule are different in different time periods. The current load of the WMSP is determined by the current user access arrival rate and the system queuing state, and the load feedback period is influenced by the time sequence change of the current user access arrival rate and the access arrival rate, so that when the WMSP is in a state of low arrival rate, small arrival rate change and low load, a fixed longer load feedback period is adopted; and when the accesses are gathered and the load change fluctuation is large, calculating and acquiring the current arrival rate and the load feedback period corresponding to the change rate of the current arrival rate through the time sequence prediction of the access arrival rate. The method and the system can reflect the current load of the cluster nodes of the WMSP server in real time, reduce the consumption of load collection on the system, improve the service throughput of the system and reduce the extra service resource consumption of the WMSP.
3. By establishing and maintaining the association table of the load and the service rate, the service processing capacity of the server under different loads is quickly and effectively acquired, and the optimization of the service efficiency and the load balance is ensured.
The invention acquires the load state of the server in real time by establishing and maintaining the association table between the load and the service rate, and selects the optimal service request distribution strategy based on the load expectation of the WMSP server cluster node, so that the service resource utilization rate and the load balance degree of the WMSP are obviously improved. The invention can realize better load balance of the service request and lower service resource consumption in the WMSP cluster environment, can obtain good response performance of the service request and higher WMSP throughput rate, and is suitable for variable-strength access of large-scale users.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the adaptive load balancing policy based on the concurrent access timing rule of the Web map service according to the embodiment of the present invention includes the following steps:
s1, acquiring an access log generated in the long-term operation of the WMSP server cluster, performing access to the access log based on different time granularities of a long period and a short period to achieve time sequence analysis, and calculating a variable load feedback cycle arrival rate threshold according to the maximum concurrency number which can be contained by the WMSP server cluster per second and the access log access to achieve time sequence analysis results;
s2, calculating a variable load feedback period according to the variable load feedback period arrival rate threshold;
s3, calculating the service rate of each node load of the WMSP server cluster, and establishing an association table of the cluster node load and the service rate; calculating load expectation of the WMSP server cluster nodes in a feedback period;
s4, generating a forwarding probability space of the cluster nodes based on load expectation, and obtaining a larger probability space by the cluster nodes with larger load expectation according to the association table of the cluster node load and the service rate; and when the access request arrives at the WMSP server cluster, a random number is generated in the [0,1] space, a target cluster node is determined according to the drop point of the random number in the probability space, and the access request is distributed to the target cluster node.
The basic idea of the method of the invention is as follows: adaptive dynamic load balancing that takes into account the aggregative and chronological nature of user access in WMSP systems. Firstly, by matching a variable feedback period strategy of user access time sequence change, load feedback reflects the current load of each node server in real time and reduces the consumption of load collection on a system; and then, an association table of the load and the service rate is established and maintained, the service processing capacity of the server under different loads is quickly and effectively obtained, an optimal service request distribution strategy is selected based on the load expectation of the WMSP server cluster node, and the service resource utilization rate of the WMSP system is obviously improved.
Compared with the common dynamic load balancing method, the key creation point of the invention is that the load feedback cycle is dynamically obtained based on the user access time sequence distribution rule, and the association table of the load and the service rate is established and maintained at the same time, so that the load state of the server is obtained in real time on the premise of consuming less service resources; based on the load expectation of the WMSP server cluster node, the optimal server is selected to process the service request task, so that the optimal service resource utilization rate is achieved.
In another embodiment of the invention:
the invention provides a self-adaptive load balancing strategy based on a concurrent access timing sequence rule of a Web map service, which comprises the following specific steps:
step 1, WMSP visit arrival rate time sequence analysis and parameter acquisition
The implementation basis of the method is the result of statistical analysis of the user and system interaction mode in the long-term operation of WMSP. And counting the access arrival rate of the user under different time granularities of a long term and a short term by collecting the WMSP system access log. Under the long-term time granularity, the user presents periodicity to the WMSP access request, the mode of the working day is similar, and the mode of the rest day is also similar; at short time granularity, user access exhibits strong burstiness and aggregativity characteristics. In different time intervals, the user has aggregation and concurrency rules for WMSP access requests. According to the queuing theory, the average number of service requests arriving in unit time is the access arrival rate lambda of the WMSP, and can be used for expressing the access load of the WMSP. If the time of a cycle is divided into a plurality of equal-length time intervals, the service request number of each time interval forms a time sequence H (R, t), as shown in formula (1):
H(R,t)＝{R(t1),R(t2)…R(tn)} (1)
wherein T is the time length of each time interval; λ (t) represents the user average arrival rate for the t period.
