JP2008059040A - Load control system and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To process a significant request on application with priority when a server is congested. <P>SOLUTION: The priority control is conducted depending on the class of a request in an application layer of a server. When the server receives a request through TCP/IP, before the request is next processed by Web application, the request is analyzed. The request determined to be significant is first processed with priority by the Web application. When the server is congested and all requests are not processed, a request with the least significance is refused. Further, when the overload of the server is detected, TCP/IP layer is controlled client by client to regulate packet receiving. At this time, it is estimated based on the request analysis result whether or not the priority of future request transmitted by the client is high. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a packet control technique when a server that receives a request from a client and returns a response is overloaded. In particular, the present invention relates to a packet control technique for performing priority control in units of clients based on request priority prediction. In this specification, the description will be given focusing on the Web server, but the application of the present invention to other servers is not necessarily limited.

  With the spread of the Internet, various services can be used via a network. Examples are mail, homepage browsing, search, online transactions, IP phone calls, video on demand, and the like. Although these network services can be provided in various forms, in recent years, the use of a Web server has become the mainstream as an interface with a client.

  The basic mechanism of a service using a Web server (Web service) is as follows. First, a client transmits a request to which a URL (Uniform Resource Locator) for identifying content to be acquired is added to a Web server. When the Web server receives the request, the content for the URL in the request is sent back to the client as a response. The Web service is provided by repeating this request-response.

  In this specification, the entire server system that performs a Web service is referred to as a Web server, and a function for generating content according to a request on the Web server is referred to as a Web application. In the Web service, TCP / IP is used as a transport layer / network layer protocol for transmitting and receiving requests and responses.

  An issue of Web services is server load control when requests are concentrated. In particular, the importance of Web services differs depending on the content of the request. Therefore, it is necessary to preferentially process important requests when the server is busy.

  For example, a web service may require a large number of requests in order to display one page on a client browser. Repeating a request for displaying one page is referred to as page processing in this specification. At this time, when the server is congested, it is desirable to de-prioritize the first page acquisition request, and the already started page processing needs to be protected.

  The basic procedure for page processing is as follows. First, the client inputs the URL of a resource (hereinafter referred to as a page root resource) that is the root of a page to be acquired to the browser. Next, the browser transmits a request to the Web server based on the input URL, and acquires a page root resource. At this time, the URL of another resource necessary for page display is indicated in the page root resource. Next, the browser automatically issues a request for the indicated URL. The above is recursively repeated until all resources necessary for page display are acquired.

  When page processing is requested from a large number of clients, a request during page processing fails due to resource contention on the server, causing a situation in which the page cannot be completely displayed on any client. In order to solve this problem, it is necessary to prioritize a new page processing request and protect the already started page processing.

  In addition, in a Web service, one service may be completed for the first time by browsing and inputting information across a plurality of pages. For example, in online shopping, a purchase procedure is completed for the first time by selecting a product to be purchased, inputting client information, and finally confirming purchase contents.

  In this specification, in a service that requires a plurality of pages to complete, a session from when the client acquires the first page until the last page is acquired is referred to as a session. The session is used when performing important processing such as dealing with gold goods or updating personal information.

JP 2002-16633 A

  In the session processing described above, there arises a problem that session processing is hardly completed when the server is congested. This problem is caused by an increase in the number of sessions that are processed in parallel on the server, resulting in an increase in the number of sessions that fail midway due to contention of server resources among the sessions. It is therefore necessary to de-prioritize new session processing requests and protect already started session processing.

  For the purpose of solving server congestion, a method of discarding packets in units of source IP addresses according to the CPU usage rate has already been proposed (for example, see Patent Document 1). However, with this method, since the priority in the Web application is not considered, the above-described page processing and session processing cannot be protected.

  It is an object of the present invention to preferentially process an important request on an application when the server is congested.

  In the present invention, first, in the request priority control means, priority control according to the type of request is performed in the application layer of the server. When the server receives the request via TCP / IP, the request is then analyzed before the request is processed by the Web application. Then, the Web application preferentially processes requests determined to be important, for example, requests belonging to ongoing page session processing. If the server is too busy to process all requests, it rejects requests with low importance.

  Furthermore, in the present invention, TCP / IP is controlled using request priority prediction so that the request priority control means itself does not become overloaded. That is, the calculation cost required for request analysis and importance determination by the request priority control means is large, and request reception processing such as TCP / IP processing is involved when receiving a request. Becomes overloaded and important requests are missed. Therefore, in the present invention, when an overload of the server is detected, the load on the request priority control unit is reduced by controlling the TCP / IP layer for each client and regulating packet reception.

  At this time, the priority prediction means predicts whether or not the priority of the request that the client will transmit in the future is high based on the request analysis result in the request priority control means. The predicted priority is reflected in the priority control in the TCP / IP layer. Thus, a request that is highly likely to be prioritized in the application layer can be processed by the Web application without being regulated by the TCP / IP layer.

  The TCP / IP layer performs priority control based on the IP address. For this reason, the priority control in the TCP / IP layer may not accurately reflect the priority control in the request unit in the application layer. However, it does not impede the purpose of protecting important requests by reducing the load of request priority control. For example, if different clients are transferring requests via the same proxy, the source IP address of the packet received by the server is all the IP address of the proxy. Therefore, when a plurality of clients are connected via a proxy, priority control cannot be performed for each individual client. However, when the server is congested, it can be predicted that there are many source IP addresses other than the proxy. Therefore, a sufficient load reduction effect can be obtained by restricting traffic from other source IP addresses. When all clients connected to the proxy are de-prioritized, the request from the proxy may be de-prioritized for the first time.

  That is, the present invention is a load control system used for a server that receives a request from a client and sends back a response. The feature of the present invention is that the request priority control means performs processing that requires a plurality of requests. Based on the progress, a means for setting the priority for each IP address of the request source client, a means for determining whether or not the server load exceeds the threshold, and when the threshold is exceeded, the priority is set to a low priority. And means for restricting reception of IP packets from the IP address.

  Alternatively, the request priority control means may be provided with means for setting a priority for each IP address of the request source client based on the user identification information stored in the request.

  The means for restricting the reception of the IP packet restricts the reception of the IP packet by, for example, limiting the rate of the TCP / SYN packet among the IP packets using the TCP / IP protocol. Alternatively, reception of IP packets is restricted by reducing the window size of the TCP header in the TCP-ACK packet transmitted from the server system using the TCP / IP protocol. Alternatively, reception of IP packets is restricted by cutting the TCP connection using the TCP / IP protocol.

  Further, when the server load exceeds a threshold value, the IP packet reception limit can be gradually increased, and when the server load becomes equal to or less than the threshold value, the IP packet reception limit can be gradually decreased.

  Further, from the viewpoint of the load control method, the present invention is a load control method used for a server that receives a request from a client and sends back a response. Based on the progress of the processing that requires the request, set the priority for each IP address of the request source client, determine whether the server load exceeds the threshold, and set the priority lower when the threshold is exceeded The IP packet reception from the specified IP address is restricted.

  Alternatively, the priority may be set for each IP address of the request source client based on the user identification information stored in the request.

  When restricting the reception of the IP packet, for example, the TCP / IP protocol is used to restrict the reception of the IP packet by limiting the rate of the TCP / SYN packet among the IP packets. Alternatively, reception of IP packets is restricted by reducing the window size of the TCP header in the TCP-ACK packet transmitted from the server system using the TCP / IP protocol. Alternatively, reception of IP packets is restricted by cutting the TCP connection using the TCP / IP protocol.

  Further, when the server load exceeds a threshold value, the IP packet reception limit can be gradually increased, and when the server load becomes equal to or less than the threshold value, the IP packet reception limit can be gradually decreased.

  Based on the present invention, the load required for request priority control when the server is overloaded can be reduced.

(Matters common to the first to third embodiments)
FIG. 1 shows a block configuration of the load control system of the present embodiment. The present embodiment includes clients 1-1 to 1-n and a server 3 with a load control function. Hereinafter, when simply referred to as a server, the server 3 with load control function is indicated. The server is connected to one or more clients 1-1 to 1-n (hereinafter, reference numerals 1-1 to 1-n are omitted) via a network. When the server receives the request, it returns appropriate content as a response. The server can process the request itself, or it can act as a proxy to request processing from the back-end server.

  After processing the request, the server sends the result back to the client as a response. It is assumed that TCP / IP is used as a protocol for transmitting / receiving a request / response between the server and the client.

The server of this embodiment appropriately controls TCP / IP processing according to the load on the server. Here, the following resources are assumed as the server load measurement targets. At this time, only one of the following resources may be measured, all resources may be measured, or a plurality of resources may be selectively measured.
CPU usage rate (U): Percentage of CPU time used per unit time in the entire server Memory usage amount (M): Memory amount used by the server Number of simultaneously connected TCP connections (C): Client and server Further, a threshold value is defined for each resource. The threshold values for the CPU usage rate U, the memory usage amount M, and the simultaneous connection TCP connection number C are expressed as LU, LM, and LC, respectively. In the present invention, it is expressed that the server is overloaded when each resource to be measured exceeds the threshold. In this embodiment, control of TCP / IP processing is attempted so that each resource is equal to or less than a threshold value.

The configuration of the server of this embodiment is shown in FIG. The server of this embodiment is configured by the following.
・ Application section 10
Request priority control unit 11
-TCP / IP processing unit 12 with load control function
・ Priority prediction unit 13
The application unit 10 creates content based on the URL specified in the request and returns it as a response. Even if the application unit 10 processes the request, the application unit 10 may be operated as a reverse proxy and the back-end server may process the request.

  The request priority control unit 11 performs priority control according to the request type in the application layer of the server. When the server receives the request via TCP / IP, the request is then analyzed before the request is processed by the Web application. Then, the Web application preferentially processes requests determined to be important, for example, requests belonging to ongoing page session processing. If the server is too busy to process all requests, it rejects requests with low importance.

  The priority prediction unit 13 predicts the priority of subsequent requests transmitted from the same transmission source IP address based on the request analysis result in the request priority control unit 11. Here, in order to simplify the description, only two stages of high priority and low priority are used as request priorities. The priority predicting unit 13 registers the transmission source IP address of the client for which the subsequent request is determined to have high priority in the high priority address table.

  In the TCP / IP processing of this embodiment, the load is controlled by referring to this high priority address table. FIG. 3 shows a high priority address table. In the high priority address table, an IP address of a transmission source that is predicted to transmit a high priority request and an expiration date of the entry are registered. The expiration date can be set as an absolute time such as 13:45:30 on May 21, 2006, or as a relative time such as how many seconds remain.

  If the same address has already been registered at the time of registration in the high priority connection table, it is only necessary to update the expiration date of the corresponding entry. In this specification, a client whose IP address is registered in the high priority address table is called a high priority client, and a client that is not registered is called a low priority client. A connection connected to a high priority client is called a high priority connection, and a connection connected to a low priority client is called a low priority connection.

The priority prediction method in a present Example is shown. All of the methods described below may be used, or only one of them may be used.
Prediction based on progress of processing requiring a plurality of requests: As described above, there are page processing and session processing as processing that requires execution of a plurality of requests to complete one processing. In this embodiment, based on the progress of page processing or session processing, the priority of a request sent in the future is predicted in units of IP addresses.

  Priority prediction based on progress of page processing: From the viewpoint of page processing protection, priority may be given to requests belonging to the page processing being processed. Therefore, when the processing of the request to the page root resource is completed, the client address of the request is registered in the priority address table. Whether or not it is a page root resource can be determined from the URL of the request by storing the page root resource and other resources in different directories, for example. Set a value sufficient for the completion of page processing as the expiration date of the high priority address table. In general, page processing is usually completed in a few seconds.

  Priority prediction based on the progress of session processing: From the viewpoint of session processing protection, requests belonging to the page processing being processed may be prioritized. Therefore, the address of the client whose session start is approved by the application is registered in the priority address table. As a method for determining whether or not the session processing has been approved, for example, there are the following methods.

  In general, a Web server issues a session identification number at the start of session processing. Then, the session identification number is given to the response using HTTP Cookie or the like and notified to the client. The client gives the session identification number notified from the Web server to the subsequent request, so that the Web server identifies the session to which the request belongs.

Therefore, the priority predicting unit 13 considers that the client has started session processing when a session identification number is given to the response. Then, the IP address of the client to which the response is returned is stored in the high priority address table. The expiration date is set to a value sufficient to complete session processing. Since session processing generally requires several minutes to several tens of minutes, it should be set longer than the expiration date of page processing.
Precedence prediction based on user identification number (user ID): The request may be preferentially controlled based on the user ID stored in the request. The user ID is generally stored in a Cookie or URL query. Therefore, when the request is received, the cookie and URL of the request are inspected to determine whether or not the user ID is included.

  If the user ID is included, it is next determined whether the user ID is an ID to be prioritized. If it is a user ID that should be given high priority, it is determined that the client that transmitted the request is likely to include the user ID in the request in the future, and the address is registered in the high priority address table. Since it is considered that the relationship between the user ID and the address does not change frequently, the expiration date can be set longer than in the case of page processing and session processing.

(First Example)
As a first embodiment, an execution procedure of the load control function-equipped TCP / IP processing unit 12 will be described. The TCP / IP processing unit 12 with load control function is TCP / IP extended so that load control based on the present invention is possible. That is, when the server is overloaded, packet reception from the client is restricted based on the prediction result of the priority prediction unit 13. Since most of the processing contents are the same as those in the normal IP processing, only the points different from the normal TCP / IP processing will be described here.

  First, at the start of this TCP / IP process, whether or not an IP packet can be received is controlled according to the server load. Here, the connection rate of the TCP connection is particularly controlled. In the case of an overload, by limiting the connection request of the TCP connection, that is, the TCP / SYN packet rate, in the IP packet, the transfer of the request from the new connection is suppressed while protecting the existing connection.

  Here, the upper limit of the connection request rate is represented by F (pps: packet per second). At this time, in this embodiment, when the server load exceeds the threshold, the upper limit F of the connection request rate is decreased. If a TCP / SYN packet exceeding the upper limit F is received, the excess is discarded.

  On the other hand, when the TCP / SYN packet is discarded even though the server load does not exceed the threshold, the upper limit F of the connection request rate is increased. Furthermore, in the present embodiment, priority control using priority predicted in the application layer is performed in connection request rate limitation. In this specification, it is assumed that TCP is used as the L4 protocol for transmitting a request. However, when UDP is used, a TCP / SYN packet may be read as a UDP packet.

  To simplify the description, it is assumed that TCP / SYN packets from high priority clients are always allowed to be received. On the other hand, TCP / SYN packets from low priority clients are subject to rate control. Further, an example of the packet rate control algorithm will be described using a token bucket method.

  In the token bucket method, tokens necessary for executing packet processing are added to a bucket of a certain capacity at a constant speed. If the number of tokens exceeds the bucket capacity, stop adding tokens. When the IP packet is received, if a token remains in the bucket, the token is consumed and packet processing is started. On the other hand, if the token is empty in the bucket, the packet is discarded. By adjusting the token addition speed and the bucket capacity, the packet rate can be appropriately limited.

  FIG. 4 shows an execution procedure for rate limiting according to the present invention. FIG. 4 is executed as pre-processing of TCP / IP processing when the server receives an IP packet from the network. First, when an IP packet is received, it is checked whether the packet is a TCP / SYN packet. If it is a TCP / SYN packet, the TCP / IP process is continued as it is.

  Next, it is determined whether or not the source IP address of the IP packet is registered in the priority address table. If it is registered in the priority address table, the TCP / IP process is continued as it is.

  Next, it is determined whether or not a token for processing the IP packet remains. If there are remaining tokens, one token is consumed and TCP / IP processing is continued as it is. On the other hand, if there is no remaining token, it is assumed that the rate of the TCP / SYN packet is higher than F, the discard flag is set, and then the IP packet is discarded.

  Next, FIG. 5 shows an execution procedure of a processing unit (token addition processing unit) that adds a token to the bucket. The token addition processing unit is activated at the same time when the server is activated, and repeats the token addition processing until the server operation is completed. It is assumed that the token addition process is performed every time D, and the execution waits until the time D elapses. Next, it is determined whether or not F needs to be updated. F may be updated at regular time intervals, or F may be updated each time the number of received packets exceeds a certain number.

  When F needs to be updated, it is first determined whether or not the server is overloaded. That is, it is determined whether the resource to be measured among the CPU load U, the memory usage M, and the number of TCP connections C exceeds the respective threshold values LU, LM, and LC. Next, when the threshold value is exceeded, F is reduced by ΔF in order to reduce the server load. On the other hand, if the threshold is not exceeded, it is next determined whether or not the current limit by F is too strong.

  In other words, if the TCP / SYN packet is discarded despite not being overloaded, it means that the rate limit is too strong. Therefore, if the discard flag is true, F is increased by ΔF to relax the rate limit. Finally, the discard flag is reset.

  If F update is not required, or when F update is complete, the token is then added to the bucket. The number of tokens added at this time is F × D. At this time, the number of tokens should not exceed the bucket capacity.

  Note that the rate control of TCP / SYN packets may be performed not by the server but by a network device (router, L2 switch, Firewall, load balancer) connected to the previous stage of the server. In this case, the priority IP address and the CPU load of the server are transmitted to the network device. Then, the TCP / SYN packet rate control is performed in the network device based on the execution procedure of FIGS.

(Second embodiment)
In the first embodiment, the load on the server is reduced by limiting the rate of the TCP connection newly established at the time of overload. However, the server load can be reduced by controlling the request rate of each TCP connection during an overload.

  As a function for avoiding buffer overflow on the TCP reception side, there is a flow control based on the window size. In flow control, when an ACK that is a response message is returned, the number of bytes remaining in the reception buffer is notified to the transmission side as a window size. On the transmission side, data larger than the notified window size is not transmitted. Therefore, the smaller the window size of the server, the smaller the amount of data transmitted from the client (that is, the request amount).

  In the present embodiment, control is performed so that the window size becomes smaller during overload. At this time, the connection connected to the low priority client may be preferentially restricted. In the present specification, for simplification of description, it is assumed that the window size control of the connection connected to the high priority client is not performed. The window size in TCP processing is set to w. When the coefficient k (0 ≦ k ≦ 1) is used for this, the window size w notified to the client by the ACK packet is expressed by the following equation.

W = w... Connection connected with high priority client W = kw... Connection connected with low priority client Here, k is periodically updated based on the execution procedure shown in FIG. First, an update is waited for a predetermined time. Next, it is determined whether or not an overload has occurred. That is, it is determined whether the resource to be measured among the CPU load U, the memory usage M, and the number of TCP connections C exceeds the respective threshold values LU, LM, and LC. When overload occurs, the value of k is decreased by Δk so that a smaller window size is notified to the client.

At this time, if the value of k becomes too small, data is transmitted by a packet having a small size, and the efficiency of the server is lowered. Therefore, if the value falls below the lower limit K _MIN of k, k is set to 0 so that a request does not flow from the client having a low priority to the server.

  On the other hand, if no overload occurs, the value of k is increased by Δk so that more requests are transmitted. First, it is confirmed whether the value of k is 0. When k is 0, the request transmission from the client to the server is stopped. Therefore, an ACK message indicating that the window size has been restored is transmitted to the client having a low priority. Next, the value of k is increased. At this time, k does not exceed 1.

  In FIG. 6, the window size of the low priority connection is uniformly reduced. However, it is possible to set k for each connection, and it is also possible to stop transmission of some requests of non-priority connections. FIG. 7 shows the execution procedure. First, an update is waited for a predetermined time. Next, it is determined whether or not an overload has occurred. That is, it is determined whether the resource to be measured among the CPU load U, the memory usage M, and the number of TCP connections C exceeds the respective threshold values LU, LM, and LC.

  When an overload occurs, a connection for stopping request transmission is selected from low priority connections in which k> 0. Here, a plurality of connections may be selected. The connection may be selected at random, or may be selected from a connection that has not been used for transmission / reception of a request / response, a connection having the longest period, or a connection that has been newly connected. Next, k of the selected connection is set to 0. This stops request transmission from the selected connection and measures the load on the server.

  On the other hand, if no overload has occurred, a connection that resumes request transmission is selected from the connections that have stopped request transmission, that is, connections with k = 0. Here, a plurality of connections may be selected at a time. Further, the connection may be selected randomly, or a connection that has been suspended for the longest time may be selected. Next, k of the selected connection is set to 1, and an ACK message indicating that the window size has been restored is transmitted to the client.

  Note that TCP / SYN rate control and window size control can be used in combination. In this case, for example, TCP / SYN rate control is first performed. If the overload is not resolved even if the value of F is sufficiently small, a method of reducing the request rate of an already established connection using window size control can be used.

  In FIG. 7, the packet reception is stopped by changing the window size, but the packet reception may be stopped by cutting the selected connection. In this case, processing when the server is not overloaded is unnecessary.

(Third embodiment)
As a third embodiment, an overload avoidance method focusing on the number C of simultaneously connected TCP connections will be described. In this embodiment, when the number C of simultaneously connected TCP connections is overloaded, in the connection establishment process (for example, accept () in C language) from the client registered in the priority address table. Establish a connection preferentially.

  FIG. 8 shows the execution procedure of the TCP connection connection process according to this embodiment. This process is called when a new TCP connection is established. First, it is determined whether or not the number C of simultaneously connected TCP connections exceeds the threshold LC by establishing a new TCP connection. If the threshold is not exceeded, the process is terminated. When the threshold value is exceeded, it is checked whether or not the IP address of the transmission source client of the newly established TCP connection is registered in the priority address table. If it is not registered in the priority address table, the newly established TCP connection is disconnected and the process is terminated.

  When registered in the priority address table, it means that a newly established TCP connection should be maintained with priority. Therefore, it is first checked whether there is a TCP connection connected to a client that is not registered in the priority address table. If it exists, one connection is selected from the TCP connections connected to the clients not registered in the priority address table and disconnected. The connection to be disconnected may be selected at random, or the connection that has not been used for request / response transmission / reception for the longest period or the connection that has been newly connected may be selected. On the other hand, if it does not exist, the new connection is disconnected.

  Based on the present invention, the request is not discarded and interrupted in the middle of the service, and the processing capability of the server can be improved.

The block block diagram of the load control system of a present Example. The block block diagram of the server of a present Example. The figure which shows an example of the high priority address table | surface of a present Example. The flowchart which shows the reception process procedure of the IP packet of a present Example. The flowchart which shows the execution procedure of the token addition process of a present Example. The flowchart which shows the Example (uniform regulation) of the Window size control of a present Example. The flowchart which shows the Example (selection regulation) of the window size control of a present Example. The flowchart which shows the connection selection process sequence at the time of the new TCP connection establishment of a present Example.

Explanation of symbols

1-1 to 1-n Client 2 Network 3 Server 10 with load control function Application unit 11 Request priority control unit 12 TCP / IP processing unit with load control function 13 Priority prediction unit

Claims (12)

In a load control system used for a server that receives a request from a client and returns a response,
Means for setting a priority for each IP address of a request source client based on progress of a process requiring a plurality of requests;
Means for determining whether the server load exceeds a threshold;
And a means for restricting reception of an IP packet from an IP address set to a low priority when a threshold value is exceeded. In a load control system used for a server that receives a request from a client and returns a response,
A means for setting a priority for each IP address of the request source client based on the user identification information stored in the request;
Means for determining whether the server load exceeds a threshold;
And a means for restricting reception of an IP packet from an IP address set to a low priority when a threshold value is exceeded.   3. The load control system according to claim 1, wherein reception of an IP packet is limited by limiting a rate of a TCP / SYN packet among IP packets using a TCP / IP protocol.   3. The load control system according to claim 1, wherein reception of an IP packet is limited by reducing a window size of a TCP header in a TCP-ACK packet transmitted from a server system using a TCP / IP protocol.   The load control system according to claim 1 or 2, wherein reception of an IP packet is restricted by disconnecting a TCP connection using the TCP / IP protocol.   6. The IP packet reception limit is gradually increased when a server load exceeds a threshold value, and the IP packet reception limit is gradually decreased when a server load becomes equal to or less than the threshold value. Load control system as described in. In a load control method used for a server that receives a request from a client and returns a response,
Based on the progress of processing that requires multiple requests, set the priority for each IP address of the request source client,
A load control method characterized by determining whether or not a server load exceeds a threshold, and restricting IP packet reception from an IP address set to a low priority when the threshold is exceeded. In a load control method used for a server that receives a request from a client and returns a response,
Based on the user identification information stored in the request, set the priority for each IP address of the request source client,
A load control method characterized by determining whether or not a server load exceeds a threshold, and restricting IP packet reception from an IP address set to a low priority when the threshold is exceeded.   The load control method according to claim 7 or 8, wherein reception of an IP packet is limited by limiting a rate of a TCP / SYN packet among IP packets using a TCP / IP protocol.   The load control method according to claim 7 or 8, wherein reception of an IP packet is limited by reducing a window size of a TCP header in a TCP-ACK packet transmitted from a server system using a TCP / IP protocol.   The load control method according to claim 7 or 8, wherein reception of an IP packet is restricted by disconnecting a TCP connection using a TCP / IP protocol.   12. The IP packet reception limit is gradually increased when a server load exceeds a threshold value, and the IP packet reception limit is gradually decreased when a server load is equal to or less than the threshold value. The load control method described in 1.

Images