KR20070032441A - Load balancing system based on fuzzy grouping and the load balancing method - Google Patents

Abstract

A system and a method for balancing load based on fuzzy grouping are provided to select the fittest server by monitoring a load state of each server in real-time, to distribute the load efficiently/stably by dynamically allocating the load to the servers, and reduce message overhead among the servers by allocating task to the server of the smallest load state. A load monitor(103) monitors load information of each server object(102) in real-time by each group. A fuzzy load balancer(104) infers service priority by using the collected load information. A global scheduler(105) allocates the task to the fittest server object based on the inferred service priority. The load monitor periodically transmits the collected load information to the fuzzy load balancer. The service priority is inferred based on a fuzzy rule. The fuzzy rule is defined by using CPU time usage and memory usage collected from each server object as an input value.

Description

Load balancing system based on fuzzy grouping and its load balancing method {LOAD BALANCING SYSTEM BASED ON FUZZY GROUPING AND THE LOAD BALANCING METHOD}

1 is a block diagram illustrating a fuzzy grouping based load balancing system according to the present invention;

2 is a diagram illustrating a fuzzy rule and a membership graph used to infer service priorities according to the present invention;

3 is a view showing a grouping method according to the present invention;

4 is a flowchart illustrating a fuzzy load balancing operation according to the present invention;

5 is a flowchart illustrating a fuzzy grouping-based load balancing method according to the present invention;

6 is a graph showing the average response time and throughput according to the number of servers and the number of jobs according to the present invention;

Explanation of symbols on the main parts of the drawings

101: Client 102: Server Object

103: load monitoring means 104: purge load balancing means

105: global scheduling means

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a Distributed Object Computing System. In particular, a fuzzy grouping-based load balancing system and its load are distributed and allocated so that loads are distributed evenly among server objects operated in parallel. To a balancing method.

In general, a distributed environment is an environment in which system resources are separated from each other in terms of physical or configuration. The three-tier concept is closely related to the distributed environment. Here, a tier is a column or hierarchy with a series of similar objects listed. Thus, Tier 3 refers to three similar objects. In computer programming, parts of a program can exist in several layers, which can also be located on different computers on the network. In the third tier, databases, middleware applications, and user interfaces are all separated, so smooth communication between layers is an important issue. Each layer is physically on another computer, or all on one computer. However, not all concepts are encompassed, so that each class should be able to make good use of each other at the same time without having strong binding power on each other.

Distributed objects used in distributed environment are individual objects that exist in various system resources that are physically or conceptually separated. The technology for smooth communication between these objects is the core of distributed objects. Distributed environment refers to all cases where hierarchical structure is divided like three tiers, or even objects belonging to the same hierarchically exist in several system resources horizontally. The technology that allows you to send and receive is distributed environment technology.

In a distributed object computing system implemented in a distributed environment, server objects process service requests received from a remote client. Therefore, a load balancing technique, which is a technology of selecting a server that handles an appropriate service and distributing work, is important. This is because proper load balancing can increase the performance, efficiency, and reliability of the entire server system.

The application has a task to do and at least a processor instruction to execute. The faster the computer running the application, the faster the instruction is executed, but the processor speed is limited. As the number of concurrent requests the server receives increases, if the workload becomes too large for the computer's capacity, the application will exceed its target response time. There are two solutions: first, using faster computers, and second, using more computers. The solution using a faster computer is not always applicable and also expensive. In addition, no matter how fast a computer is, eventually it will need faster performance. Any efficient design should include the possibility that the application will need to be installed on multiple servers. This is called horizontal scalability.

One technique for achieving horizontal scalability is load balancing. Load balancing is a mechanism to ensure that different computers efficiently use different processor capacities. Load balancing techniques should pass requests to the least busy resources. However, in some load balancing techniques such routing may be difficult, and may not be suitable for some applications, especially session-based applications. Therefore, you need to determine the load balancing mechanism that is most appropriate for your application.

Conventional load balancing techniques mainly use random and round-robin techniques. Because it's easy to implement, simple, and easy to manage. The random scheme allocates requests in random order when there are service requests. This technique takes a method of unconditionally allocating requests to any selected server without considering server load conditions. Round-robin is a method of allocating requests according to a predetermined order of servers when there are service requests. Like the random technique, this technique is a static algorithm that unconditionally allocates service requests according to a predetermined allocation order without considering server load conditions.