Wherein, R and t are the user arrival time sequence H (R, t) arrival rate factor and time factor respectively; r (t)i) Represents tiThe number of service requests, i.e. the user access strength, that arrive within the time period. For the time series H (R, t), the expected value expression of the mean is:
E(H(R,t))＝λ(t)*T (2)
wherein T is the time length of each time interval; λ (t) represents the user average arrival rate for the t period.
And obtaining a time sequence of the service request access arrival rate based on the time sequence H (R, t), and further analyzing the periodicity of the user access, the access intensity change in different time periods, and high-intensity burstiness and aggregativity characteristics. Meanwhile, based on the maximum concurrency number M and the access log time sequence analysis which can be contained by WMSP per second, the variable load feedback cycle arrival rate threshold is taken as M0β · M, where β is an empirical value.
Step 2, variable load feedback period strategy
In the step 1, different feedback cycle strategies are adopted for arrival rates under different time sequences based on the feedback arrival rate threshold obtained in the step. The total load of the WMSP server cluster is positively correlated with the access strength (arrival rate), for WMSPs with lower load or more stable load, the load capacity in each time sequence period is similar, the load value in the previous period can be adopted to replace the load value in the next period, and the tasks are distributed, so that unnecessary system overhead caused by load feedback and information collection is avoided. For the WMSP with the load changing along with the sequence, the access aggregation, the load intensity and the change rule in different time periods are different, the load feedback period changes accordingly, and the utilization rate of system service resources can be maximized by dynamically matching the load feedback period.
Calculating a variable load feedback period T in a variable load feedback period strategykThe method comprises the following two substeps:
1) if the current arrival rate lambda (t)i)＜M0Then with T0As a load feedback period. Where λ (t)i) Is tiArrival rate of the system at time interval, T0At a fixed, larger value of the duty feedback cycle.
2) If λ (t)i)≥M0Calculating the current period TkAverage rate of change of inter-arrival rate
Suppose period TkStarting at time tiThe period is n time intervals, tiRate of change of arrival delta (t) at timei) Comprises the following steps:
wherein, λ (t)i-1) Is tiLast moment ti-1Access arrival rate of the user. To obtain
The following were used:
further calculating the load feedback period TkComprises the following steps:
wherein, the value L is a correlation coefficient of the feedback period and the load change rate.
The access intensity of the user changes according to a certain periodic rule; the distribution of the access arrival rate in the same access time period in different periods has certain stability, so that the predicted access arrival rate, the time sequence change of the access arrival rate and the corresponding load feedback period can be reasonably delayed. Considering that the prediction complexity is reduced, the accuracy of the change prediction of the user access arrival rate is ensured, and the consumption of service resources is reduced, the method is based on a one-time exponential smooth prediction method, and the average change rate of the user access arrival rate is predicted on the basis of a small amount of prior data, as shown in a formula (6):
and (3) by calculating an exponential smoothing value, giving a larger influence weight to the user access arrival rate change actual value close to the time on the user access arrival rate change predicted value in the next period, and predicting the average change rate of the user access arrival rate in the next period. Wherein, FkIs a period TkIs an exponentially smooth predicted value of the average rate of change of the arrival rate, Fk-1Is a period Tk-1The index smoothing trend prediction value of (1);
period Tk-1α is a weighting factor and is an empirical value.
