JPH06110851A - Load distribution control method for computer system - Google Patents

Abstract

PURPOSE:To distribute the load of an entire computer in a work environment where the multiwindows of plural users are used. CONSTITUTION:The load information on the computers 141 and 142 are controlled by a load table 139 of each computer. Thus the computer that carries out a window application can be automatically decided based on the load information. Therefore the load distributing efficiency is improved. In particular plural window applications are simultaneously carried out when a multi-window environment is used by a user. Thus the highly efficient load distribution control is secured for each application.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a load balancing control method between computer systems connected via a network.

[0002]

2. Description of the Related Art In a client / server type multi-window system, a window server process (hereinafter referred to as a server process) that manages an input / output device is started to execute an actual application program on a network-connected computer group. Can be run.

FIG. 12 shows a block diagram of a conventional multi-window system. The computer 108 is a communication unit 101 capable of communicating with all computers on the network.
And an input / output device 103 that receives an input from a user and passes it to a window server process (hereinafter referred to as a server process), and an input / output device 103 or an input / output device of another computer through the server process. The window application 102, the startup file 111 used to start the window application when starting the server process, and the input / output of the window application are managed, and the input / output device 103 is used by using another computer and the communication means 101. It has a server process 105 for exchanging the input received from the server and the output information received from the window application.

The user 106 first sets the server process 10
5 is started by the computer 108. Furthermore, the window application 11 on the computer 109 is used as an application program.
0 can be activated. This window application 11
The input / output of 0 is passed to the server process 105 of the computer 108 through the communication means 107, and is exchanged with the user 106 through the input / output device 103. In this way, the application program in any computer on the network can be executed, and its input / output can be performed by the input / output device of the computer in which the server process is activated. The user 106 can statically specify the application and the computer that starts the application in the startup file 111. When the user 106 starts the server process,
The server process 105 starts the application program on the specified computer based on the startup file 111, and the started application program is the input / output device 103 of the computer 108.
Exchange information between them.

Conventionally, a computer for activating an application program has been designated by a user or statically designated in a file executed when a server process is activated. FIG. 13 shows a situation where the user is using a plurality of application programs.

Both the users 85 and 86 execute the application program statically specified by the start files 87 and 88 when the server process is started. In the computer 1 (82) and the computer 2 (83), users 85 and 86, respectively.
Runs a server process to run multiple application programs on another computer. The user 85 executes the application program c1 on the computer 5 (80) and executes the application programs c2 and c3 on the computer 6. Further, the user 86 uses the application program c2.
Is executed by the computer 5 (80) and c1 is executed by the computer 6.

However, this method has the following problems.

When a new computer is connected to the network A state in which a new computer 7 is added to the state shown in FIG. 13 is shown in FIG. In this case, the users 155, 15
Since the computer 6 statically specifies that the application programs are started by the computers 5 and 6, to execute the application program on the computer 7 and distribute the load to the computers on the network, the startup file of FIG. It is necessary to rewrite 87 and 88.

FIG. 15 shows a case where the user 85 does not start the application program in FIG. 13 when the user does not use the application program. In this case, only the application program started by the user 86 is executed. In this case, the computer 80 is activated with only one c2, and the computer 81 is activated with two c1. If c1
Is a program with a very large load, even though the load on the computer 80 is small, the
Since 1 is executed twice, the load balance becomes poor.

As described above, when the computer for starting the application program is statically designated, even if the load of the computer group connected by the network fluctuates, the computer for starting the program corresponding to the fluctuation is selected. It was difficult to change and balance the load.

Further, when the user designates a computer and starts the application program, in order to distribute the load on the computer on the network, the user checks the load on each computer and then selects the computer that executes the program. It was necessary to make a decision.

[0012]

Since the computer for executing the application program is designated by the user at the time of start-up or statically designated, the computer is biased toward a part of a plurality of computers, or the new computer is executed. Even if a calculator is added,
If the user does not recognize it, the computer cannot be used, and if the user does not modify it, it is difficult to distribute the load among the computers connected by the network. Further, when trying to balance the load, the user had to decide which computer to start the application on the basis of the load status of all the computers on the network, and the operation was difficult.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to efficiently perform load distribution of all computers on a network in a work environment using a multi-window of a plurality of users. .