An example of such load balancing is disclosed in Korean Patent Registration Publication No. 10-0271199 (Distributed Computing System Using Logical Processing and Method).

The technique disclosed in this publication provides load balancing of a distributed network system by providing server processes that include two levels of abstraction, both logical and physical, whereby physical processes have addresses supported by a transport mechanism, On the other hand, since logical processes do not have such addresses, each logical process is implemented by a series of physical processes, the system maintains a mapping between the two, and the system also allows the client to send messages to the logical process. And a message is disclosed for a distributed computing system and method using logical processing that provides interfaces that automatically redirect to an appropriate physical process.

In addition, an example of load balancing is disclosed in Korean Patent Laid-Open Publication No. 2005-0043616 (method and system for operating a cluster of servers and a computer-readable recording medium).

The technique disclosed in this publication discloses a system, method, and program product for operating a server cluster, wherein a load balancer that is separate from the server selects a server that processes each work request assigned to the cluster based on a load balancing algorithm. Send the work request to the selected cluster, the load balancer receives the request from a server other than the current member of the cluster, and the request requests that the other server be a member of the cluster that processes the work request, In response to the request, the load balancer joins the other server as a member of the cluster to process a predetermined job request that is subsequently assigned to the cluster, and the other server load thresholds for the other server against the load balancer. , The other server is configured for the load balancer. Specify a timeout to be used for the existing server, the timeout operating a cluster of servers representing the time allotted to the other server to send the hello message after sending the previous hello message and the computer read A possible recording medium is disclosed.

However, in the technique disclosed in the above-mentioned publication 10-0271199, it is load balancing using two levels of logical and physical abstraction, and in the technique disclosed in the above publication 2005-0043616, the load for allocating work requests to servers in a cluster. There was a problem in that load balancing was performed in a predetermined order or in an arbitrary order without considering the state of the entire system by balancing.

For example, there is a distributed processing system where the first request is assigned to the first server, the second request is assigned to the second server, and the third request is assigned to the third server. If this process is repeated and the fourth request goes back to the first server, if the first server receives a very large data file request that takes several minutes to process and fails to process the first request, the processing of the fourth request will be delayed. There is no choice but to. However, at the time when the fourth request comes in, even if the second server has already completed the second request, the prior art and the above-mentioned technologies are assigned to the first server without any consideration of the server status. That is, some servers are in a rest state while some servers are overloaded, resulting in a problem of using the server quite inefficiently.

An object of the present invention is to solve the problems as described above, to provide a fuzzy grouping-based load balancing system and load balancing method that can monitor the load status of each server in real time to select the most suitable server will be.

Another object of the present invention is to provide a fuzzy grouping-based load balancing system and a load balancing method for dynamically allocating load to a server to enable more efficient and stable load balancing.

Another object of the present invention is to provide a fuzzy grouping-based load balancing system and a load balancing method capable of reducing message overhead between servers by allocating tasks to servers having the least load using the concept of grouping.

In order to achieve the above object, a fuzzy grouping-based load balancing system according to the present invention includes a plurality of server objects connected to a client through a network to perform a request of the client, and each server object operated in parallel to performance. A load balancing system for dynamically allocating loads to respective server objects by grouping the loads, wherein the load monitoring means monitors load information of each server object in real time for each grouped group; Fuzzy load balancing means for inferring, and global scheduling means for allocating work to a server object most suitable according to the inferred service priority, wherein the load monitoring means periodically transmits the collected load information to the fuzzy load balancing means. Characterized by The.

In addition, in the fuzzy grouping-based load balancing system according to the present invention, the service priority is inferred by a fuzzy rule defined by using CPU time and memory usage collected from each server object as input values. It is done.

In addition, in the fuzzy grouping-based load balancing system according to the present invention, the grouping is calculated by dividing the number of activated server objects by the number of server objects per group, calculating the number of groups, and calculating the group number in the rank number of server objects to be grouped. The number of server objects to be included in the same group is calculated by adding the number.

In addition, in the fuzzy grouping-based load balancing system according to the present invention, when a new server object is added, the global scheduling means adds the new server object to the highest load group.

