JP3897843B2 - Network load balancing method in data transfer of client server system - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a method for maintaining the soundness and efficiency of a client server system by distributing a load concentrated on the server and the network when a large number of client computers access the server simultaneously in the client server system.
[0002]
[Prior art]
Recently, a client server system has been widely used in place of an information processing system including a conventional host computer and a terminal.
[0003]
The client server system includes a “server” that manages and provides data and software in an integrated manner, and a “client computer” that makes a request to the server and uses the data and software.
[0004]
These servers and client computers perform information processing independently in each device, but are configured to perform unified information processing as a whole by exchanging processing results and processing data with other devices.
[0005]
In this way, the client server system performs unified information processing while the server and client computers in the system exchange data and software with each other, so that each client computer in the system naturally has the same software (in the same software). Must be the same version).
[0006]
For this reason, in the current client / server system, generally all client computers access the server at regular intervals (for example, once or twice a week), and the software and data stored in the server at that time are stored. Updates its own file with a file.
[0007]
FIG. 6 shows a system configuration and method for the client computer to access the server and receive distribution of software and data. In FIG. 6, the flow of instructions and information is indicated by arrows.
[0008]
As shown in FIG. 6, a client server system 1 includes a server 2 and a plurality of client computers 3 connected via a network 4 such as a LAN.
[0009]
When it is time to update the file, many client computers 3 access the server 2 all at once and try to download the file. In the conventional download, after the client computer 3 accesses the server 2 and a connection procedure such as confirmation of the communication partner is established, a command such as file transfer is sent from the client computer 3 to the server 2.
[0010]
Transfer commands from these client computers 3 are processed by the server CPU 2 a of the server 2. In this case, the CPU 2a of the server controls the I / O device 2b, takes out the designated file from the storage device 2c and transfers it to the client computer 3 by the action of the server I / O device 2b.
[0011]
In the client computer 3, the I / O device 3b under the control of the CPU 3a stores the file transferred from the server 2 in the storage device 3c, and updates the old file if it exists.
[0012]
The conventional client server system 1 includes a flow rate control program 5 so that the processing of the server CPU 2a does not overflow. This is because the CPU 2a of the server has a limit in its processing capability, and if more commands from the client computer 3 are input, the processing becomes impossible and the entire system stops functioning. For this reason, when a predetermined number of client computers 3 are connected, the flow rate control program 5 denies access to the requested client computer 3 for further connection requests.
[0013]
FIG. 6 shows a state where the server 2 is performing a full process. In FIG. 6, the number of client computers to which the server 2 can be connected simultaneously is assumed to be three. In this case, the client computers C1 to C3 are in a communication state, but the other client computers C4 to Cn are in a standby state (WAIT state) and keep trying to access the server 2.
[0014]
When any one of the client computers C1 to C3 finishes communication with the server 2, any one of the client computers C4 to Cn that attempted to access immediately after that is connected.
[0015]
[Problems to be solved by the invention]
However, the conventional client-server system and its file distribution method cannot efficiently utilize the capacity of the entire system due to the limitation of the processing capacity of the CPU of the server.
[0016]
In other words, in the conventional client-server system, as described above, the CPU of the server intensively processes access requests and file transfer instructions from the client computer, so that the upper limit of the processing capacity is reached immediately. .
[0017]
Due to the limitation of the processing capability of the CPU of the server, the file transfer was eventually performed to a limited number of client computers. On the other hand, the CPU of the client computer during this period uses only a part of its processing capacity and is in a state of playing.
[0018]
In addition, since the client computer in the WAIT state repeatedly tries to access the server, a load is imposed on the CPU and the network on the server side in order to send a response to the access request (response to reject access).
[0019]
Therefore, the present invention solves the above-mentioned problems of the prior art, eliminates the concentration of the load on the CPU of the server, and distributes a large number of files from the server to a large number of client computers simultaneously by the processing capability of the entire client-server system. It is an object of the present invention to provide a network load distribution method in data transfer of a client server system that can be performed.
[0020]
[Means for Solving the Problems]
The network load distribution method in the data transfer of the client server system according to claim 1 of the present application is:
The network load control program with each client computer client-server system, data is transferred by directly controlling the server of the I / O device, the network load control program on each client computer, the network load from load time data The data block size to be transferred at one time and the waiting time during one data transfer are controlled according to the estimation result of the network load.
