JP2012247971A - Load information unshared type load leveling server, system with the server, and method thereof - Google Patents

Abstract

An object of the present invention is to smooth the processing load between servers without introducing a distribution node between the server and the client, and without directly reporting the load status to other servers.
A server receives a processing request from a client terminal having an equiprobable server selection function, and the server determines whether the processing request message is an original message or a redirect message, and the original In the case of a message, it is accepted at the acceptance rate and stored in the original request queue. The original message that is not accepted is redirected, and the redirect message is stored in the redirect request queue. In addition, the server dynamically redirects processing request messages stored in the original request queue and the redirect request queue.
[Selection] Figure 3

Description

  The present invention provides a server that smoothes a processing load between servers without sharing load information when a plurality of servers are distributed and content is provided (transmitted) to a large number of clients, and a system including the server It relates to that method.

  Many techniques for smoothing the processing load between WEB servers have been proposed. For example, in Patent Document 1, when distributing a load to a plurality of servers, a distribution node is arranged in front of the WEB server, and content requests are distributed according to the load status of the server. In Patent Document 1, a distribution node is essential in addition to the WEB server.

  Patent Document 2 discloses a WEB server system and a load distribution method capable of distributing a load to a WEB server with a small load, and Patent Document 3 discloses a free recording capacity of a content recording unit for recording content data. Disclosed is a storage instruction device that instructs a node device having a large storage to store. In Patent Document 2 and Patent Document 3, a message exchange function for transmitting and receiving a load status or a storage free status between WEB servers or between file storage nodes is essential.

  Patent Document 4 discloses a method of switching the connection destination of a WEB server or a cache server in accordance with the load status of the WEB server. The WEB server and the cache server form a hierarchical structure, and the load status is exchanged between nodes corresponding to the nodes of the tree, and the WEB server or the cache server to which the client is connected is switched.

Japanese Patent Laid-Open No. 2008-40718 JP 2009-237835 A JP 2010-67073 A JP 2009-237637 A

  However, in the above-described conventional technology, the load status is directly notified between the load balancing nodes or WEB (cache) servers and the load balancing processing is performed. Therefore, the configuration becomes complicated.

  In Patent Document 1, since a distribution node is used, in order to eliminate a single point of failure, a redundant configuration of the distribution node is essential, and the complexity of introducing the distribution node increases. Furthermore, if the server stops due to a failure, the distribution node must immediately delete the server from the distribution candidates, and the distribution node must always check the operability of the WEB server.

  In Patent Documents 2, 3 and 4, in order to notify each other of the load status between servers, a function for associating WEB servers with each other is indispensable. The operation status monitoring of the server and the association of the load status notification must work closely, and the entire system becomes complicated.

In addition, DNS (Domain Name), which is a common means on the Internet, is used to distribute server load.
In the case of using System) load balancing, the above-mentioned patent document has a drawback that the load cannot be appropriately distributed among servers. For example, in the DNS load distribution, requests are distributed evenly to N WEB servers. Therefore, when the processing performance of the WEB server is not uniform, the load is biased.

  In the DNS load distribution, the DNS performs name resolution to obtain an IP address from the WEB server name before the client connects to the server. The DNS can associate a plurality of IP addresses with the same WEB server name, and can respond to all the corresponding IP addresses to the client. In many cases, a client that has received notification of a plurality of IP addresses selects one IP address with equal probability and connects to the WEB server. The operations of the DNS and the client are not specific to the WEB service, but are used in various services on the Internet.

  Therefore, the present invention does not introduce a distribution node between the server and the client, and the server does not directly notify the load status with other servers, so that a plurality of servers are distributed and contents are distributed to many clients. An object of the present invention is to provide a server that smoothes the processing load between servers, a system including the server, and a method thereof.

  The server that smoothes the processing request load to achieve the above object is a server that smoothes the processing request load from the client terminal having the equiprobable server selection function, and the processing request message is redirected. A determination means for determining whether the original message is a redirected message or a redirected redirect message; an original acceptance means for accepting the original message at a reception rate and redirecting the original message that was not accepted; An original request queue for storing an original message, a redirect request queue for storing the redirect message, and a process for dynamically redirecting messages of processing requests stored in the original request queue and the redirect request queue And a redirection means.

  The dynamic redirect means preferably operates at time intervals based on a redirect rate.

  The original message acceptance rate and the redirect rate are preferably updated at regular intervals based on the number of redirect messages transmitted by the server and the number of redirect messages received from other servers.