Substituting equations (4) and (6) into equation (5) can result in the predicted load feedback period Tk。
Step 3, load expectation calculation of WMSP server cluster nodes in a feedback period
Based on the dynamic load feedback period obtained in the step 2, the WMSP server cluster node can reasonably and flexibly perform load feedback. When the WMSP is in different feedback periods, the load and remaining processing power of the cluster nodes are also different, and the load expectation for the subsequent time is also different, and the load expectation also has time variability. The service request is distributed to cluster nodes in the WMSP cluster through a load balancer, if the current cluster node is in an empty load state, the task is directly executed, and if the current cluster node is in a load state, the task is queued to wait for the previous task to be executed. Therefore, if the load expectation of the cluster node is small, the current load of the system is high, the residual processing capacity is small, and vice versa, so that the load expectation can represent the load state of the cluster node and serve as the basis for load collection. The process of calculating the load expectation includes the following two substeps:
1) calculating the service capability, namely the service rate, of the cluster nodes of the WMSP server:
generally, the server load is related to the input index of the server, the current CPU load, the current disk usage, the current memory service condition, the current number of processes of the server, and the service response time. However, the server service rate and the load thereof have no function relationship which can be quantified, and no matter the service rate of the server is manually tested or calculated, huge extra calculation consumption is caused for a large-scale cluster system. In order to quantify the service rate of the WMSP server cluster node, the invention establishes an association table of the cluster node load and the service rate based on an equal interval division method.
For a set of cluster nodes in WMSP S ═ SiI is more than or equal to 1 and less than or equal to S }, and cluster nodes SiIs of workload capability Loadi(Loadmin,Loadmax)，Dividing the working load range of the cluster node into Q intervals at equal intervals, then Si([Loadmin,Load1],(Load1,Load2]…(Loadq-1,Loadq]…(LoadQ-2,LoadQ-1](LoadQ-1,Loadmax]) Wherein (Load)q-1,Loadq]Is a load interval corresponding to the q-th interval and the corresponding service rate is muiqAnd the value is the average value of the service rates corresponding to the corresponding load intervals.
2) Computing load expectations of WMSP server cluster nodes
Clustering nodes S for WMSP serveriLet its user arrival rate be λiCluster node service rate is muiMean latency between requests of cluster nodes is Di(ii) a Wherein the delay time of the request comprises a queuing time of the task and a task processing time. Then according to Little's law, there are:
the user arrival rate and service rate at a cluster node determine the average queue length and average response time within the cluster node. Suppose that the average user response time expected by WMSP is TeCalculating to obtain cluster node SiAverage arrival rate E (λ) expected when this condition is satisfiedi) As shown in formula (9):
wherein muiRepresenting a cluster node SiService capabilities. When the load of the cluster node is larger, the residual processing capacity is smaller, and muiThe value size decreases as cluster node load increases. Will service rate as muiqSubstituting formula (9) to calculate the average user arrival rate λ expected by the node under the current loadiq：
And 4, reasonably scheduling tasks and distributing loads based on the load expectation of the WMSP server cluster nodes.
Service requests arriving in WMSPs are typically distributed to the least loaded server to minimize its queue latency in order to process the request faster and return the result. Therefore, based on the server load and the service rate table in step 3, the real-time maintenance and update are performed in the variable load feedback period obtained in step 2, so that the task distribution can be more balanced, and the load balance is realized.
The specific process of load distribution in this step is as follows:
for a set of cluster nodes in WMSP S ═ SiI is more than or equal to 1 and less than or equal to S, based on the scheduling strategy with expected load in step 3, the corresponding forwarding probability is P ═ P1,p2,...,psWhere cluster node SiForward probability P ofiAs in equation (11), with a forwarding probability space of
When the load distributor receives a service request, a random number is generated in the [0,1] space. And determining the target server according to the falling point of the target server in the probability space. Based on the association table of the cluster node load and the service rate established in the step 3, the cluster node with the expected large load obtains a larger probability space and requests service for more services. Because the forwarding of the service request is independent random events, the dynamic independent random probability forwarding also ensures the scheduling balance.