In addition, in the fuzzy grouping-based load balancing system according to the present invention, when a failed server object occurs, the work of the failed server object is reassigned to another server object according to a load state of another group.

In addition, in the fuzzy grouping-based load balancing system according to the present invention, when a new group is added, the global scheduling means allocates a service to the newly added group according to the updated service priority.

In addition, in order to achieve the above object, the fuzzy grouping-based load balancing method according to the present invention includes a plurality of server objects connected to a client through a network to perform a request of the client, and each server object operated in parallel. A load balancing method for dynamically allocating load to each server object by grouping according to performance, the method comprising: receiving a service request from the client; monitoring and collecting load information of the server objects by load monitoring means; Inferring the service priority of each server group by the fuzzy load balancing means, and allocating the request of the client to the server object most suitable for the global scheduling means according to the service priority inferred in the fraud inference step. The load monitor Means is characterized in that it periodically transmits the collected information to the load balancing means for the purge rod.

In addition, in the fuzzy grouping-based load balancing method according to the present invention, the inferring of the service priority may infer the usage of the CPU time and the memory usage collected by the load monitoring unit from each server object using a fuzzy rule. Characterized in that.

In addition, in the fuzzy grouping-based load balancing method according to the present invention, the grouping is calculated by dividing the number of activated server objects by the number of server objects per group, calculating the number of groups, and calculating the group number in the rank number of server objects to be grouped. The number of server objects to be included in the same group is calculated by adding the number.

In addition, in the fuzzy grouping-based load balancing method according to the present invention, when a new server object is added, the global scheduling means adds the new server object to the group having the highest load state.

In addition, in the fuzzy grouping-based load balancing method according to the present invention, when a failed server object occurs, the failed server object may be reassigned to another server object according to a load state of another group.

Further, in the fuzzy grouping-based load balancing method according to the present invention, when a new group is added, the global scheduling means allocates a service to a newly added group according to the updated service priority.

Hereinafter, the most preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily practice the present invention. .

1 is a block diagram illustrating a fuzzy grouping based load balancing system according to the present invention.

As shown in FIG. 1, 101 is a client requesting a service, 102 is a server object performing a request of the client 101, and a plurality of server objects 102 are each server object 102. They are grouped according to their performance and run in parallel.

Numeral 103 is a load monitoring means for monitoring load information of each server object 102 of the grouped group in real time, and 104 is inferred of service priority by receiving load information from the load monitoring means 103. The fuzzy load balancing means 105 is a global scheduling means for receiving a service priority from the fuzzy load balancing means 104 and assigning a task to a server object most suitable for a task to be executed.

The server objects 102 shown in FIG. 1 group two server objects into one group and place one load monitoring means 103 per group. Since load balancing requires accurate system state information, the load monitoring means 103 in each group serves to collect load information of such server object state. The load monitoring means 103 also transmits load information to the purge load balancing means 104. The load monitoring means 103 measures, for example, the CPU time usage and the memory usage, which are load information of the two server objects 102, every two seconds, and the measured load information is fuzzy load balancing in the form of an asynchronous event. Send to means 104. Through this load information, the fuzzy load balancing means 104 infers the service priority through the fuzzy rule and transmits the result to the global scheduling means 105. The global scheduling means 105 then selects the most appropriate server object 102 based on the service priority and assigns the request of the client 101 to the selected server object 102. In the load balancing system based on the fuzzy grouping according to the present invention, instead of applying a polling policy in which the fuzzy load balancing means 104 queries the load monitoring means 103 periodically, the load monitoring means 103 ) Uses a push policy that periodically measures and sends load information to the fuzzy load balancing means 104.

Next, a rule applied to the system shown in FIG. 1 will be described with reference to FIG. 2.

2 is a diagram illustrating a fuzzy rule and a membership graph used to infer service priorities according to the present invention.

The fuzzy rules and membership graphs shown in FIG. 2 are used by the fuzzy load balancing means 104 to determine service priorities. The fuzzy load balancing means 104 can efficiently control complex and ambiguous nonlinear systems using fuzzy rules using fuzzy logic control based condition control (IF-THEN) rules. Therefore, the fuzzy load balancing means 104 uses this algorithm for efficient load balancing, and sets the membership graph and the fuzzy rule necessary for this as shown in FIG. 2 (a, b) shows the usage of the CPU time and the memory usage in a membership graph, which is used as an input value of the fuzzy rule. In addition, Figure 2 (c) shows the service priority in a membership graph, which is used as a result of the fuzzy rule.