[0014]

Means and Actions for Solving the Problems
In a computer system composed of a plurality of computers connected via a network, a communication means for communicating with all the computers on the network, and a load state for managing the load state of the computer and other load information using the communication means. Load information management means for exchanging with other computers and storing and managing it.
The display device of the computer is capable of displaying a plurality of windows at the same time, and includes a window server means for managing the input / output of the windows, and a means for determining a computer to execute a window application according to load information managed by the load information managing means. This is the first feature.

Further, according to the present invention, in a computer system comprising a plurality of computers connected via a network, a plurality of windows can be simultaneously displayed on a display device of the computer, and a window server means for managing the input / output thereof is provided. , History management means for storing the history of the load applied to the computer by the window application, load information management means for recording the load of the computer and transmitting information to other computers, and server means or load information management means for each computer Communication means for exchanging information transmitted and received by the computer, and means for determining a computer that executes a window application by associating a history with a window by the history management means and the history management means. This is the second feature.

In this computer system, the load information of a plurality of computers is managed, and the computer that executes the window application can be determined according to the load information. Therefore, efficient load distribution can be achieved. In particular, when a user uses a multi-window environment, a plurality of window applications are simultaneously executed, and this load balancing control is effective.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a system configuration diagram of this load balancing control method. The computer 141 grasps the communication means 131 capable of communicating with all the computers on the network and the load state of each computer, exchanges the information with other computers using the communication means 131, and exchanges the information. Load information management means 13 accumulated in the load table 139 of the external storage device
2, an input / output device 133 that receives an input from a user and passes it to a window server process (hereinafter referred to as a server process), and an input / output device 133 or an input / output device of another computer through the server process. A window application 135 and a server process 136 that manages the input / output of the window application 135 and exchanges the input received from the input / output device 133 and the output information received from the window application by using another computer and the communication unit 131. ing. The input / output device 133 provides a user interface such as a display screen and a keyboard.

FIG. 2 shows a state in which a plurality of computers 41 having the configuration shown in FIG. 1 are connected by a network 42.

As shown in FIG. 3, the load table stores information obtained from the load information management means of all computers connected to the network. The name of each computer and the current load are stored in the table.

In FIG. 1, the load information management means 132 of the computer 141 periodically checks the current load information of the computer 141 and writes it in the load table 139. Further, the load information of the computer 141 is transmitted to the load information management means of all the computers on the network using the communication means 131. Further, the information sent from the load management information means of another computer through the communication means 131 is stored in the load table 139. There are various definitions of computer load,
Here, the load of the computer is considered to be the CPU usage rate, and the CPU usage rate per unit time is represented by a decimal number between 0 and 1. In a computer, the load is 0
When is 1, the CPU is not used at all, and when 1 is the CPU
Indicates that it is always used.

Next, referring to FIG. 4, it will be explained how the user 72 starts the server process and the window application, and then how the server process uses the history file. .

The user 72 first activates the server process 66 on the computer 60. Next, the window application 65 is activated. Conventionally, when the user starts the window application 65, the computer to be started and the application name are specified, and this is specified by the server process 66.
And the window application was running. In the present invention, the user does not have to specify the computer name in order to distribute the load to the computers on the network as described later. This is because the server process 66 automatically selects the computer that starts the window application 65 in order to distribute the load on the computers on the network. Therefore, when the user 72 activates the window application (cl1), the part which has conventionally used the computer name is replaced with the keyword “anonymous” and the command “anonymous cl1 parameters” is sent to the server process 66 through the input / output device 63. However, if it is desired to avoid automatic selection of a computer to be executed by the server process 66, a computer name may be designated, but in this case, load distribution will not be performed. Also, the parameter
The window application 6 is specified by rs.
These are various parameters passed when 5 is executed.
When the user 72 activates the window application 65, a window display position and a value indicating the size of the window are designated as parameters. In each window application, standard values of parameters such as position and size are set, and if not specified, the standard values are used.

Next, the load balancing method in the present invention will be described.

As already explained by the user 72, a command for starting the window application 65 is sent to the server process 66 of the computer 60 through the input / output device 63, and how the server process balances the load will be described. The window application that the user is trying to start is cl1. As the computer for starting cl1, the computer with the smallest load in the load table 67 is selected. Here, it is host2 (computer 71). The server process 66 activates the window application 69 of host2, and its input / output is performed through the input / output device 63 of the computer 60.