The method realizes a variable feedback period strategy by matching with the access time sequence change of the user, so that the load feedback can reflect the current load of each server cluster node in real time, the consumption of load collection on the system can be reduced, and the service throughput of WMSP is improved; in the task scheduling strategy, an association table of load and service rate is established and maintained, the problem that the real-time service processing capacity of the server cannot be rapidly and effectively obtained is solved, and meanwhile, the server with the optimal processing capacity is selected to process a task request based on the whole load expectation of the WMSP, so that the CPU utilization rate of the WMSP is remarkably improved, and the load imbalance is small. The method highly accords with the time sequence distribution characteristics of concurrent access of users in the WMSP, and the proposed variable feedback period strategy effectively coordinates and balances the contradiction relationship between WMSP service resource consumption and load state acquisition, ensures the optimization of service efficiency and load balance, and effectively improves the service performance and service resource utilization efficiency of the WMSP.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (7)
1. A self-adaptive load balancing method based on a Web map service concurrent access time sequence rule is characterized by comprising the following steps:
s1, acquiring an access log generated in the long-term operation of the WMSP server cluster, performing access to the access log based on different time granularities of a long period and a short period to achieve time sequence analysis, and calculating a variable load feedback cycle arrival rate threshold according to the maximum concurrency number which can be contained by the WMSP server cluster per second and the access log access to achieve time sequence analysis results;
the method for calculating the variable load feedback cycle arrival rate threshold in step S1 includes:
counting WMSP user access logs based on different time granularities of a long period and a short period, wherein the long period counting of the access logs is used for extracting period characteristics of user access, and the short period counting of the logs is used for extracting variable strength characteristics of the user access;
defining the access arrival rate sequence of the user as a time sequence:
H(R,t)＝{R(t1),R(t2)…R(tn)}
wherein R andt is the arrival rate factor and time factor of the user arrival time sequence H (R, t), respectively; r (t)i) Represents tiThe number of service requests arriving in a time period, i.e. the user access strength;
based on the maximum concurrency number M which can be accommodated by WMSP per second and the WMSP long-period access log time sequence analysis, the user arrival rate threshold value which meets the load feedback period change is taken as M0β M, where β is an empirical value;
s2, calculating a variable load feedback period according to the variable load feedback period arrival rate threshold;
s3, calculating the service rate of each node load of the WMSP server cluster, and establishing an association table of the cluster node load and the service rate; calculating load expectation of the WMSP server cluster nodes in a feedback period;
s4, generating a forwarding probability space of the cluster nodes based on load expectation, and obtaining a larger probability space by the cluster nodes with larger load expectation according to the association table of the cluster node load and the service rate; and when the access request arrives at the WMSP server cluster, a random number is generated in the [0,1] space, a target cluster node is determined according to the drop point of the random number in the probability space, and the access request is distributed to the target cluster node.
2. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 1, wherein the method for calculating the variable load feedback period in step S2 is as follows:
s21, if the current arrival rate lambda (t)i)＜M0Then with T0As a duty feedback period, where λ (t)i) Is tiArrival rate of WMSP in time slot, T0A fixed value of the duty feedback period;
s22, if the current arrival rate lambda (t)i)≥M0Calculating the current period TkAverage rate of change of inter-arrival rate
Wherein T iskIndicating the kth load feedbackA period; suppose period TkStarting at time tiThe period is n time intervals, tiThe arrival rate change rate at the time is
Wherein, λ (t)i-1) Is tiLast moment ti-1User access arrival rate; thus obtaining
Further calculating the load feedback period
Wherein, the value L is a correlation coefficient of the feedback period and the load change rate; then, a load feedback period prediction method based on one-time exponential smoothing is used for predicting the subsequent load of the WMSP and calculating a feedback period based on the load
Wherein Fk-1For the load feedback period Tk-1The average rate of change of arrival rate of (a) is an exponential smoothing prediction value, α is a weight coefficient, is an empirical value,
for the load feedback period Tk-1Is measured by the average rate of change observation of the arrival rate of (1).
3. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 2, wherein the load feedback prediction method based on the first exponential smoothing in step S22 specifically comprises:
predicting the average change rate of the user access arrival rate on the basis of the prior data as follows:
giving the users with close time access arrival rate change actual values by calculating an exponential smoothing valuePredicting the average change rate of the user access arrival rate in the next period by using the weight with larger influence on the change predicted value of the user access arrival rate in the next period; wherein, FkIs a period TkIs an exponentially smooth predicted value of the average rate of change of the arrival rate, Fk-1Is a period Tk-1The index smoothing trend prediction value of (1);
is a period Tk-1α is a weight coefficient, is an empirical value, and can obtain a load feedback period matching the current load state
With the proviso that λ (t)i)≥M0And the L value is a correlation coefficient of the feedback period and the load change rate.
4. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 1, wherein the method for calculating the load expectation of the WMSP server cluster node in one feedback cycle in step S3 comprises:
s31, calculating the WMSP server cluster node service capacity, namely the service rate:
in order to quantify the service rate of cluster nodes of the WMSP server, an association table of the load and the service rate of each cluster node is established by a method based on equal interval division; for cluster node SiIs of workload capability Loadi(Loadq-1,Loadq) With a corresponding service rate of muiq；
S32, calculating load expectation of the WMSP server cluster nodes;
clustering nodes S for WMSP serveriBased on the Little law, queuing theory and the obtained service rate muiqAnd calculating to obtain the current cluster node SiLoad expected value E (λ)i) Wherein λ isiIs the user arrival rate.
5. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 4, wherein the method for calculating the service rate in step S31 specifically comprises:
for a group of cluster nodes S ═ { S ] in a WMSP server clusteriI is more than or equal to 1 and less than or equal to S, and the WMSP server cluster node S is madeiHas a workload capacity of Si(Loadmin,Loadmax) Dividing the workload range of the node into Q intervals at equal intervals:
Si([Loadmin,Load1],(Load1,Load2]…(Loadq-1,Loadq]…(LoadQ-2,LoadQ-1](LoadQ-1,Loadmax])
wherein (Load)q-1,Loadq]A load interval corresponding to the q-th interval; with corresponding service rate of muiqAnd the value is the average value of the service rates corresponding to the corresponding load intervals.
6. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 5, wherein the method for calculating the load expectation of the cluster node of the WMSP server in step S32 specifically comprises:
clustering nodes S for WMSP serveriWith a user arrival rate of λiCluster node service rate is muiMean latency between requests of cluster nodes is Di(ii) a The delay time of the request comprises the queuing time of the task and the task processing time;
then according to Little's law, get
The user arrival rate and the service rate on the cluster node determine the average queuing time and the average response time in the cluster node;
suppose that the average user response time expected by WMSP is TeAnd calculating to obtain cluster node SiThe desired average arrival rate of is
Wherein muiRepresenting a cluster node SiA service capability;
when the load of the cluster node is larger, the residual processing capacity is smaller, so that the muiThe value size decreases as the cluster node load increases;
the obtained service rate muiqIntroduction of E (. lamda.)i) And calculating to obtain cluster node SiExpected average arrival rate of users under current load
I.e. cluster node SiIs desired.
7. The adaptive load balancing method based on the concurrent access timing rule of the Web map service as claimed in claim 1, wherein the method for generating the forwarding probability space of the cluster node in step S4 is:
set of cluster nodes S ═ S in WMSP server clusteriI is more than or equal to 1 and less than or equal to S, and the corresponding forwarding probability is P ═ P based on the scheduling strategy expected by the load1,p2,...,ps}；
Wherein the cluster node SiHas a forwarding probability of
The forward probability space is
Wherein P is1+P2+…+Ps＝1。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711024188.3A CN107566535B (en) | 2017-10-27 | 2017-10-27 | Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711024188.3A CN107566535B (en) | 2017-10-27 | 2017-10-27 | Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107566535A CN107566535A (en) | 2018-01-09 |
| CN107566535B true CN107566535B (en) | 2020-05-05 |
Family
ID=61032630
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711024188.3A Active CN107566535B (en) | 2017-10-27 | 2017-10-27 | Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107566535B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110035122A (en) * | 2019-04-04 | 2019-07-19 | 科讯嘉联信息技术有限公司 | A kind of load-balancing method based on dynamic probability model, device and system |
| CN112995265A (en) * | 2019-12-18 | 2021-06-18 | 中国移动通信集团四川有限公司 | Request distribution method and device and electronic equipment |
| CN113377539A (en) * | 2021-06-10 | 2021-09-10 | 中国工商银行股份有限公司 | Processing method and device for realizing load balance |
| CN114090394B (en) * | 2022-01-19 | 2022-04-22 | 山东卓朗检测股份有限公司 | Distributed server cluster load abnormity analysis method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102143022A (en) * | 2011-03-16 | 2011-08-03 | 北京邮电大学 | Cloud measurement device and method for IP network |