CPU time usage is 'low, L', 'less than moderate (LMO)', 'moderate (MO)', 'high, H', and 'very high' (very high, VH) 'can be defined and classified. In addition, the memory usage is defined as a fuzzy set of 'small (S)', 'less than medium (LME),' medium (ME), 'large (L)' Can be classified. In order to determine the most suitable server using the two load information, service priority is classified into seven categories as shown in FIG. The fuzzy rule required to infer this priority is defined as shown in FIG. VL is 'very low', L is 'low', LME is 'less than medium', ME is 'medium' and MME is 'more than normal' than medium) ', H means' high' and VH means' very high '.

For example, if the CPU time usage is LMO and the memory usage is S, the service priority will infer the VL value. The fuzzy load balancing means 104 infers and selects the most suitable server object to enable more efficient load balancing using this fuzzy logic control.

Next, grouping of servers of the system shown in FIG. 1 will be described with reference to FIG. 3.

3 is a diagram illustrating a grouping method according to the present invention.

As shown in FIG. 3, six server objects are grouped into three groups. Grouping is based on the performance of the CPU and memory. Since the performance of each server object is different, it is necessary to check the performance of each server object in order to equalize the performance of the group.

First, in order to group, a rank number is determined according to each CPU and memory performance, and the number of groups to be generated is calculated according to the following formula.

[Number of groups = number of active server objects / number of server objects entering per group]

Next, the rank number of the server object to be included in the same group is calculated by adding the calculated number of groups to the rank number of the server object to be grouped.

[Rank number of server objects to be included in the group = rank number of server objects to be grouped + number of groups]

The server objects with the rank number of the server object to be grouped and the rank number of the server object to be included in the group are created as one group.

For example, if the number of active server objects is six and the number of server objects entering per group is two, the number of groups is three. Now add the rank number and the number of groups to group. Rank number 1 equals 4, the number of groups. That is, the server object with rank number 1 becomes the same group as the server object with rank number 4. This calculation makes {1, 4}, {2, 5}, and {3, 6} the same group.

If a new server object 102 is added, the global scheduling means 105 adds the new server object 102 to the group of highest load.

In addition, when a failed server object 102 occurs, the work of the failed server object 102 is reassigned to another server object 102 according to the load status of another group.

In addition, when a new group is added, the global scheduling means 105 allocates a service to the newly added group according to the updated service priority.

Next, a fuzzy grouping-based load balancing method according to the present invention will be described with reference to FIGS. 4 to 5.

4 is a flowchart illustrating a fuzzy load balancing operation according to the present invention, and FIG. 5 is a flowchart illustrating a fuzzy grouping based load balancing method according to the present invention.

First, the global scheduling means 105 receives a service request from the client 101 (ST 501). Then, the load monitoring means 103 monitors and collects load information of the server objects 102 (ST 502), and transmits the collected load information to the fuzzy load balancing means 104. The fuzzy load balancing means 104 infers the service priority of each group based on the transmitted load information (ST 503), and transmits the inferred service priority to the global scheduling means 105. The global scheduling means 105 allocates the request of the client 101 to the most suitable server object 102 according to the transmitted service priority (ST 504).

As mentioned above, although the invention made by this inventor was demonstrated concretely according to the said Example, this invention is not limited to the said Example and can be variously changed in the range which does not deviate from the summary.

In addition, when the present invention is commercialized, it can be applied to process a transaction in a large e-business enterprise or an e-Commerce organization in which a plurality of servers are grouped and operated. It is very useful for Internet Service Providers (ISPs) that host many servers as well as many websites. In addition, due to the activation of the Internet shopping mall and the online game industry that are accessed by a large number of users, related companies need to use the server efficiently. In addition, load balancing is an important part of VPNs and is expected to be utilized in the future.

The present invention compared and analyzed conventional load balancing techniques to build a more advanced optimization model and to generate a better intelligent load balancing model through simulation.