[0021]
The network load distribution method in data transfer of the client server system according to claim 2 of the present application is the method according to claim 1,
In a client-server system including client computers with different processing speeds,
A time obtained by subtracting the data transfer time of each client computer from the reference data transfer time based on the data transfer time when a client computer having a predetermined processing speed receives data transfer using the maximum bandwidth of the network. In the positive case, the time difference is set as the time interval of each data transfer operation.
[0022]
The medium according to claim 3 of the present application is:
Each client computer of the client server system directly controls the server I / O device to transfer data,
Each client computer estimates the network load from the data read time, and controls the data block size to be transferred at one time and the waiting time during one data transfer according to the estimation result of the network load. This is a record of the program.
[0023]
DETAILED DESCRIPTION OF THE INVENTION
Next, embodiments of the present invention will be described below.
The “network load distribution method in data transfer of a client server system” according to the present invention includes a load sharing by a CPU of a client computer, a file distribution simultaneously to a large number of client computers, and a file transfer at an efficient transfer rate of a network. Realize three. This will be described below.
[0024]
FIG. 1 shows the system configuration of a client server system that implements the method of the present invention and the state of its processing.
[0025]
In FIG. 1, the client server system 1 includes a server 2, a client computer 3, and a network 4 such as a LAN that connects the server 2 and the client computer 3.
[0026]
This client server system 1 is different from the conventional client server system (see FIG. 6) in that there is no flow control program on the server 2 side, and the client computer 3 has a network load control program 6, and this network load control The program 6 directly controls the server I / O device 2b.
[0027]
As shown in FIG. 1, in this client server system 1, a large number of client computers 3 control the server I / O device 2b by the CPU 3a, and the server I / O device 2b is stored in the server storage device 2c. A predetermined file is taken out and transferred to the client computer 3. In this case, the CPU 2a of the server is only slightly involved in the control of the I / O device 2b and is released from most loads.
[0028]
On the other hand, the file transferred to the client computer 3 is stored in the storage device 3c by the I / O device 3b of the client computer.
[0029]
That is, according to the client server system 1, the network load control program 6 takes over the processing performed by the CPU 2a of the conventional server by the CPU 3a of the client computer and distributes the load on the CPU 2a of the server.
[0030]
In the client server system 1, the network load control program 6 controls the load on the network 4 simultaneously with the file transfer to all the client computers 3 that have requested connection. This will be described below.
[0031]
FIG. 2 shows a communication state in which the client server system 1 is transferring files to a large number of client computers 3 at the same time.
[0032]
The shaded bar chart portion of the figure indicates the time during which data is transferred. As shown in the figure, the network load control program 6 of this system, when a small number of client computers 3 try to access the server 2 (referred to as “normal”), Similarly, after transferring a file to one client computer, the file is transferred to the next client computer.
[0033]
However, when a large number of client computers 3 try to access the server 2 at the same time (referred to as “rush time”), the network load control program 6 of each client computer 3 calculates the load on the network 4. However, the transfer file is divided, only a small amount of data is transferred at one time, and the time for receiving the transfer is controlled so as not to overload the network 4 as a whole.
[0034]
That is, in this system, the network load control program 6 of each client computer 3 that has accessed the server 2 calculates the load state of the network 4, divides the transfer file based on the calculation result, and performs one READ time (TR ) The WAIT time (TW) in one data transfer and the WAIT time (TS) between each data transfer are changed.
[0035]
By finely dividing the transfer file, data can be alternately transferred to a large number of client computers 3, and data can be apparently transferred to a large number of client computers 3 simultaneously.
[0036]
In addition, by dividing the transfer file in this way and adjusting the interval between the transfer times and the length of one transfer time, data transfer can be performed at the most efficient transfer rate of the network 4. .
[0037]
In general, the file is subdivided so that the access requests to the server 2 are crowded, and the interval between data transfers is increased.
[0038]
Here, the most efficient transfer rate of the network 4 will be described.
[0039]
In general, the network 4 has a unique information transfer capacity (referred to as bandwidth).
[0040]
However, at the limit of the bandwidth limit, the number of times of re-transmission or the like is increased due to signal interference, and the overall communication efficiency is lowered. It is known that this communication efficiency decreases exponentially when it exceeds a certain ratio with respect to the entire bandwidth.