  The original message acceptance rate is obtained by obtaining an adjacent server of the server, and adding the number of redirect message transmissions from the adjacent server divided by the number of redirect message transmissions of the server divided by the total number of servers. By applying a bias, a request number index is calculated, and the request number index is multiplied by an acceptance factor. If this result is 0 or more and less than the maximum acceptance rate, this result is regarded as the original message acceptance rate. 0 is calculated by setting the maximum acceptance rate as the original message acceptance rate when the maximum acceptance rate is equal to or greater than the maximum acceptance rate, and when the request number index is negative, the redirect rate includes a redirect response factor and − Multiply by 1 and when this result is greater than or equal to the reciprocal of the time interval, And Irekutoreto, the request index number if non-negative or the result is less than the inverse of the time interval, it is also preferable to calculate the reciprocal of the time interval as a redirect rate.

  In addition, service processing means for processing a processing request message is further provided, and the dynamic redirecting means redirects the processing request message in the original request queue with priority when the processing request message is stored in the original request queue. Preferably, the service processing means preferentially processes the message in the redirect request queue when the message is stored in the redirect request queue.

  A client / server system that smoothes the load of processing requests to achieve the above object is a client terminal having an equiprobable server selection function and a processing request message from the client terminal is an original message that is not redirected. Determining means for determining whether the message is a redirected redirect message, an original acceptance means for accepting the original message at a reception rate and redirecting the original message that was not accepted, and an original request for storing the accepted original message A queue, a redirect request queue for storing the redirect message, and a dynamic request for dynamically redirecting the processing request messages stored in the original request queue and the redirect request queue. And a server and a Irekuto means.

  A method of smoothing a processing request load to achieve the above object is a method of smoothing a processing request load from a client terminal having an equiprobable server selection function, wherein the processing request message is redirected A determination step for determining whether the message is an original message that has not been redirected or a redirected redirect message; an original acceptance step for accepting the original message at a reception rate and redirecting the original message that has not been accepted; Storing an original message in an original request queue; storing the redirect message in a redirect request queue; and processing request messages stored in the original request queue and the redirect request queue. Manner and a dynamic redirection step to redirecting.

  The above-described present invention uses server content request redirection, so that a distribution node is not introduced between the server and the client, and the server does not notify the load status directly with other servers. Thus, it is possible to smooth the processing load.

The equal probability distribution of content requests by client terminals is shown. The equal probability distribution of the redirect message by the client terminal is shown. The structure of the WEB server of this invention is shown. The flow of content request message acceptance processing by the WEB server of the present invention is shown. 4 shows a message redirection process flow in a queue by the WEB server of the present invention.

  The best mode for carrying out the present invention will be described in detail below with reference to the drawings. In the present embodiment, it is assumed that all the plurality of WEB servers can provide the same content and a plurality of clients request the content. Also, only static pages are used for WEB content provided by the WEB server, and the WEB content is acquired from a database behind the server.

  In addition, a distribution node is not introduced between the WEB server and the client terminal, and the equiprobable server selection of the client is used for distributing processing requests. FIG. 1 shows equal probability distribution of content requests by client terminals.

As described in the background art, the client terminal is provided with an equal probability server selection function using DNS.
(1) The client terminal performs name resolution from the WEB server name to the IP address using DNS before connecting to the WEB server. A list of IP addresses from the DNS is returned to the client terminal, and the client terminal recognizes the number of servers n. When the number of WEB servers is changed, the IP address of the WEB server handled by DNS is changed.
(2) The client terminal selects one IP address with equal probability 1 / n from the list of IP addresses and connects to the WEB server. If there is no response from the selected server, another server is selected and an attempt is made to acquire WEB content. This processing enables the client terminal to autonomously connect to another server even when the server does not respond due to a failure.
(3) The content request of the client terminal is stored in the processing waiting queue of the WEB server, and is processed from the request previously stored in the queue.

In the present invention, the WEB server does not directly notify the load status with other servers. The WEB server is provided with a dynamic redirect function for redirecting a content request message according to its processing load. This redirection is realized by using a redirection process used in the HTTP protocol. The client terminal selects the server with equal probability even for the redirect message from the server. FIG. 2 shows equal probability distribution of redirect messages by client terminals.
(1) When the WEB server redirects the content request message of the client terminal, the WEB server adds its own IP address to the message.
(2) The client terminal that has received the redirect message again selects the WEB server with equal probability and transmits a redirect content request to the selected server. At this time, the same WEB server may be selected again.

  FIG. 3 shows the configuration of the WEB server of the present invention. The WEB server includes a determination unit 1, an original reception unit 2, a redirect request queue 3, an original request queue 4, a dynamic redirect unit 5, and a service processing unit 6.