In this simulation, we compared the average response time and throughput for our work with the existing random and round-robin techniques. The task arriving at each server modeled the Poisson distribution with λ = 6 and assumed that the task arrived at the same rate on all servers. In addition, service time, CPU time usage, and memory usage for work requests are regularly distributed. This simulation environment compares the average response time and throughput when there are 6 servers when there are 1500 to 2 jobs and 6 servers when there are 100 to 3000 jobs. same.

6 is a graph showing the average response time and throughput according to the number of servers and the number of jobs.

In Figure 6 it can be seen that the average response time and throughput is superior to other conventional techniques. As shown in FIG. 6, the throughput increases as the number of servers increases, and the throughput decreases as the number of tasks increases. In addition, the purge group according to the present invention has the highest throughput. This result is because the present invention uses the grouping concept to allocate work in the least load state, and the grouping concept also reduces the message overhead between servers.

As described above, according to the fuzzy grouping-based load balancing system and the load balancing method according to the present invention, it is possible to monitor the load state of each server in real time to select the most suitable server.

In addition, according to the fuzzy grouping-based load balancing system and the load balancing method according to the present invention, an effect of improving the stability and performance of the system can be obtained by dynamically allocating load to the server.

In addition, according to the fuzzy grouping-based load balancing system and the load balancing method according to the present invention, it is also possible to reduce the message overhead between servers by assigning tasks to the server with the least load using the grouping concept. Lose.

Claims (12)

A load balancing system having a plurality of server objects connected to a client through a network to perform a client's request, and dynamically assigning a load to each server object by grouping each server object operated in parallel according to performance. Load monitoring means for monitoring load information of each server object in real time for each of the grouped groups; Fuzzy load balancing means for inferring a service priority using the collected load information; Global scheduling means for allocating work to the most suitable server object according to the inferred service priority; The load monitoring means is a fuzzy grouping-based load balancing system, characterized in that for periodically transmitting the collected load information to the fuzzy load balancing means. The method of claim 1, The service priority is inferred by a fuzzy rule defined using the CPU time and memory usage collected from each server object as input values. The method of claim 1, The grouping calculates the number of groups by dividing the number of activated server objects by the number of server objects per group, and calculates the rank number of server objects to be included in the same group by adding the calculated number of groups to the rank number of server objects to be grouped. Fuzzy grouping based load balancing system, characterized in that made. The method of claim 1, The fuzzy grouping-based load balancing system, wherein when the new server object is added, the global scheduling means adds the new server object to the group under the highest load. The method of claim 1, A load balancing system based on fuzzy grouping, wherein when a failed server object occurs, the work of the failed server object is reassigned to another server object according to a load condition of another group. The method of claim 1, If a new group is added, the global scheduling means assigns a service to the newly added group according to the updated service priority. A load balancing method comprising a plurality of server objects connected to a client through a network to perform a client request, and dynamically assigning a load to each server object by grouping each server object operated in parallel according to performance. Receiving a service request from the client, Monitoring and collecting load information of the server objects in a load monitoring means; Inferring the service priority of each server group by the fuzzy load balancing means according to the load information, Allocating a request of the client to a server object most suitable for global scheduling means according to the service priority inferred in the reasoning step; The load monitoring means is a fuzzy grouping-based load balancing method, characterized in that for transmitting the collected load information to the fuzzy load balancing means periodically. The method of claim 7, wherein And inferring the service priority according to the fuzzy grouping-based load balancing method, wherein the load monitoring means infers the CPU time usage and the memory usage collected from each server object by using a fuzzy rule. The method of claim 7, wherein The grouping calculates the number of groups by dividing the number of activated server objects by the number of server objects per group, and calculates the rank number of server objects to be included in the same group by adding the calculated number of groups to the rank number of server objects to be grouped. Fuzzy grouping based load balancing method, characterized in that made. The method of claim 7, wherein The fuzzy grouping-based load balancing method according to claim 1, wherein when the new server object is added, the global scheduling means adds the new server object to the highest load group. The method of claim 7, wherein A method for load balancing based on fuzzy grouping, wherein when a failed server object occurs, the work of the failed server object is reassigned to another server object according to a load condition of another group. The method of claim 7, wherein If a new group is added, the global scheduling means allocates a service to the newly added group according to the updated service priority.

Images