[0041]
On the other hand, when data is transferred at a low transfer rate with respect to the bandwidth, retransmission due to signal interference is eliminated, but the network 4 is idle.
[0042]
In the network load distribution method according to the present invention, the size of data to be transferred at one time (referred to as a data transfer unit) and the time for receiving the transfer are changed according to the load state of the network 4 that changes every moment. By appropriately distributing the data transfer, the file transfer is performed at the most efficient transfer rate of the network 4.
[0043]
Below, the process of the network load control program 6 for implement | achieving the said function is demonstrated.
[0044]
FIG. 3 shows the flow of processing by the network load control program 6. As shown in FIG. 3, when the client computer 3 tries to access the server 2, the network load control program 6 is activated (step S <b> 100), and a setting value is first acquired from the server 2. Here, the set value refers to the retry interval, the number of retries, the first data transfer unit, and the standard one-time data transfer time for efficient data transfer in the network 4 (this is the standard data transfer time. Etc.).
[0045]
The setting value is stored in the server 2 and acquired by the activated network load control program 6 in consideration of the ease of change when changing the setting value. You may make it store in the computer 3. FIG. In this case, this step S110 is omitted.
[0046]
Next, the network load control program 6 executes file transfer (step S120), controls the data block size / waiting time in one data transfer (step S130), and controls each data transfer interval (step S140). The file transfer is executed while performing the above operation for each data transfer.
[0047]
Details of the control of the data block size and waiting time in step S130 will be described in detail with reference to FIG.
[0048]
As shown in FIG. 4, first, the first read WAIT time TW = T1 and the read data block size BS = B1 are acquired (step S110).
[0049]
Next, data READ, WAIT, and WRITE are repeated n times based on T1 and B1 (steps S131 to S133). During the n data transfers, the READ time TR of each data transfer is measured.
[0050]
Next, the data block size BS transferred at one time and the WAIT time TW in one data transfer are calculated and controlled by the n READ times TR (step S134). The process in step S134 will be described in further detail below.
[0051]
First, an average of n times of TR time is obtained. TA = ΣTR / n where the average of TR is TA.
[0052]
This TA represents the load state of the network 4. That is, if TA is long, it means that it takes time to read data (READ), which indicates that the network 4 is in a rush state.
[0053]
Therefore, when TA> Tmax (Tmax is a set value), it is determined that the network 4 is in a high load state, and the data block size BS is reduced to the current ½.
[0054]
When the data block size BS becomes smaller than the predetermined value BL (lower limit of the data block size) as a result of reducing the data block size BS, the WAIT time is doubled without further reducing the data block size.
[0055]
In other words, when the network 4 is in a congested state by the above processing, the data block size becomes smaller or the WAIT time becomes longer, eventually the load on the network 4 is temporally distributed, and an efficient transfer rate is achieved. It will calm down.
[0056]
Contrary to the above, when TA <Tmin (Tmin is a set value), it is determined that the network 4 is in a low load state, and the WAIT time TW is reduced to the current half or the data block size. Double the BS.
[0057]
In other words, when the network 4 is in a low load state, the WAIT time becomes shorter or the data block size becomes larger, and eventually the load on the network 4 is concentrated in time and settles to an efficient transfer rate. Become.
[0058]
However, since the above process is a process in each client computer 3, it cannot be adjusted due to a difference in processing speed between the client computers 3.
[0059]
That is, when the processing speeds of the client computers 3 connected to the server 2 are the same, no problem occurs. However, when there is a difference in processing speed between the client computers 3, the client computer 3 having a fast processing speed is connected to the network 4. Judge that the load is low and concentrate data transfer more and more. On the other hand, the client computer 3 having a low processing speed distributes data transfer more and more because the READ time TR is long.
[0060]
Therefore, the data transfer interval of each client computer 3 is controlled in step S140 (see FIG. 3). The control of the data transfer interval will be described below with reference to FIG.
[0061]
First, the number of bytes BT that can be transferred per second of the standard client computer 3 is calculated from the bandwidth of the network 4 (step S141).