The determination unit 1 determines whether the content request is original (not redirected) or redirected. Original receiving portion 2, accept the message the original message at the reception rate r _t, to redirect did not accept the original message. The redirect request queue 3 stores redirect messages, and the original request queue 4 stores original messages. The dynamic redirect unit 5 is means for dynamically redirecting messages stored in the redirect request queue 3 and the original request queue 4. The service processing unit 6 processes a processing request from a client terminal.

FIG. 4 shows a content request message acceptance processing flow by the WEB server of the present invention. The processing of accepting a WEB server will be described with this processing flow.
(S41) When the processing request message arrives, it is first checked whether it is a redirect message. In the case of a redirect message, the process proceeds to S45 in order to accept it unconditionally. In the case of an original message, the process proceeds to S42.
(S42) accept the message the original message at the reception rate _{r t.} When accepting the message, the process proceeds to S44. If not accepted, the process proceeds to S43.
(S43) The original message is redirected. At this time, a message storing the server IP address is returned to the client. Also, the number of redirects is updated.
(S44) The original message is stored in the original request queue. At this time, the number of original message arrivals is updated.
(S45) Store the redirect message in the redirect request queue. At this time, the number of redirect message arrivals for each server is updated so that the redirect source server can be distinguished.

Further, WEB server, a queue message independently of S45 from the processing S41, the redirect redirect rate _{q t.} FIG. 5 shows an in-queue message redirection processing flow by the WEB server of the present invention. With this processing flow, the redirect processing of the WEB server will be described.

(S51) This redirect process loops over the redirect rate q _t at a time interval 1 / q _t . When q _t is 0, the redirect time interval is I (parameter update interval). When the processing time comes, the process proceeds to S52.
(S52) In-queue message redirection processing is performed. At this time, when the original request message is stored, the original request message is redirected. When the original request message is not stored, the redirect message is redirected and the number of redirects is incremented by one.
(S53) The next processing time is updated.

  Further, the WEB server performs a service process (FIG. 3) independently of the processes S41 to S45 and S51 to S53. At this time, the WEB server preferentially processes the request in the redirect request queue, and processes the original message when the message is not stored in the redirect request queue. When service processing is performed, the processed number is updated.

The WEB server updates the original message reception rate r _t and the in-queue message redirection rate q _t as follows each time the parameter update interval I elapses.

(1) The adjacent server that has transmitted the redirect message from the previous update to the present is identified by the IP address stored in the redirect message. These adjacent servers are arranged in a number line based on the IP address of the server. Here, the front and back of the server of the WEB server _{S prev,} and _{S next.} When the WEB server has the minimum IP address or the maximum IP address, the number line is assumed to be circular, and the adjacent server is read as the server with the maximum IP address or the minimum IP address. Also, refrain from the number N of servers that have sent redirect messages.
(2) Let g _prev and g _next be the difference between the number of redirect messages received in the previous cycle of S _prev and S _next and the number of redirect messages received in the current cycle. In calculating this difference, the number of redirects in the previous period is subtracted from the number of redirects in the current period.
(3) Let g _out be the number of redirect message transmissions from this WEB server.
(4) First, the request number index u _t = g _prev + g _next −g _out / N + A is obtained. A (non-negative number) is given by the server administrator as an acceptance bias.
(5) If _{u t} is, if positive, seek the acceptance rate as _{r t} = _{αu t.} Here, if r _t > R _m , r _t = R _m is set. That, and 0 ≦ _r t _{<R m.}
If u _t is non-positive, let r _t = 0. Α (positive number) is given by the server administrator as a reception factor. R _m (R _m <1) is also given by the server administrator as the maximum acceptance rate.
(6) Store the number of redirect messages received this week as the number of redirect messages received in the previous period.
(7) If u _t is negative, determine the in-queue message redirect rate as q _t = −βu _t . If u _t is non-negative or 1 / I> q _t , q _t = 1 / I and 1 / I ≦ q _t (to perform redirection with the minimum period I). Note that β (positive value) is given by the server administrator as a redirect response factor.
(8) Reset (return to zero) the number of redirect messages received for each server in this cycle and the number of redirect messages received by this WEB server.