Thus, conventionally, even if the load on the computer on the network fluctuates, the load cannot be distributed because the window application is statically specified, but the load on the network is small. The load can be distributed to computers. In addition, the user does not need to check the load status and decide the computer to start.

However, this embodiment has the following problems.

There is a computer 1 and a computer 2, and it is assumed that the computer 1 has a higher current load as shown in the graphs of FIGS. In the graph, the horizontal axis represents time and the vertical axis represents the load on the computer. It is assumed that time 0 on the horizontal axis indicates the present. The application A running on the computer 1 is heavily loaded, but it ends shortly, and the load on the computer 1 thereafter becomes very small. On the other hand, it is assumed that the application B running on the computer 2 is expected to take a long time to execute although the load is small. Here, it is assumed that the application C is to be newly executed by the computer 1 or 2. When the computer that executes the application C only with the current load is determined, the computer 2 with the light load is selected. However, if it takes a long time to execute the application C, it is expected that the load will be distributed more when the application C is executed by the computer A as shown in the graph.

Therefore, a second embodiment will be described below in which an expected history file of the load of a computer is used to carry out efficient load distribution so as to solve this problem.

FIG. 10 is a system configuration diagram of the load balancing control system of the second embodiment.

The computer 21 grasps the load state of each computer and the communication means 11 capable of communicating with all the computers on the network, and exchanges the information with other computers by using the communication means. Information is stored in the load table 19 of the external storage device, and the load information management unit 12 that manages the process table 14 that manages the CPU time information of the window application 15 executed on the computer 21 and the past of the user 17 are stored. A history file 18 that holds a history of application usage, and an input / output device 1 that receives input from a user and passes it to a window server process (hereinafter referred to as a server process)
3, a window application 15 for displaying on the input / output device 14 or the input / output device of another computer through a server process, and management of the input / output of the window application, and the input / output device 13 by using another computer communication means 11. It has a server process 16 for exchanging the input received by and the output information received from the window application.

Similar to the load table of the first embodiment, the load table 19 stores information obtained from the load information management means of all computers connected to the network. The table 19 stores the name of each computer and the current load.

The load information management means 12 of the computer 21 periodically checks the current load information of the computer 21 and writes it in the load table 19. Further, the load information of the computer 21 is transmitted to the load information management means of all the computers on the network using the communication means 11. In addition, the information sent from the load management information means of another computer through the communication means 11
It is stored in the load table 19. There are various definitions for the load of the computer, but here, the load of the computer is the CPU
The CPU usage rate per unit time is represented by a decimal number between 0 and 1. In a certain computer, when the load is 0, the CPU is not used at all, and when the load is 1, the CPU is always used.

The process table 14 records the CPU time used by the window application 15 currently executed on the computer 21 for each process of the window application. Load management information means 12
Assigns a unique number to the window application currently being executed (this makes it possible to determine even if the same application is being executed more than once), periodically checks the CPU time used by each, and uses the process table 14 Write in. The contents of the process table 14 are shown in FIG.

A history file 1 for each user
There are eight. Using FIG. 10, the history file 18
Will be described. In the history file, the window application name 51, the size 52 and the position 53 displayed on the display device at the last execution, the computer name 54 started last time, the statistical information of the last execution (start time 55, end time 56, The CPU usage time 57) is recorded for each window application. What is CPU time 57?
CP will be launched by the time the application is started
Represents the time used U.

Next, referring to FIG. 8, a description will be given of how the user 192 starts the server process 186 and the window application 185, and then how the server process 186 uses the window history file 188. I will explain.

The user 192 first activates the server process 186 on the computer 180. Next, the window application 185 is activated. Conventionally, when a user starts a window application, a computer to be started and an application name are specified, and this is sent to the server process 186, and the window application 185 is executed. In the second embodiment, the user does not have to specify the computer name in order to distribute the load to the computers on the network as will be described later. This is because the server process 186 automatically selects the computer that starts the window application 185 in order to distribute the load on the computers on the network. Therefore,
User 192 is a window application (cl1)
When you start the
The keyword "nymous" is replaced and a command "anonymous cl1 parameters" is sent to the server process 186 through the input / output device 183. However, if it is desired to avoid automatic selection of the computer in which the server process is executed, the computer name may be specified, but in this case the load will not be distributed.

