KR100566809B1 - Method for Controlling Cluster System and Computer-Readable Recording Medium Where Program Executing the Method Is Recorded - Google Patents

Abstract

The present invention relates to a cluster system control method or a computer readable recording medium having recorded thereon a program for performing such a control method. The method for controlling a cluster system according to the present invention relates to a method for controlling a cluster system configured using a plurality of nodes and a load balancing server, the method comprising: (1) reading and recording IP addresses and host names of the plurality of nodes; (2) reading and recording the usage amount of the central processing unit of each node to which the IP address and the host name are assigned; and (3) loading the load balancing server by referring to the usage amount of the recorded central processing unit. Changing the step; and (4) repeating the step at predetermined time intervals.

Cluster, real-time, Linux

Description

Method for Controlling Cluster System and Computer-Readable Recording Medium Where Program Executing the Method Is Recorded}

1 is a diagram showing the configuration of a cluster system according to the present invention;

2 is a flowchart illustrating a control method of a cluster system according to the present invention;

The present invention relates to a control method of a cluster system and a recording medium on which a program on which the control method is performed is recorded. In particular, the method of controlling a node distribution of a parallel system constituting a cluster system having a plurality of nodes, and performing such a control method A computer readable recording medium having recorded thereon a program therefor.

As a computing device that performs a complex and vast amount of operations, there has been a single high performance computer, a so-called supercomputer, and is currently used in many research institutes and institutions. However, such supercomputers are extremely expensive and their performance is reaching its limit.

As a replacement for this single high-performance supercomputer, parallel computing technology has emerged, in which multiple central processing units (CPUs) cooperatively calculate a problem. However, computers with this parallel computing technology were also expensive.

Recently released microprocessors are coming out with very high performance, and high-speed networking technology has also evolved so much that it is possible to build a new concept of parallel computers by connecting them to a network. In particular, using an operating system in which source code such as Linux is open, the performance can be freely tuned by a user or a network management expert, which is advantageous for system design.

The cluster system used in the present specification refers to a system in which a plurality of computers are connected to each other so as to be connected to each other and viewed as a system image when viewed from the outside. The computers that make up the cluster are referred to as "cluster nodes" or simply "nodes". In this way, the nodes constituting the cluster system can share storage devices or memory. Cluster nodes are typically connected so as to be networkable using a local area network (LAN) or a wide area network (WAN).

As described above, it is not efficient to connect a plurality of cluster nodes to form a cluster system and to use it only for a specific purpose. Typically, cluster nodes can be operated to provide web services and high-speed computation services. In this case, the resource usage of each cluster node should be properly distributed so that the web services and high-speed computation services do not collide with each other.

In a conventional cluster system, an administrator used a method of manually designating a load of each cluster node by operating a load balancer. In this method, however, the load balancer cannot check the actual usage of the central processing unit (CPU) of each cluster node in real time, so that a cluster node is received when a request such as a web service or a high speed computation service is received from the outside. There was a problem of not properly distributing work among the workers.

The present invention solves such problems of the prior art, and the load balancing server can know the resource usage of each cluster node at predetermined time intervals, and when the work request from the outside comes in, A control method capable of determining which cluster node to allocate a task based on actual resource usage information and controlling a system, and a computer readable recording medium having recorded thereon a program for executing such a control method.

The method for controlling a cluster system according to the present invention relates to a method for controlling a cluster system configured using a plurality of nodes and a load balancing server, the method comprising: (1) reading and recording IP addresses and host names of the plurality of nodes; (2) reading and recording the resource usage of each node to which the IP address and the host name are assigned; (3) changing the load of the load balancing server with reference to the recorded resource usage; 4) repeating the above steps at predetermined time intervals.

According to this method, the resource usage of each cluster node is read and the process is repeated at predetermined time intervals so that the load balancer monitors the usage of each cluster node in near real time and based on the request for work from outside. Correspondingly, you can schedule cluster nodes. Therefore, it is possible to schedule according to the resource usage of the actual cluster node, thereby improving the efficiency and stability of the cluster system.

The method for controlling a cluster node according to the present invention may further include sending a service processing request to a node of a low load amount according to a predetermined rule with reference to the changed load balance server load amount when a new service request is received.

In addition, the cluster system according to the present invention is preferably Linux-based. Since the source code of Linux is disclosed, system administrators and the like can freely tune the system, so that the cluster system can be configured according to the actual system environment.

The usage amount of the central processing unit (CPU) of each node may be used as the resource usage amount.

When configuring a cluster system based on Linux, the step of repeating at predetermined time intervals is preferably performed using a cron program. If you use this cron program, you can register the program in the crontab and execute the resource usage monitoring program of the cluster node at a certain time. In particular, the administrator can control the time to check the resource usage in the crontab. have. In other words, if the program is compiled and compiled, it is easier to manage than other programs that cannot be modified or adjusted later unless the source code is known.

In addition, the cluster nodes may perform a task of processing a web service and a high speed computation service.