As described above, according to the present invention, redirection of a content request is positioned as information exchange of load status between servers, and the original message acceptance rate r _{t of} the WEB server and in-queue message redirection based on the number of redirect messages transmitted and received. to update the rate _{q t.}

For this reason, when the processing performance of the WEB server is different, a difference appears in the number of stored processing queues, and the processing waiting number of a server with low processing performance tends to increase. At this time, since g _out increases and redirects increase, further increase in the number of waiting processes is suppressed. Also, if the arrival of original messages decreases despite the large number of stored messages, the waiting for requests is redirected from the queue for processing to reduce the number of queues for processing. By increasing the parameter β, it is possible to improve the response performance for reducing the number of processing waits.

  Since the present invention accepts all redirect messages, it is possible to quickly increase the number of processing waits when the number of processing waits is smaller than that of the adjacent server. Furthermore, it can be increased more quickly by increasing the acceptance rate of the original message (increasing the parameter α).

  The acceptance bias A affects the increase / decrease of the acceptance rate and the decrease / increase of the redirect rate, and affects the size of the queue length. If A is increased, the queue length becomes longer.

In the present invention, when a WEB server receives a redirect message, the queue length of another WEB server can be estimated. The maximum reception rate R _m to be smaller than 1, it is possible to redirect levels in the original message, it is possible to maintain the accuracy of methods.

  In the above description, the embodiment of the present invention has been described in the form where the server is a WEB server and the client terminal requests content from the WEB server. However, the present invention is not specialized for the WEB service, and the client terminal includes an equiprobable server selection function using DNS and the like, and selects a server again with an equal probability for request redirection. If the server can provide the same service, it can be applied to other services other than the WEB service.

  Moreover, all the embodiments described above are illustrative of the present invention and are not intended to limit the present invention, and the present invention can be implemented in other various modifications and changes. Therefore, the scope of the present invention is defined only by the claims and their equivalents.

DESCRIPTION OF SYMBOLS 1 Discriminating part 2 Original acceptance part 3 Redirect request queue 4 Original request queue 5 Dynamic redirect part 6 Service processing part

Claims (7)

A server that smoothes the load of a processing request from a client terminal having an equal probability server selection function,
Determining means for determining whether the message of the processing request is an original message not redirected or a redirected redirect message;
An original receiving means for accepting the original message at an acceptance rate and redirecting the original message that was not accepted;
An original request queue to store accepted original messages;
A redirect request queue for storing the redirect message;
A server comprising: the original request queue; and dynamic redirect means for dynamically redirecting a message of a processing request stored in the redirect request queue.   The server according to claim 1, wherein the dynamic redirect unit operates at time intervals based on a redirect rate.   The original message acceptance rate and the redirect rate are updated at regular intervals based on the number of redirect messages transmitted by the server and the number of redirect messages received from other servers. server. The original message acceptance rate is
A request number index is obtained by obtaining an adjacent server of the server, adding a number obtained by dividing the number of redirect message transmissions of the server by the total number of servers to the increase in the number of redirect message receptions from the adjacent server, and adding an acceptance bias. When the request number index is multiplied by the acceptance factor, and this result is 0 or more and less than the maximum acceptance rate, this result is taken as the original message acceptance rate. Calculates the maximum acceptance rate as the original message acceptance rate,
The redirect rate is
When the request number index is negative, the request number index is multiplied by a redirect response factor and −1. When the result is equal to or larger than the reciprocal of the time interval, the result is set as a redirect rate, and the request number index is non-negative. The server according to claim 3, wherein when the result is smaller than the reciprocal of the time interval, the reciprocal of the time interval is calculated as a redirect rate. Service processing means for processing the processing request message is further provided,
The dynamic redirect means preferentially redirects the processing request message of the original request queue when the processing request message is stored in the original request queue,
5. The server according to claim 1, wherein, when a message is stored in the redirect request queue, the service processing unit preferentially processes the message in the redirect request queue. 6. . A client terminal having an equal probability server selection function;
Determining means for determining whether the processing request message from the client terminal is an unredirected original message or a redirected redirect message;
An original receiving means for accepting the original message at an acceptance rate and redirecting the original message that was not accepted;
An original request queue to store accepted original messages;
A redirect request queue for storing the redirect message;
A client / server system comprising: a server comprising: the original request queue; and a dynamic redirection means for dynamically redirecting a processing request message stored in the redirect request queue. A method of smoothing a load of a processing request from a client terminal having an equal probability server selection function,
A determining step of determining whether the message of the processing request is an original message that has not been redirected or a redirected redirect message;
An original acceptance step of accepting the original message at an acceptance rate and redirecting the original message that was not accepted;
Storing the accepted original message in the original request queue;
Storing the redirect message in a redirect request queue;
A dynamic redirection step for dynamically redirecting said original request queue and a message of a processing request stored in the redirect request queue.

Images