The parameters specified by parameters are various parameters passed when the window application is executed. When the user 192 starts the window application, the start-up parameters that specify its position and size are used.
In each window application, standard values of parameters such as position and size are set, and if not specified, the standard values are used.

Input is sent to the server process 186 to launch the window application 185 to perform load balancing. The server process 186 searches the history file 188 for an entry having a parameter data that matches the startup parameter of the window application 185 and the application name. If there is a matching entry, the name of the computer on which the application was last executed, the startup time, the end time, and the CPU time can be obtained from that entry as information on the previous execution. Based on this, as will be described later, the server process 186 determines a computer on the network that executes the window application 185 to perform load distribution.

In the second embodiment, when the window is activated, the user's history file entry is searched. As a search method at this time, the displayed position and size of the window are supposed to match exactly,
If there is no exact match, you can choose the closest match.

The following is a description of how the server process stores information in the history file 188.

Information is stored in the history file 188 when the window application is terminated. Therefore, in FIG. 8, the user 192 terminates the execution of the window application 189 which is being executed by the computer 191. Think about the case. The server process 186 is a window application 189.
Since it manages the position and size of, it first writes this information to a new entry in the history file. next,
How to write start time and end time of window application. Server process 186
Checks the CPU time, and obtains the value from the communication means 193, 18
Notify the server process 186 using 1. The information obtained by this is written in the entry of the history file. With the above, all necessary information is written in the entry.

Next, the load distribution method based on the load prediction of each computer on the network according to the present invention will be described with reference to FIG.

The server process 186 can calculate the number of window applications (scheduled startup number) to be executed by the computer on the network at an arbitrary time from the information on the startup time of the history files of all users. Furthermore, from the entry of each window application, the expected value of the load of each computer at any time can be calculated by the following formula. However, here, it is assumed that all users launch approximately the same number and types of applications every day, and when multiple window applications are used, the users may have different usage patterns depending on the location of the window applications. I'm assuming.

The expected CPU usage time L (T) of the computer A at a certain time is the CPU usage time C (for the window application Y, which is expected to be executed by the user X at the time T on the computer A. T, X,
Y) can be calculated as follows. However, C (T,
X, Y) is the value of the CPU time recorded in the history file of the user X when the fact that the user X used the application Y at the time T is recorded in the history file of the user X. Time shall be 0.

[0045] L (T) indicates that a user X has a computer A at time T.
The CPU usage time per unit time is calculated during the execution of the application Y executed in 1., and it is totaled by all users and applications. That is, it indicates the expected value of the CPU usage time per unit time.

Further, the average Av (t) of the predicted CPU usage time of the computer A after t hours from the present is calculated by the following formula. Here, k represents the dragon degree of the time interval for obtaining the average. Av (t) takes a value that reflects the fluctuation of the load on the computer A as k decreases.

Av (t) = 1 / k.SIGMA.x _{= 1, k} L (N + t.x / k) This is the CPU usage per unit time of the window application predicted to be executed by the computer A. Shows the cumulative total of time. Further, the expected average load (LDAV (g) of the computer A can be obtained from the load equation using AV (t). Here, the current load of the computer A is set to ld, and α
Is a constant.

The larger the value of α, the more the current load is reflected in the expected average load LDAV (t). On the other hand, if the value is small, the influence of the current load is small. Although α is a constant here, it may be a function inversely proportional to the length of time t. If a function inversely proportional to time is used, the current load value becomes LD when the time interval t increases.
It does not affect AV (g), and when t is small, the effect becomes large.

As the user 192 has already described, input / output the command to start the window application 1
A description will be given of how the load is distributed to the server process 186 of the computer 1 through 83 and how the server process performs load distribution. The window application the user is trying to launch is cl1, and the parameter is size 100 ×
50 and positions 10 and 10. The server process 186 searches the entry in the window history file for a match with the parameter of the received command.
Here, the first entry in the history file 188 matches.