[0062]
Next, the data READ, WAIT, and WRITE processes are repeated n times, and the total time TTsum and the data transfer amount BSsum for n times are obtained (steps S142 to S144). Note that the time for performing one READ, WAIT, and WRITE processing is TT, and the amount of data transferred by one READ, WAIT, and WRITE is BS.
[0063]
Next, the data transfer interval is controlled by the TTsum and BSsum (steps S145 and S146).
[0064]
In step S145, when the standard client computer 3 first uses the maximum bandwidth, the time TW2 required to transfer BSsum is calculated (TW2 = BSsum * 1000 / BT).
[0065]
Next, TS is calculated as TS = TW2-TTsum (step S145). In a computer with high processing speed, TTsum becomes short and TS becomes long. On the other hand, in a computer with slow processing, TTsum becomes long and TS becomes short.
[0066]
The calculated TS is used as a time between data transfers so that a time TS is opened between the data transfers (step S146). When TW2 <TTsum, each data transfer interval is set to zero.
[0067]
With the above processing, the client computer 3 that is fast in processing transfers a large amount of data within a short time by one data transfer, but the data transfer interval becomes long. On the other hand, the slow-processing client computer 3 transfers a small amount of data within a long time by one data transfer, but the data transfer interval is shortened.
[0068]
Due to the above balance, even when client computers 3 having different processing speeds are connected to the server 2, data can be evenly distributed from the server 2 to the entire client computers 3.
[0069]
【The invention's effect】
As is apparent from the above description, according to the “network load distribution method in data transfer of a client / server system” of the present invention, each client computer directly controls the server I / O device, so that The load can be distributed by the CPU of each client computer.
[0070]
Also, in the method of the present invention, the network load control program of each client computer determines the network congestion status based on the read time of one data transfer, and when the network is congested, the data block to be transferred Reduce the size or increase the waiting time in one data transfer. Conversely, when the network is not congested, the data block size to be transferred is increased, or the waiting time in one data transfer is shortened.
[0071]
By setting the above-mentioned transfer data block size and the upper and lower limits of the waiting time so as to match the highly efficient transfer rate of the network, each client computer autonomously adjusts the data transfer interval and transfer data block size. As a result, it is possible to receive data transfer from the server at a highly efficient transfer rate of the network.
[0072]
Further, by controlling the data transfer interval, a client computer having a high processing speed opens the interval, and conversely, a client computer having a low processing speed receives the data transfer at a close interval. As a result, data is evenly distributed throughout the client server system.
[Brief description of the drawings]
FIG. 1 is a diagram showing a means for realizing a “network load distribution method in data transfer of a client server system” according to the present invention and a flow of processing between those means.
FIG. 2 is a diagram showing a state of data transfer according to a “network load distribution method in data transfer of a client server system” according to the present invention.
FIG. 3 is a flowchart showing a processing flow of a network load control program of “network load distribution method in data transfer of client server system” according to the present invention.
FIG. 4 is a flowchart showing a method for controlling the data block size and waiting time in one data transfer according to the present invention.
FIG. 5 is a flowchart showing a data transfer interval control method according to the present invention.
FIG. 6 is a diagram showing a data transfer means of a conventional client-server system and a flow of processing between those means.
[Explanation of symbols]
1 Client server system 2 Server 2a Server CPU
2b Server I / O device 2c Server storage device 3 Client computer 3a Client computer CPU
3b Client computer I / O device 3c Client computer storage device 4 Network 5 Flow control program 6 Network load control program

Claims (3)

The network load control program with each client computer client-server system, data is transferred by directly controlling the server of the I / O device, the network load control program on each client computer, the network load from load time data The data of the client server system is characterized in that the data block size received at one time and the waiting time during one data transfer are controlled according to the estimation result of the network load. Network load balancing method for forwarding. In a client-server system including client computers with different processing speeds,
A time obtained by subtracting the data transfer time of each client computer from the reference data transfer time based on the data transfer time when a client computer having a predetermined processing speed receives data transfer using the maximum bandwidth of the network. 2. The network load distribution method in data transfer of a client server system according to claim 1, wherein the time difference is set as a time interval of each data transfer operation when positive . Each client computer of the client server system directly controls the server I / O device to transfer data,
Each client computer estimates the network load from the data read time, and controls the data block size to be transferred at one time and the waiting time during one data transfer according to the estimation result of the network load. A medium on which the program is recorded.

Images