The present invention also proposes a computer readable recording medium having recorded thereon a program on which the above-described control method is performed. The recording medium includes a plurality of nodes and a load balancing server that provide a web service and a computing service. A computer readable recording medium having recorded thereon a control program for performing control of a cluster system, the program comprising: (1) reading and recording IP addresses and host names of the plurality of nodes; and (2) the IP Reading and recording resource usage of each node to which an address and a host name are assigned; (3) changing a load of the load balancing server by referring to the recorded usage of the central processing unit; and (4) The step is repeated at predetermined time intervals.

Hereinafter, with reference to the accompanying drawings will be described in detail an embodiment of the present invention.

1 shows a block diagram of a cluster system 1 in which a control method according to the invention is carried out.

As shown in FIG. 1, the cluster system 1 according to the present invention includes a load balancing server 10, a plurality of cluster nodes 30-1, 30-2,..., 30-n, and a load. LAN / WAN network that connects the distributed server 10 and a plurality of cluster nodes (30-1, 30-2, ..., 30-n) to enable networking.

The cluster system 1 is connected to the external user terminal 100-1, 100-2, 200-1, 200-2, 200-3 by the Internet or an intranet.

The cluster system 1 is seen from the outside as a system, that is, a virtual server 2. The cluster nodes 30-1, 30-2,..., 30-n perform the function of a real server performing real work in the cluster system 1.

The virtual server 2 according to the present invention operates using IP tunneling, which is a cluster node 30-1, 30-2, .. ., 30-n). That is, when the user 100-1, 100-2, 200-1, 200-2, 200-3 requests a service, the user first connects to the external IP address of the load balancer 10, that is, the virtual IP address. The distributed server 10 checks the destination address and port number of the packet desired by the user. If the checked content matches the service described in the predetermined virtual server rule table, each cluster node 30-1, 30-2, ..., 30-n is selected according to the scheduling algorithm. At the same time, a new connection is added to the hash table that records the connection made. After adding the connection, the load balancing server 1 wraps the packet in the IP datagram and cluster nodes 30-1, 30-2. .., 30-n). The cluster nodes 30-1, 30-2,..., 30-n receiving the wrapped packet decompress the packet again, process the service requested by the user, and finally the cluster nodes 30-1, 30-n. 2, ..., 30-n) return the result directly to the user according to the routing table.

However, it should be understood that the scope of the present invention is not limited to the virtual server using the IP tunneling scheme. In addition, in describing the present invention, a Linux-based system is described, but it should be clearly understood that the scope of the present invention is not limited thereto.

When the load balancer 10 receives a service request from an external user 100-1, 100-2, 200-1, 200-2, 200-3, the cluster node 30-1, 30-2 goes through a scheduling process. , ..., 30-n). In the present invention, a weighted least-connection scheduling method is used, which weights each cluster node 30-1, 30-2, ..., 30-n. To receive more requests from higher nodes.

According to the present invention, the resource usage of each cluster node (30-1, 30-2, ..., 30-n) is periodically checked at predetermined time intervals to automatically determine the weight to be used in the load balancing server 10. The method to give is proposed.

2 shows a flow diagram of this method.

In order to perform the control method of the cluster system according to the present invention, an IP address and a hostname of each cluster node 30-1, 30-2, ..., 30-n must first be known. To know the IP address of each node, get information from the "etc / hosts" file, a directory in Linux, and record it (step S21). In the process, each node's IP and hostname are registered as shown below.

127.0.0.1 localhost.localdomain localhost

210.125.177.xxx node01.cluster.co.kr node01

210.125.177.xxx node02.cluster.co.kr node02

210.125.177.xxx node03.cluster.co.kr node03

210.125.177.xxx node04.cluster.co.kr node04

210.125.177.xxx node05.cluster.co.kr node05

210.125.177.xxx node06.cluster.co.kr node06

In this example, 127.0.0.1 is specified as the local host network number, which allows the local system to run network software in a networkless environment.

Each Internet address has a corresponding hostname, which allows users 100-1, 100-2, 200-1, 200-2, and 200-3 to recognize the system as a hostname. A hostname can have one or more aliases, which are names that can recognize the host on the network. In this example, the alias is created as node **.

In this way, the IP address and hostname of each cluster node are obtained from the "etc / hosts" file of the Linux system. The execution code of the program written in C ++ language may be as follows.

ifstream real_server_name;

real_server_name.open ("etc / hosts");

Next, the flow advances to step S22 to read the usage amount of the central processing unit (CPU) of each node.

In the present invention employing Linux, a remote shell (rsh) command is used to obtain the central processing unit usage of each cluster node 30-1, 30-2, ..., 30-n. An example of a specific command is as follows.

rsh real_server_name ps aux │awk ＼ '{print $ 3} ＼'> real_server_name_cpu

In this command, rsh is the name of the command, and real_server_name is the hostname of the server to get the central processing unit usage. And ps aux | awk ＼ '{print $ 3} ＼' refers to the option of the rsh command to receive the usage of the central processing unit from another computer. > real_server_name_cpu means a command to record the usage of the central processing unit of each cluster node obtained in a file called real_server_name_cpu.

Next, the flow advances to step S23 to change the load of the load balancing server by referring to the usage of the central processing unit.

In the related art, when weighting each cluster node in the load balancing server 10, a command such as the following example is input to a command input window to be manually assigned.

$ IPVSADM -a -t 210.125.177.155 -r 192.168.1.5 -g -w 16

$ IPVSADM -a -t 210.125.177.155 -r 192.168.1.6 -g -w 7

$ IPVSADM -a -t 210.125.177.155 -r 192.168.1.7 -g -w 1

In other words, the command to manually weight each node in the Linux system,

$ IPVSADM -a -t [IP address of load balancing server] -r [node IP address] -g -w [weight].

In the present invention, the final weight portion is given with reference to the central processing unit usage of each cluster node read in step S22 described above. For example, if the central processing unit of the cluster node 30-2 that is performing the high-speed operation is 99%, the weight of 99% is applied to the node to request the service from the outside. Do not send work requests.

However, in the cluster system 1, the amount of use of the central processing unit of each of the cluster nodes 30-1, 30-2, ..., 30-n may change over time. The above steps are repeated at time intervals of.

When the above step is finished, the process proceeds to step S24, after which time is read, the process proceeds to step S25. If it is determined in step S25 that the predetermined time has elapsed, the process returns to step S22 to read the usage amount of the central processing unit of each node again and repeat the above-described steps.

If it is determined in step S25 that the predetermined time has not elapsed, the process returns to step S24 and waits until a predetermined time interval is reached.

This kind of repetition at regular intervals uses Linux's cron. Automatic task scheduling with cron is divided into creating a simple shell script for the task and defining the executable script to be automatically executed in cron using the crontab.

To use cron, use the command etc / crontab. The following is the contents of / etc / crontab.

SHELL = / bin / bash

PATH = / sbin: / bin: / usr / sbin: / usr / bin

MAILTO = root

HOME = /

# run-parts

01 * * * * root run-parts /etc/cron.hourly

02 4 * * * root run-parts /etc/cron.daily

22 4 * * 0 root run-parts /etc/cron.weekly

42 4 1 * * root run-parts /etc/cron.monthly

* The table means all, which means run every time.

The content of 01 * * * * means to run every month, every week, every day, every 01 minutes. That means run it once an hour.

run-parts The contents of the /etc/cron.hourly section mean to run scripts in the /etc/cron.hourly/ directory. In other words, if you put the script file in the /etc/cron.hourly/ directory, the script will run once an hour.

Similarly, the second line means to run the script file in the /etc/cron.daily/ directory at 4:02 every day, and the third line to run the script file every 4:22 every Sunday.

When a program for performing the method according to the present invention is registered in the crontab, the program is executed at predetermined time intervals to periodically check the central processing unit of each cluster node 30-1, 30-2, ..., 30-n. The usage is checked to control the load balancing server 10.

Using crontabs like this allows administrators to actively control the program. That is, according to the general program, once compiled and converted into an executable file, the contents of the program cannot be changed unless it has source code. All you have to do is make it very easy to manage your cluster systems.

The above-described control method of the cluster system is preferably performed by a computer program executable on a computer, and such computer program is recorded and stored in a computer-readable recording medium.

While preferred embodiments of the present invention have been described with reference to the accompanying drawings, the scope of the present invention is defined by the description of the appended claims, and should not be construed as limited to the embodiments or drawings described above. . In addition, it should be clearly understood that improvements, changes, and modifications apparent to those skilled in the art described in the claims are included in the scope of the present invention.

According to the present invention, in controlling a cluster system consisting of a plurality of cluster nodes that provide a web service and a high-speed computation service, the resource usage of each cluster node can be periodically checked in real time or at regular time intervals. The load balancing server adjusts the weight of each node with reference to the resource usage of the checked cluster nodes, and thus it is possible to control the system organically without conflicting resource usage between nodes.

Claims (7)

In the control method of a cluster system configured using a plurality of nodes and a load balancing server, Reading and recording the IP addresses and host names of the plurality of nodes; Reading and recording resource usage of each node to which the IP address and host name are assigned; Changing the load amount of the load balancing server by referring to the recorded resource usage; Repeating said step at predetermined time intervals. The control method according to claim 1, further comprising sending a service processing request to a low load node according to a predetermined rule with reference to the changed load balancing server load amount when a new service request is received. The method according to claim 1 or 2, wherein the cluster system is Linux-based. The control method according to claim 1 or 2, wherein the resource usage is usage of a central processing unit (CPU) of each node. 4. The method of claim 3, wherein the step of repeating at predetermined time intervals is performed using a cron program. 3. The method of claim 1, wherein the nodes process a web service and a high speed computation service. 4. A computer-readable recording medium having recorded thereon a control program for performing control of a cluster system configured by using a plurality of nodes and a load balancing server, wherein the program includes: Reading and recording the IP addresses and host names of the plurality of nodes; Reading and recording resource usage of each node to which the IP address and host name are assigned; Changing the load amount of the load balancing server by referring to the recorded resource usage; And repeating said step at predetermined time intervals.

Images