From the information in the first entry, the application use time T1 at the previous execution (here, 5 hours 23
Minutes) is required. As a result, the expected average load LDAV
Calculate (T1) for all computers. The computer with the smallest expected average load LDAV (T1) calculated is selected as the computer expected to have the least load while using the window application cl1.
(Here, it is assumed to be host2 (191)). The server process 186 activates the window application 189 of the host2 (191), and its input / output is the computer 1
(180) through the input / output device.

There are other ways to calculate the expected load than those listed above. Here, in order to determine the computer that starts the application, the average value is calculated for each computer based on the contents of the history file.
However, a value integrated by the time t may be used instead of the average value.

Further, although the load is calculated only by the CPU usage time, the computer resource is not limited to that, and I
There are various things such as / O peripherals and memory, and statistical information (I / O frequency per unit time, memory usage rate, etc.) can be collected for them. This information is also
As in the case of writing to the history file and calculating the expected load, the expected utilization rate of resources of each computer can be obtained. At the same time, the resource usage rate of the window application which is currently activated and used can also be obtained from the history file, and it is possible to compare which computer can supply more resources frequently used by the window application to be activated. From the expected resource usage rate of each computer, the computer expected to be able to supply the most resources can be determined and the load can be distributed.

Thus, conventionally, even if the load on the computer on the network fluctuates, the load cannot be distributed because the window application is static, but the load on the computer on the network is small. The execution of the window application can be well distributed to the computer, and the load can be distributed to the computers on the network. In addition, the user does not need to check the load status and decide the computer to start.

Further, even if the current load of a computer is very large, it is possible to detect a computer whose load is expected to be reduced after a short time by using the predicted load information. By executing the program on the selected computer in this way, a computer with an average load is selected throughout the user's use of the window application, compared with the case where the load is distributed only by the current load information. It will be.

The computer in this embodiment is realized by a host computer, a workstation or the like to which a terminal is connected.

[0056]

As described above, according to the present invention, a computer with a low load can be determined from the prediction of the load status and the usage status, and efficient load distribution can be performed without the user's hand. .

[Brief description of drawings]

FIG. 1 is a diagram showing the configuration of a computer system according to a first embodiment of the present invention.

FIG. 2 is a diagram showing how the computers of the first embodiment are connected.

FIG. 3 is a diagram showing an example of a load table provided in each computer of the first embodiment.

FIG. 4 is a diagram for explaining the operation of the first embodiment.

FIG. 5 is a diagram showing an example of load transition of the first computer according to the first embodiment.

FIG. 6 is a diagram showing an example of load transition of the second computer according to the first embodiment.

FIG. 7 is a diagram showing the configuration of a computer system according to a second embodiment of the present invention.

FIG. 8 is a diagram for explaining the operation of the computer system according to the second embodiment.

FIG. 9 is a diagram showing an example of a process management table provided in each computer of the second embodiment.

FIG. 10 is a diagram showing an example of a history file provided in each computer of the second embodiment.

FIG. 11 is a diagram showing a system configuration of a conventional computer system.

FIG. 12 is a diagram showing a first operation state of a conventional computer system.

FIG. 13 is a diagram showing a second operation state of a conventional computer system.

FIG. 14 is a diagram showing a third operation state of a conventional computer system.

[Explanation of symbols]

131 ... Communication means, 132 ... Load information management means, 133
... I / O device, 135 ... Window application,
136 ... Server process, 137 ... User, 139 ... Load table, 141, 142 ... Calculator.

Claims (2)

[Claims] 1. A computer system comprising a plurality of computers connected via a network, a communication means for communicating with all the computers on the network, and managing the load status of the computers to communicate the load information. Load information management means for transmitting and receiving to and from another computer using the means, storing and managing the same, a window server means for simultaneously displaying a plurality of windows on the display device of the computer and managing the input and output thereof, the load information And a means for determining a computer that executes a window application according to load information managed by the management means. 2. In a computer system composed of a plurality of computers connected via a network, a plurality of windows can be simultaneously displayed on a display device of the computer, and a window server means for managing the input / output thereof and a window application History management means that stores the history of the load on the applied computer, load information management means that records the load of the computer and sends information to other computers, and information sent and received by the server means or load information management means of each computer A computer for exchanging data, and a means for determining a computer for executing a window application by associating a history with a window by the history management means and the history management means. System load balancing control method.