| CN102447719A (en) * | 2010-10-12 | 2012-05-09 | 上海遥薇(集团)有限公司 | Dynamic load balancing information processing system for Web GIS service |
| CN102624922A (en) * | 2012-04-11 | 2012-08-01 | 武汉大学 | Method for balancing load of network GIS heterogeneous cluster server |
| CN103188346A (en) * | 2013-03-05 | 2013-07-03 | 北京航空航天大学 | Distributed decision making supporting massive high-concurrency access I/O (Input/output) server load balancing system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110231517A1 (en) * | 2010-03-20 | 2011-09-22 | Sudharshan Srinivasan | Smart download system for mobile devices with multiple data interfaces using enhanced HTTP proxy server |
-
2017
- 2017-10-27 CN CN201711024188.3A patent/CN107566535B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102447719A (en) * | 2010-10-12 | 2012-05-09 | 上海遥薇(集团)有限公司 | Dynamic load balancing information processing system for Web GIS service |
| CN102143022A (en) * | 2011-03-16 | 2011-08-03 | 北京邮电大学 | Cloud measurement device and method for IP network |
| CN102624922A (en) * | 2012-04-11 | 2012-08-01 | 武汉大学 | Method for balancing load of network GIS heterogeneous cluster server |
| CN103188346A (en) * | 2013-03-05 | 2013-07-03 | 北京航空航天大学 | Distributed decision making supporting massive high-concurrency access I/O (Input/output) server load balancing system |
Non-Patent Citations (3)
| Title |
|---|
| A Load-balancing method for network GISs in a heterogeneous cluster-based;Rui Li et al;《Future Generation Computer Systems》;20130228;全文 * |
| 互联网泛在地理信息自动发现关键技术研究;陈万志;《中国博士学位论文全文数据库》;20161231;全文 * |
| 公共地图服务的群体用户访问行为时序特征模型及预测;吴华意等;《武汉大学学报(信息科学版)》;20151008;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107566535A (en) | 2018-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110275758B (en) | Intelligent migration method for virtual network function | |
| CN107566535B (en) | Self-adaptive load balancing method based on concurrent access timing sequence rule of Web map service | |
| CN105491138B (en) | Distributed load scheduling method based on load rate graded triggering | |
| Mapetu et al. | A dynamic VM consolidation approach based on load balancing using Pearson correlation in cloud computing | |
| CN108182105B (en) | Local dynamic migration method and control system based on Docker container technology | |
| CN110764912A (en) | Self-adaptive task scheduler and method | |
| Fu et al. | Layered virtual machine migration algorithm for network resource balancing in cloud computing | |
| CN110933139A (en) | System and method for solving high concurrency of Web server | |
| CN107220108B (en) | Method and system for realizing load balance of cloud data center | |
| CN112685170A (en) | Dynamic optimization of backup strategies | |
| CN112003660B (en) | Dimension measurement method of resources in network, calculation force scheduling method and storage medium | |
| CN113822456A (en) | Service combination optimization deployment method based on deep reinforcement learning in cloud and mist mixed environment | |
| CN112835698A (en) | Heterogeneous cluster-based dynamic load balancing method for request classification processing | |
| CN111752678A (en) | Low-power-consumption container placement method for distributed collaborative learning in edge computing | |
| CN110198267A (en) | A kind of traffic scheduling method, system and server | |
| Gu et al. | A multi-objective fog computing task scheduling strategy based on ant colony algorithm | |
| CN111176784A (en) | Virtual machine integration method based on extreme learning machine and ant colony system | |
| Li et al. | Self-adaptive load-balancing strategy based on a time series pattern for concurrent user access on Web map service | |
| US11374869B2 (en) | Managing bandwidth based on user behavior | |
| CN106888237B (en) | Data scheduling method and system | |
| CN114785692A (en) | Virtual power plant aggregation regulation and control communication network flow balancing method and device | |
| Zhu et al. | Load balancing algorithm for web server based on weighted minimal connections | |
| Beraldi et al. | On the impact of stale information on distributed online load balancing protocols for edge computing | |
| Gao et al. | Deep Reinforcement Learning Based Rendering Service Placement for Cloud Gaming in Mobile Edge Computing Systems | |
| Bao et al. | Nginx-Based Combined Dynamic and Static Load Balancing Algorithm Research |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |