JP4485428B2 - Network system, management computer, cluster management method, and computer program - Google Patents

Description

The present invention relates to an efficient distributed computing technique performed by a plurality of computers connected to a computer network in a broadband environment.
More particularly, the present invention relates to a network system capable of distributed computing, its components, and a cluster management method in a network.

Today, it is common for a plurality of computers connected to a network to collaborate to process one job in a distributed manner. In the case of distributed processing of jobs, conventionally, in order to determine which job is assigned to which computer, it is indispensable to have a server that knows the processing capabilities of all computers that can be connected to the network in advance.
The server identifies the size of the job load and the surplus processing capacity (calculation resources) of each computer connected to the network when performing distributed processing, and selects a computer having surplus processing capacity according to the load. Sequential assignment is performed, and job execution results are received from the assigned computers.

In a conventional distributed processing method that requires a server, it is very difficult for the server to quickly grasp the surplus processing capacity of the computer that is connected to the network or disconnected at any time. . Further, since the server must receive the execution result from the computer that requested the job distribution processing and transfer it to the job request source, the overhead of the server increases. For this reason, there has often been a problem that the time required for job execution and the time required for data transmission via the network are substantially increased.
The main object of the present invention is to provide a distributed processing mechanism capable of solving such conventional problems.

The present invention relates to a network system capable of efficiently distributing and executing information processing, which is expected to change every moment, in various sizes, types, and numbers of jobs to be executed by a plurality of computers. The above problem is solved by the component and the cluster management method.
The “computer” mentioned here refers to a device including a processor that operates according to a computer program, but a device, processor board, processor chip, or processor itself that does not take the form of the device is also included in the concept of “computer”. For example, an individual processor, such as a multiprocessor system, can also be a “computer”.

The network system of the first configuration provided by the present invention is a network system in which a plurality of computers that can be clustered with other computers can freely join and leave.
This network system includes a first table in which information on whether or not each computer can be clustered is recorded, and a cluster that is already formed by one or more computers. And a second table for recording additional ease information indicating how easy it is to add a computer.
Any one of the computers forms a cluster including self and other computers whose clusterability information in the first table indicates a clusterable state, and for all the computers included in the formed cluster. Cluster forming means for updating the clusterability information to information indicating a non-clusterable state and further recording the additional ease information on the formed cluster in the second table. Further, the computer that has formed a cluster has the ease of addition recorded in the second table as to whether or not to add the candidate computer to the cluster when there is a candidate computer to be added to the cluster. It has a cluster growth means which determines based on information.

In the network system having such a configuration, any one of the computers autonomously forms a cluster based on the clustering availability information of each computer recorded in the first table, and is easy to add in the cluster. Information is recorded in the second table. The number of computers included in the cluster may be the default number determined at the time the cluster is formed, or a number that is determined flexibly according to the size of the job that was actually submitted or expected to be submitted in the near future. It may be. In any case, a computer whose clusterability information is clusterable is selected to form a cluster.
After the cluster is formed, if there is a candidate computer that is an additional candidate, whether the computer that formed the cluster adds the candidate computer to the cluster based on the additional ease information recorded in the second table Decide if.

  The addability information is information for the computer forming the cluster to autonomously determine whether it is appropriate to add the candidate computer, and various types of information can be used. Conditions such as adding only a few computers that have applied for the addition first, adding only a computer having identification number as a number generated by a random number, etc. can also be adopted as additional ease information. However, from the viewpoint of simplifying the process of determining whether or not to add, a measure such as ease of addition is quantified by a numerical value so that it can be compared with this numerical value. It is desirable to facilitate the addition of computers to the cluster. As an example, the numerical value in this case is one-time response to an inquiry about whether or not to add five times that occur irregularly, and the addition is permitted (the numerical value in this case is 1/5: 20%) In this case, addition of candidate computers is allowed only for the accepted time period of 10% of the system operation time of one day (in this case, the numerical value is 24 (hours) × 60 (minutes) × 60 (seconds) × 0 .1 × α (probability value)).

  Note that the numerical value may be held at a constant value regardless of addition of candidate computers, or may be variable after being recorded in the second table. If the numerical value can be changed, it is easier to gather candidate computers by changing the candidate computer added to the cluster as a trigger, and it becomes easier to collect other candidate computers. This is advantageous in that it is easy to form a cluster.

  In this way, as candidate computers are sequentially added to the cluster, clusters of various sizes are formed and grown on a plurality of computers in the network system. As a result, distributed computing is realized that can efficiently execute information processing that is expected to vary in size, type, and number of jobs to be executed. The problem is solved.

In one embodiment of the network system of the first configuration, the computer that formed the cluster deletes the cluster when the job execution by the cluster is completed, and is recorded in the second table. It further has cluster extinguishing means for restoring the clusterability information recorded in the first table for all computers belonging to the added cluster information and the extinguished cluster to the state before the cluster formation.
Until the job is executed, the cluster grows based on the additional ease information described above. However, since it disappears when the job execution is completed, compared to the case where the cluster is fixedly prepared, The computer can be used effectively.

In the network system of the second configuration provided by the present invention, a plurality of management computers each having one or a plurality of computers that can be clustered with other computers can freely join and leave. It is a network system.
“Managing one or a plurality of computers as computers under their control” means controlling the operations of a plurality of computers connected to the computers and monitoring their operation states.
In this network system, a first table in which information on whether or not each computer can be clustered is recorded, and a cluster already formed by one or more computers is another computer. And a second table in which easy-to-add information indicating how easy it is to add is recorded.
In addition, at least one of the management computers includes a computer in which the clusterability information in the first table represents a clusterable state among computers under its own control and computers under the control of another management computer. Cluster forming means for forming a cluster and updating the clustering availability information for all computers included in the formed cluster to information indicating a non-clusterable state. Further, the management computer that formed the cluster determines whether or not to add the candidate computer to the cluster when there is a candidate computer to be added based on the additional ease information recorded in the second table. Cluster growth means to be determined.

  In the network system having such a configuration, any of the management computers autonomously forms a cluster based on the computer clusterability information recorded in the first table, and additional ease information in the cluster. Is recorded in the second table. After the cluster is formed, if there is a candidate computer that is an additional candidate, the management computer that formed the cluster adds the candidate computer to the cluster based on the additional ease information recorded in the second table. Decide whether or not. The criteria for handling clusterability information and additional ease information, clustering, and cluster growth are the same as those of the network system of the first configuration.

In one embodiment of the network system of the second configuration, the management computer annihilates the cluster when job execution by the formed cluster is completed, and belongs to the second table and the erasure cluster The apparatus further includes cluster extinguishing means for restoring the recording information of the first table for all the computers that have been used to the state before the cluster formation.
Until the job is executed, the cluster grows based on the additional ease information described above. However, since it disappears when the job execution is completed, compared to the case where the cluster is fixedly prepared, The computer can be used effectively.

In the network system of the second configuration, the second table may exist at an arbitrary place, but normally, it is held by any management computer that will generate the second table. In this case, the management computer holds the second table of the number of computers under its control.
The second table is a master table that is generated for the first computer when the management computer that owns it forms a first cluster that includes the first computer under its control, and the management computer that owns the master table, It includes at least one of the slave tables generated for the second computer when the operation of the second computer under its control added to the second cluster formed by another management computer is monitored and controlled. Additional ease information is recorded in the master table. The management computer that holds the master table behaves as a master management computer that mainly performs information processing related to formation of the first cluster, change of the number of computers in the first cluster, and disappearance of the first cluster, and holds the slave table. The management computer behaves as a slave management computer for the second cluster.

  From the viewpoint of promoting clustering, the master management computer searches for a candidate computer to be added to the first cluster by inquiring of any management computer whether there is a computer in a clusterable state. It shall have means. Further, one of the management computers determines whether or not to add a candidate computer belonging to the management computer to the first cluster based on the additional ease information about the first cluster formed by the master management computer. Constitute.

  The network system of the third configuration provided by the present invention is a network system in which a plurality of computers that can be clustered with other computers can freely join and leave; already one or a plurality of computers For each of the clusters formed by the above, including a table listing identification information of other clusters to which the computers belonging to the cluster are related; the computer forming the cluster is a candidate computer that is an additional candidate A network system having cluster growth means for determining whether to add the candidate computer to another cluster related to the own cluster, based on the identification information listed in the table It is. “Associated” means a state in which, for example, it is possible to communicate with each other and perform cooperation processing. In the network system having such a configuration, clustering according to the power distribution can be performed regardless of the additional ease information.

The present invention also solves the above problem by providing a management computer for configuring the network system of the second configuration, for example.
Each of these management computers is a management computer that owns one or a plurality of computers that can be clustered with other computers; freely participate and leave with other management computers having similar functions. A network connection means for connecting to a computer network capable of performing a clustering; a first table recording clustering availability information indicating whether each computer is in a clusterable state, and one or more already Table management means for allowing access to a second table for recording additional ease information indicating how easily a cluster formed by computers can add another computer; A computer and other management computers A cluster including computers whose clusterability information in the first table of the computers indicates a clusterable state is formed, and the clusterability information for all computers included in the formed cluster is clustered. A cluster forming unit that updates the information indicating an incapable state and records the additional ease information about the cluster in the second table; and when there is a candidate computer that is an additional candidate, Cluster growth means for determining whether to add to the cluster based on the additional ease information recorded in the second table. When the execution of the job by the formed cluster is completed, the cluster is deleted, and the additional ease information recorded in the second table and the first computer for all computers belonging to the deleted cluster. It is also possible to further have a cluster disappearance means for returning the clustering availability information recorded in the table to the state before the cluster formation.

The present invention also solves the above problem by, for example, a cluster management method executed by a plurality of computers included in the network systems having the first configuration and the third configuration described above.
A first cluster management method is a cluster management method in a network system in which a plurality of computers that can be clustered with other computers can freely join and leave; each computer has its own Recording the clusterability information indicating whether or not the state is clusterable in a first table; any one of the computers can cluster the clusterability information of itself and the first table; Forming a cluster including other computers representing the state, updating the clustering availability information for all computers included in the formed cluster to information representing a non-clusterable state, and Addability information that indicates how easy it is to add other computers Determining whether to add the candidate computer to the cluster when there is a candidate computer to be added based on the additional ease information recorded in the second table. A cluster management method comprising:
When the computer that formed the cluster finishes executing the job by the formed cluster, the cluster disappears, and the first table for all computers belonging to the second table and the disappeared cluster. And a step of returning the recorded information of the table to the state before the cluster formation.

  A second cluster management method is a cluster management method in a network system in which a plurality of computers that can be clustered with other computers can freely join and leave, respectively; A computer that forms a cluster lists identification information of other clusters to which a computer that belongs to the cluster is associated in a predetermined table; a candidate computer in which the computer that forms the cluster is an additional candidate Determining whether to add the candidate computer to another cluster associated with the cluster based on the identification information recorded in the table. is there.

The present invention also solves the above problems by providing a computer program for giving a computer a predetermined function.
Each of the first computer programs is read and executed by any computer in the network system in which a plurality of computers that can be clustered with other computers can freely join and leave. A first table in which information on whether or not each computer can be clustered is recorded, and a cluster already formed by one or more computers. Table management means for enabling access to the second table for recording additional ease information indicating whether or not it is easy to add; self indicating that the clusterability information of the first table indicates a clusterable state And forming a cluster containing other computers In addition, the clustering availability information of the first table for all computers included in the formed cluster is updated to information indicating a non-clusterable state, and the additional ease information for the cluster is further updated. Cluster forming means for recording in the second table; based on the additional ease information recorded in the second table, whether or not to add the candidate computer to the cluster when there is a candidate computer to be added It is a computer program for functioning as a cluster growth means to be determined.

  The second computer program is read and executed by a management computer that owns one or more computers each capable of clustering with another computer, thereby executing the management computer; A network connection means for connecting to a computer network that can freely join and leave with a management computer having; a clustering availability information indicating whether each computer is clusterable or not Allows access to one table and a second table to record additional ease information representing how easy a cluster already formed by one or more computers can add another computer Table management means; the computer under its own umbrella Among the computers under the control of the computer and other management computers form a cluster including computers in which the clustering availability information in the first table indicates a clusterable state, and all of the computers included in the formed cluster Cluster forming means for updating the clusterability information on the computer to information indicating a non-clusterable state, and further recording the additional ease information on the cluster in the second table; candidate to be an additional candidate A computer program for functioning as cluster growth means for determining whether to add a candidate computer to the cluster when a computer exists based on additional ease information recorded in the second table. .

  Each of the third computer programs is read and executed by any computer in the network system in which a plurality of computers that can be clustered with other computers can freely join and leave. Cluster management means for listing identification information of other clusters associated with computers belonging to the cluster when the computer is already formed with one or a plurality of computers in a predetermined table, an additional candidate, A cluster growth means for determining whether or not to add a candidate computer to another cluster related to the own cluster based on the identification information listed in the table. Con to make A-menu data program.

  These computer programs are recorded on a portable recording medium and distributed in the market, or downloaded from a program server or the like through a computer network accessible by a computer or a management computer.

  According to the present invention, clusters of various sizes are formed in any one of a plurality of computers of the network system, and grow based on recorded information in a predetermined table. It will be obtained. As a result, there are various job size, type, and number of jobs to be executed, and even if it is information processing of uncertainty that is expected to change from moment to moment, it is efficiently executed A unique effect is achieved in that distributed computing is realized.

<Network system architecture>
First, the architecture of a network system to which the present invention is applied will be described.
FIG. 1 is an overall view of a network system 101 to which the present invention is applied. The network system 101 includes a computer network 104. Examples of computer network 104 include a local area network (LAN), a global network such as the Internet, or other computer network.

  A management computer 106 having one or more computers under its control can be connected to the computer network 104 at an arbitrary timing. Each management computer 106 can perform bidirectional communication with another management computer 106 and another computer 108 under the control of the other management computer 106. Each computer 108 can be part of any management computer 106 at any time. That is, these computers 106 and 108 can participate in the network system 101 in a form that can be freely detached at any time.

  The “computer” in this embodiment refers to an apparatus including a processor that operates according to a computer program, but a device, a processor board, or a set of these that does not take the form of the apparatus is also included in the concept of “computer”. Further, “having one or a plurality of computers 108 as a subsidiary” means controlling the operations of the plurality of computers 108 connected to the computer 108 and monitoring their operation states.

Examples of the management computer 106 include a computer having a server function, a game console with a communication function, a computing device, and a processor board.
Examples of the computer 108 under the management computer 106 include a personal computer, a game console with a communication function and other wired or wireless computers, a computing device, a processor board, and the like.

There are various connection forms of the computer 108 to the management computer 106. For example, as shown in FIG. 1, a star connection type in which a plurality of computers 108 are directly connected around a management computer 106 connected to a computer network 104, via a local network to which the management computers 106 are connected. There are a local network type in which a plurality of computers 108 are connected, a wide area network type in which a plurality of computers 108 are connected to a management computer 106 via a computer network 104, and the like.
There are various types of local networks in the local network type. Some networks are directly connected to the computer network 104, and some networks are connected to the computer network 104 via the management computer 106.

  As shown in FIG. 2, each management computer 106 includes a storage device 1061 such as a hard disk, a communication device 1063, a semiconductor memory 1065, and a processor 1067 connected to each other via a bus B 11. The management computer 106 having such hardware resources allows the processor 1067 to read and execute a computer program loaded into the semiconductor memory 1065 through a recording medium such as a CD-ROM, and as shown in FIG. A plurality of management tables 206 (to be described later) are stored in the device 1061, and the functions of the table management unit 212, cluster management unit 216, communication control unit 236, and job execution unit 256 are built in the computer main body.

  The table management unit 212 accesses the management table 206 and updates the recorded contents of the management table 206. The cluster management unit 216 performs information processing related to the formation, growth, and disappearance of a cluster to which the computer 108 under its control and the computer 108 under the control of the other management computer 106 belong. That is, the cluster management unit 216 functions as a cluster forming unit, a cluster growing unit, and a cluster extinguishing unit in accordance with a change in the cluster state.

  The communication control unit 236 enables communication between the computer 108 under its control and the other management computer 106 through the communication device 1063. The job execution unit 256 submits a job to a cluster including the computer 108 under its control and executes it. The job execution unit 256 may have a function of calculating the job size, processing time, and the like in order to easily identify the optimal cluster size, that is, the number of computers.

  The computer 108 has the same storage device, communication device, semiconductor memory, and processor as the management computer 106. In this embodiment, a status table described later is stored in the storage device. Also, various functions related to cluster management and job execution are built in the computer main body by the processor reading and executing the computer program loaded into the semiconductor memory through a recording medium such as a CD-ROM. Preferably, the other computer 108 and the instruction set architecture (ISA) are all considered to be the same or the same, and the required processing can be executed according to the same instruction set.

The number of subordinate computers 108 in each management computer 106 is arbitrary. In a certain management computer 106, the number of computers 108 is assigned depending on the processing capability required to execute jobs given by various applications.
When the ISAs of a plurality of computers 108 under the management computer 106 are considered to be the same or the same, adaptability in the management computer 106 or the network system 101 can be dramatically improved.

  Each management computer 106 includes one or more of the computers 108 under its own umbrella or the computers 108 under the control of the other management computer 106 to form a cluster so that jobs can be executed in units of clusters. There are no performance differences, restrictions, etc. in forming at least one cluster between the computer 108 under its own umbrella and the computer 108 under the umbrella of the other management computer 106. By handling in this way, it is not so important which computer 108 of which management computer 106 executes the job. It is only necessary to specify the recipient of the job execution result as the management computer 106 that requested the job, the computer 108 under the management computer 106, or an arbitrary computer 108 that executes the subsequent job. Therefore, individual jobs can be easily executed in a distributed manner among computers 108 belonging to a plurality of management computers 106 connected to the computer network 104.

  If each computer 108 has an ISA that is all or considered the same as described above, the computational burden of an additional layer of software to achieve compatibility between computers 108 is also avoided. However, many of the problems of mixing heterogeneous computer networks can be prevented. Therefore, it is desirable to configure such a network system 101 according to the application. In such a network system 101, the difference in architecture between the plurality of management computers 106 and the computers 108 connected to the computer network 104 is absorbed, and is under the control of each management computer 106 as shown in FIG. A large-scale information processing integrated body WO capable of realizing broadband processing, in which each of a large number of computers 108 functions as a cell of information processing (Cell), is formed.

  Each computer 108 in the large-scale information processing integrated body WO is physically managed by the management computer 106 to which the computer belongs, and operates as a single computer or together with other computers 108 under the management computer 106. They are clustered and work together. However, logically, there is no wall by the management computer 106, and clustering with other computers 108 belonging to different management computers 106 is also possible. When clustered in such a form, one job can be distributedly executed by a plurality of computers 108 belonging to the same cluster.

<Distributed computing>
The present invention provides an efficient distributed computing mechanism using the architecture of the network system 101 described above. In order to enable such a mechanism, in the present invention, for example, each of a plurality of computers 108 constituting a network such as the large-scale information processing integrated body WO shown in FIG. 3 becomes a node, and starts from a small number of nodes. Thus, “growth” in which the number of nodes increases with the passage of time and “selective connection” as a link connection method between nodes can be repeated.
“Selective connection” means that when a node newly joined to the network establishes a link, it becomes easier to connect to a node that already has more links as a criterion for selecting a link destination node. It is. The two characteristics of growth and selective connectivity make the network grow and have a scale-free nature. It is detailed in Albert Barabasi, Reka Albert, Hawoong “Mean-field theory for scale-free random networks” that the distribution of the number of links of a node becomes a power distribution by repeating such growth and selective connection. Is shown in

FIG. 5 is a graph in which the measured values are plotted with the horizontal axis representing the number of links of nodes when the nodes are randomly connected and the case of selective connection, and the vertical axis representing the number of nodes having the number of links. is there. FIG. 5A shows an example in which nodes are connected at random. In this example, a representative node having a typical number of links appears. FIG. 5B is an example in the case of selective connection, a representative node does not appear, and the number of links extends over a wide range. That is, it is called scale free in the sense that a typical scale does not appear. The distribution of FIG. 5B is expressed as follows, and when written in a log-log graph, the distribution is very wide and forms a straight line.
P (k) = AK ^ (− r) (1)
Where k: number of links, P (k): number of nodes

  The present invention applies the above two features, interprets each node as a cluster, and interprets a link as two computers 108 included in a cluster corresponding to nodes connected to each other. Thus, clustering is performed in a form that can have an arbitrary distribution, of course, the power distribution shown in FIG. 5B, thereby enabling efficient distributed computing. In a network that grows in this way, the number of links is the number of computers 108 included in the cluster, that is, the cluster size or the computing capacity of the cluster. If the number of computers 108 included in the cluster is plotted on the horizontal axis and the number of clusters or the frequency of clustering is plotted on the vertical axis, the distribution should be similar to the power distribution shown in FIG.

  Hereinafter, in order to enable clustering as described above, how each computer 108 operates, and how each management computer 106 monitors and controls the operation of the computer 108 or other similar management computer 106. Describe what you should do.

Each computer 108 is currently managed by a computer cluster status (clustering availability information) indicating whether or not its own state is in a clusterable state. The computer cluster status is recorded in the status table, and is updated by the management computer 106 that manages the computer 108 or the computer 108 in accordance with changes in the state of the computer 108. The status table suffices if it exists in an arbitrary memory area accessible by the computer 108 or the management computer 106, but may be provided in the storage device of the computer 108.
In this embodiment, three types of information “clustered”, “run”, and “free” are used as examples of computer cluster statuses recorded in the status table. “Clustered” indicates that the computer is already clustered and is currently waiting for job execution. “Run” indicates that the computer is executing a job. “Free” indicates that the computer is in a non-operating state. When “clustered” or “run”, the computer 108 cannot be clustered. Therefore, “clustered” and “run” are information indicating a non-clusterable state. On the other hand, “free” is information indicating a clusterable state.

  One of the roles of the management computer 106 is to manage the computer 108 under its control as described above. The management table 206 is used for that purpose. As described above, the management table 206 is stored in, for example, the storage device 1061 included in the management computer 106. This management computer 106 has as many management tables 206 as the number of computers 108 under its control.

  There are two types of management table 206. The management table 206 classified into a certain type is a “master table” for recording information regarding the state of the cluster when the management computer 106 first forms a cluster including the computer 108 under its management. The other is a “slave table” for recording information about the state of the cluster when there is a computer 108 added to the cluster formed by another management computer 106.

  The management computer 106 behaves as a “master management computer” when the computer 108 included in the cluster managed by the management computer 106 is monitored and controlled. On the other hand, when the slave table supervises and controls the computer 108 belonging to the cluster managed by the other management computer 106 using the slave table, it behaves as a “slave management computer”. That is, although it is one management computer 106, it behaves as many types of management computers as many as the number of computers 108 under its control.

  The master table is generated by the management computer 106 that behaves as a master management computer, and the slave table is generated by the management computer 106 that behaves as a slave management computer. The presence of a slave table means that a corresponding master table exists in some management computer 106.

Here, specific examples of contents of the master table and the slave table in the present embodiment will be described. FIG. 6A shows an example of the contents of the master table 2161, and FIG. 6B shows an example of the contents of the slave table 2162.
The master table 2161 includes a cluster ID (ID means identification information, the same shall apply hereinafter) 2361, cluster size 2362, computer list 2363, cluster status 2364, maximum number of computers 2365, minimum number of computers 2366, total number of computers 2367, The value of the cluster connection ratio 2368 (recorded value in the field on the right side of FIG. 6A) is recorded by the master management computer as a component (value) of the cluster.

The cluster ID 2361 is a unique ID assigned to the formed cluster. As long as the cluster continues to exist, it may be unique information. For example, a set of the ID of the master management computer and the ID of the computer 108 to be included in the cluster first can be used as the cluster ID.
The cluster size 2362 is the total number of computers 108 newly or added to be included in the cluster. This can be specified by the master management computer measuring the number of computers in the cluster.

The computer list 2363 is a list of identification information of the computers 108 that are to be included in the cluster.
The cluster status 2364 is information representing the current state of the cluster. In this embodiment, three types of states “idle”, “run”, and “wait” are used as the cluster status. “Idle” is a state in which the cluster is not executing a job. When the cluster is in this state, the computer 108 can be added to the cluster. “Run” is a state in which the cluster is executing a job. “Wait” is a state in which the number of held computers 108 reaches the maximum number of computers. In this state, the computer 108 cannot be added to the cluster.

The maximum computer number 2365 is the maximum number of computers that can be held by the cluster, and is defined by the user or defined as a constant held by the system. In the illustrated example, it is indicated that the cluster can hold a maximum of 200 computers.
The minimum number of computers 2366 is the minimum number of computers that must be held as a cluster, and is defined by the user or defined as a constant held by the network system 101. In the illustrated example, a cluster is formed by three computers 108.
The total computer number 2367 is a total value of the computers 108 that may be included in the cluster, and can be obtained from an expected value based on statistics. In the illustrated example, it is shown that 48 computers 108 may be included.

  The connection ratio 2368 is information that is an example of additional ease information. In this embodiment, for example, when there is a computer 108 (candidate computer) as an additional candidate, it is added to the cluster with a probability of 20%, or added as in the required cluster size / total number of computers. A numerical value that is determined by a probability value or the like is used as a connection ratio. The higher this value is, the easier it is for the computer 108 to be added to the cluster and the easier the cluster grows. In the illustrated example, the connection ratio (= 0.48) when the current cluster size is 23 and the total number of computers is 48 is shown.

The connection ratio 2368 can be a fixed value or a variable value. What was set to a constant value can be replaced later with a variable value, and vice versa. This may be automatically set as a default value when forming a cluster, or an application program for causing a cluster to execute a job may be set in the connection ratio 2368 of the master table 2161 as one of the parameters. Anyway.
As shown in FIG. 5, the distribution ratio when the number of computers 108 included in the cluster is plotted on the horizontal axis and the number of clusters or the frequency of clustering is plotted on the vertical axis is determined by using this connection ratio. The place which depends on is big. For example, when trying to obtain a normal distribution as shown in FIG. 5A, the connection ratio is set to a constant value. Thereby, the above distribution becomes a normal distribution centered on the representative value. The connection ratio may be changed as the computer is added and the cluster size is increased. For example, the connection ratio may be increased every time one or several computers are added. In this way, a power distribution as shown in FIG. 5B can be easily obtained.
Note that the manner of holding the above-described components in the master table 2161 is merely an example, and can be increased or decreased as appropriate.

In the slave table 2162, a cluster ID 2461 and a computer id 2462 are recorded. The cluster ID 2461 is an ID for identifying a cluster (the same as the cluster ID 2361 recorded in the master table 2161). A computer id 2462 is an ID for identifying the computer 108 included in the cluster.
If only the cluster ID 2461 and the computer id 2462 are known, the slave table 2162 is linked to the master table 2161 that makes the cluster ID 2461 identical. Therefore, the component (value) recorded in the slave table 2162 is more than the master table 2161. It is simple.
Note that the way of holding the above components (values) in the slave table 2162 is also an example, and can be increased or decreased as appropriate.

<Operational form>
Next, an example of an operation mode of distributed computing by the network system 101 of this embodiment will be described.
In the distributed computing according to the present embodiment, as shown in FIG. 7, a job is distributedly executed while repeating a three-stage cluster state change cycle. The state change in the first stage is caused by the formation and growth of a new cluster (step S1). The state change in the second stage occurs when a job requested by an application program or the like is input (step S2). The state change in the third stage occurs due to the disappearance of the cluster after the job is executed (step S3). When the cluster disappears, all the computers 108 that belonged to the cluster until then become candidate computers (nodes) that can newly belong to another cluster at that time.

  The first stage state change in this cycle occurs regardless of the presence or absence of a job. Then, a certain management computer 106 becomes a master management computer, forms a cluster, and grows in cooperation with the slave management computer. How much the cluster grows depends on the operating state and connection ratio of each computer 108. As the operating state of the computer 108 is “free” and the connection ratio is higher, it becomes easier to connect to another computer 108. Hereinafter, the state of the above-described three-stage state change will be described in detail.

<First stage state change>
[Formation and growth of new clusters 1]
FIG. 7 is a procedure explanatory diagram showing the process of cluster formation and growth by the master management computer (the cluster management unit 216 of the management computer 106 that operates in this manner).
The master management computer checks the operating state of the computer 108 under its control. Specifically, the computer cluster status recorded in the status table held by each computer 108 is checked (step S101). Then, one cluster is formed by the computers 108 that are not operating, that is, several computers 108 whose computer cluster status is “free” (step S102). The number of computers included in the initially formed cluster is determined by, for example, prior settings. Normally, some of the computers 108 under their own group are left and clustered. The remaining computer 108 is set in a state where it can be added to other clusters as much as possible in order to facilitate the growth of other clusters. This state is shown in FIG.
In the example of FIG. 12A, one cluster C11 is formed by three computers. At this time, the master management computer generates one master table and sets the components (values) of this master table (step S104). If the master management computer is to manage the status table of each computer 108, the computer cluster status of all computers 108 included in this cluster is updated from “free” to “clustered” (step S104). ).

Thereafter, the master management computer determines whether there is no candidate computer to be added to the cluster formed by itself among the computers 108 under the randomly selected other management computer 106, that is, the computer cluster status is “free”. It is searched by inquiring whether there is any other computer 108 that is "" (step S105). The search is performed, for example, by randomly selecting several other management computers 106 having a short logical distance starting from themselves and sequentially expanding the range. If there is a candidate computer (step S106: Yes), the connection ratio of the master table is checked, the candidate computer is added to the cluster based on the connection ratio, and information about the candidate computer is recorded in the master table (step S106). S107). Thereafter, the computer cluster status of the added candidate computer is updated from “free” to “clustered” (step S108).
The management computer 106 that manages the added candidate computer becomes a slave management computer for the candidate computer 108 and generates a slave table. The state when the candidate computer exists is shown in FIG. In the example of FIG. 12B, two computers 108 have been added as candidate computers, and have grown into a cluster C14 that includes five computers 108. The master management computer adds information on the added new computer 108 to the master table.
In step S106, when there is no candidate computer (step S106: No), the process returns to step S105.

[Formation and growth of new clusters 2]
FIG. 9 is a procedure explanatory diagram showing a cluster growth process by the management computer 106 that does not become the master management computer.
As shown in FIG. 8, the growth of the cluster is basically realized by the master management computer acting actively, but the management computer 106 that is not the master management computer accesses the master management computer. And you can do it.
In other words, the management computer 106 having a computer 108 that is not clustered searches for other clusters to which the computer 108 can be added by, for example, inquiring to another management computer 106 selected at random (step S100). S201). Then, based on the connection ratio of the destination cluster, it is determined whether or not to add the computer 108 belonging to the cluster to the cluster (step S202). If not added, the process returns to step S201 (step S203: No).
When adding, since it becomes a slave management computer with respect to the cluster of a connection destination about the computer 108, the slave table about the computer 108 is produced | generated and each component (value) is set to this slave table (step S204). Then, the computer cluster status of the computer 108 to be added is updated from “free” to “clustered” (step S205). Further, the master management computer of the cluster to be added is updated with the master table (step S206). This update may be performed by accessing the master management computer from the slave management computer.

<Second stage state change>
[Submit Job]
The state change in the second stage starts from an action by any management computer 106 to which a job has been submitted by a request from an application or the like. When executing a job, the amount of calculation and the time required for executing the job, for example, the number of computers 108 are parameters. This parameter may be automatically generated by the processor of the management computer 106 based on the size of the input job, or may be appropriately given by a user who uses an application or the like. You may make it use the parameter which an application has.

  FIG. 9 is an explanatory diagram of a procedure for using a cluster by the management computer 106 to which a job has been submitted. The management computer 106 searches for a master management computer forming a cluster of a necessary size by referring to the cluster list 208 or inquiring another management computer based on the above calculation amount (step). S301). If found, the master management computer is requested to execute the job (step S302: Yes, S303). The job execution request is performed by, for example, sending a packet including a job, a program and data necessary for the execution, and a designated address for sending the execution result to the master management computer.

The master management computer updates the computer cluster status of each computer 108 belonging to the cluster executing the job from “idol” or “wait” to “run”, and then performs distributed processing by each computer 108. Start the job execution. FIG. 12C shows a state where the master management computer that manages the cluster C14 grown as shown in FIG. 12B is executing a job. The job execution result is sent from the master management computer to the requesting management computer 106. The management computer 106 that has received the execution result transmits the execution result to the job submission source (step S304: Yes, S305).
When a job is directly submitted to the master management computer that manages a cluster of a required size, the above steps S301 to S304 are processing procedures performed by the job. In step S302, if it is determined that a job needs to be executed across a plurality of clusters, for example, if it is determined that image processing and audio processing need to be executed in separate clusters, Each job is requested to be executed by the master management computer that manages the job.

<3rd stage state change>
[Disappearance of cluster]
The disappearance of the cluster is performed by the master management computer that manages the cluster that has executed the job. FIG. 11 is a processing procedure diagram in the master management computer when this cluster is extinguished.
When the job ends (step S401: Yes), the master management computer updates the computer cluster status for all computers 108 included in the cluster that has executed the job so far to “free” (step SS402). Also, the component (value) of the master table of the cluster is cleared. At the same time, the components (values) of the slave table are cleared through the slave management computer under the control of the computer belonging to the cluster. That is, all computers are returned to the state before cluster formation (step S403). As a result, all the computers 108 belonging to the cluster are in a non-operating state and can be clustered. One computer 108 is immediately clustered, and another computer 108 is added to another cluster. It will be in a state to be. FIG. 12 (d) shows this state.

As described above, according to this embodiment, in a network system with many uncertainties, which is expected to vary in size, type, or number of jobs to be input and changes from moment to moment. Even if there is a plurality of management computers 106, each of the computers 108 under their control acts as a master management computer or a slave management computer while forming a cluster of an optimal size and growing it. Therefore, job execution can be assigned to a cluster of an appropriate size, and the job can be executed effectively. In addition, since a large-scale process to a very small process can be handled in a uniform and uniform procedure, distributed computing with high flexibility can be realized. In particular, when the process size distribution is a power distribution, the additional ease information can be changed and the cluster size can also be set to a power distribution, so the process can be processed most effectively. , Increase the use efficiency of computer resources.
<Modification>
In the above example, for the sake of convenience, the description has been given on the assumption that the status table is held by each computer 108 and the management table 206 (master table / slave table) is held by the management computer 106. The computer 108 and the management computer 106 only need to exist in a memory area that can access them. For example, the management computer 106 may have a status table of the computer 108 under its control, or the management table 206 may be stored in a batch on a server or the like accessible by the table management unit 212 of the management computer 106. May be.
In the above example, the slave management computer decides whether or not to add a candidate computer under the slave management computer to another cluster, but whether or not to add a candidate computer to the cluster. All may be determined by the master management computer.

<Example in case of not using a management computer>
In the above description, an example of the network system 101 including a set of management computers 106 under the control of a plurality of computers 108 has been shown. However, in the present invention, each of the plurality of computers 108 is directly connected without using the management computer 106. Implementation as a network system that can join and leave is also possible.

  When implemented as such a network system, the connection ratio is also recorded in the status table of each computer 108, and the status table is also managed by itself. For example, when a self is added to a cluster formed by another computer 108, the self computer cluster status is updated to “clustered”. Then, when the execution of the job by the cluster is completed, the own computer cluster status is returned to “free”. Further, any of the computers 108 has the same function as the management computer 106 described above. That is, when a cluster is formed by itself, a master table that records the contents of the cluster is generated.

  The computer 108 that formed the cluster searches for other additional candidate computers whose latest computer cluster status is “free” in the status table held by the other computer, and the candidate identified by this is searched. Decide whether to add computers based on the cluster connection ratio. The computer 108 that has formed the cluster causes the cluster that has grown in this way to execute a job, and after the execution is completed, the cluster disappears.

<Embodiment as a multiprocessor / multi-core processor>
The present invention can also be implemented as a network system with a plurality of processors or multi-core processors. In this case, as illustrated in FIG. 13A, the internal bus functions as the computer network 104 described above, and any one of the processors or multi-core processors connected to the internal bus or Some may be operated as the management computer 106 described above, and the remaining processors and the like may be operated as the computer 108 described above.
The individual multi-core processors themselves illustrated in FIG. 13B can also be implemented as a network system. In this case, the operation of the management computer 106 described above is realized by any one or several cores (processor cores) connected to the internal bus, the input / output control unit (I / O), and the cache memory. The remaining cores may be operated as the computer 108 described above.
[Other embodiments]

The network system of the present invention may be configured such that when a computer 108 of a cluster accesses another cluster, the computer 108 is added to another cluster associated with the other cluster. Such a mode of the embodiment is referred to as a “Duplication Mode”. This “duplication mode” is described in detail in the following document.
Title: Growing networks with local rules: preferential attachment, clustering hierarchy and degree correlations Author: Alexei V´azquez Journal title: Phys Rev E Stat Nonlin Soft Matter Phys. 2003 May; 67 (5 Pt2): 056104. Epub 2003 May 7 .

  In the duplication mode, in place of the connection ratio 2368 described above, a cluster list that lists identification information (connection address, etc.) of other clusters associated with computers belonging to the accessed cluster is used. The cluster list is provided, for example, in the master table.

FIG. 14 shows an example of the contents of the master table in this case. This master table 2163 includes a cluster ID 2361, a cluster size 1262, a computer list 2363, a cluster status 2364, a maximum number of computers 2365, a minimum number of computers 2366 and the same contents as the master table 2161 shown in FIG. It has a list 2369. Each time a new cluster is associated with a cluster managed by the master management computer, the master management computer lists the identification information of the new cluster in the cluster list 2369. “Relation” in this embodiment means that a computer belonging to a certain cluster and a computer belonging to another cluster are logically connected (contacted), and can perform cooperation processing or the like.
The master management computer reads the identification information of one cluster listed in the cluster list 2369 in response to a request from the slave management computer, for example, and notifies the slave management computer of it. For applications where security is not important, the cluster list 2369 may be browsed by, for example, a slave management computer.

  The order of reading from the cluster list 2369 may be random or a predetermined order. In the former case, the order of listing the cluster identification information in the cluster list 2369 may be arbitrary. On the other hand, in the latter case, the ranking is weighted. For example, other clusters having higher relevance to the processing contents of the cluster managed by the master management computer are read out first. The relevance in this case can be determined by, for example, the number of accesses between clusters, the number of times of communication between clusters within a predetermined time, the degree of relevance of distributed processing between clusters (audio processing that should be synchronized with image processing, etc.), etc. it can.

For example, the operation mode in the duplication mode is as follows.
Assume that the network system is in an initial state as shown in the upper part of FIG. In the example of FIG. 15, a plurality of clusters C6 to C11 are already formed. One or a plurality of computers (the computer 108 described above, the same applies hereinafter) belong to each of the clusters C6 to C11.

Assume that a new cluster C20 is formed in the network system in this state as shown in the middle part of FIG. The management computer that forms the cluster C20 manages not only the computers belonging to the cluster C20 but also several computers for facilitating the growth of other clusters, that is, candidate computers (not shown) as their own companies. It shall be. As described above, when this candidate computer is provided to another cluster, the management computer that manages the candidate computer becomes a slave management computer.
The management computer that manages the cluster C20 searches for the cluster C7 from the clusters C6 to C11. Then, the master management computer that manages the cluster C7 is inquired, and information on the association between the cluster C7 and another cluster in the master table 2163 is obtained. In the middle of FIG. 15, the cluster C7 is related only to the cluster C11. Therefore, the management computer that manages the cluster C20 adds the candidate computer to the cluster C11 and associates the candidate computer with the cluster C20 (a computer belonging to the computer) (lower row in FIG. 15). Also, the contents of the slave table owned by itself are updated.
On the other hand, the master management computer that manages the cluster C11 adds a list of newly added computers (candidate computers before addition) to the master table 2163 to the computer list 2363 and identifies the cluster C20 in the cluster list 2369. Add information.

  The above is an example in which the cluster C20 and the cluster C11 (each computer) are associated with each other on a one-to-one basis. However, the association is not necessarily one-to-one, but one-to-N (N is a natural number of 2 or more). ). This association is not related to the connection between computers in the physical console.

For example, FIG. 16A shows a case where a management computer that manages a cluster C20 searches for an addition destination of a candidate computer under its control when a cluster C20 to which a computer newly added to the network system belongs is formed. Suppose that the cluster C11 is reached. In the cluster list of the cluster C11, in addition to the clusters C6 and C7 that are associated with the computers of the cluster C11 on a one-to-one basis, C8 and C10 that are associated with a plurality of computers of the cluster C11 are listed. Then, the candidate computer is added to one of the clusters C6, C7, C8, and C10. The same applies to the case of FIG.
In these cases, the recording information of the slave table and the master table is updated as in the example shown in FIG.

In the duplication mode, an operation mode in which the cluster list 2369 and the computer list 2363 are used together is also possible. Hereinafter, an example of the operation mode in this case will be described.
When a new cluster C20 that joins the network system is formed and the management computer that manages the cluster C20 searches for the addition destination of the candidate computer under its control, it reaches the cluster C11 until it reaches the cluster C11. It is the same as the example of b).
According to the computer list 2363 of the cluster C11, as shown in FIG. 17, computers # 1 to # 5 belong, and according to the cluster list 2369, the computers # 1 and # 3 are the cluster C10 and the computer # 4. Are associated with cluster C7 and computer # 5 are associated with cluster C6, respectively, and computer # 2 is not associated with anything. Although it is not always necessary to do so, if the computer list 2363 is read at random, when the computer # 1 is selected from the computer list 2363, the candidate computer is added to the cluster C10 related thereto. Added. When computer # 2 is selected, it is reselected because there is no cluster associated with it.

  In the duplication mode, it is possible to perform clustering according to the power distribution without setting the connection ratio 2368, so that there is an effect that the addition of the computer to the cluster becomes smoother.

  The present invention can be applied to a network system for facilitating clustering in which the number of nodes connecting computers is, for example, a normal distribution or a power distribution. In addition, the present invention can be applied to all distributed computing that can efficiently execute information processing in which the size, type, and number of jobs to be submitted vary and change from moment to moment.

1 is an overall view of a network system to which the present invention is applied. The figure which shows the general | schematic architecture of the management computer by this embodiment. Explanatory drawing of a large-scale information processing integrated body. The functional block diagram of the management computer by this embodiment. It is the actual measurement figure plotted with the number of links on the horizontal axis and the number of nodes with the number of links on the vertical axis, where (a) is an example of a random connection, and (b) is an example of a selective connection. It is. (A) is a master table, (b) is the figure which showed the example of the content of the slave table. FIG. 3 is a schematic procedure diagram of distributed computing according to the present embodiment. Explanatory drawing of the procedure of cluster formation and growth by a management computer. Explanatory drawing of the procedure of cluster formation and growth by a management computer. Explanatory diagram of job submission and execution by the management computer. Explanatory drawing of the process sequence at the time of the cluster extinction by a management computer. (A)-(d) is explanatory drawing which showed the process of formation of the cluster by this embodiment, growth, and extinction. (A) is a schematic diagram of a network system realized by a multiprocessor, (b) is a schematic diagram of a network system realized by a single multi-core processor. The figure which showed the example of the content of the master table used by duplication mode. The figure which showed the example of the operation | use form in duplication mode. (A), (b) is the figure which showed the other example of operation form in duplication mode. The figure which showed the other example of operation form in duplication mode.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 ... Network system 104 ... Computer network 106 ... Management computer 108 ... Computer (candidate computer)
206: Management table 208 ... Cluster list 212 ... Table management unit 216 ... Cluster management unit 236 ... Communication control unit 256 ... Job execution unit 315 ... Shared DRAM
1061 ... Storage device 1063 ... Communication device 1065 ... Semiconductor memory 1067 ... Processor 2161, 2163 ... Master table 2162 ... Slave table 2361 to 2369 ... Components (values) of the master table
2461-2462 ... Slave table components (values)
B11 ... Bus WO ... Large-scale information processing integrated body

Claims (27)

A network system in which a plurality of computers each capable of clustering with other computers can freely join and leave;
A first table that records clusterability information indicating whether each computer is in a clusterable state;
A second table for recording additional ease information indicating how easily a cluster already formed by one or more computers can add another computer;
Any one of the computers forms a cluster including self and other computers whose clusterability information in the first table indicates a clusterable state, and for all the computers included in the formed cluster. A cluster forming unit that updates the clusterability information to information indicating a non-clusterable state, and further records the additional ease information about the formed cluster in the second table;
The computer that has formed the cluster indicates whether or not to add the candidate computer to the cluster when there is a candidate computer to be added to the cluster in the additional ease information recorded in the second table. Having a cluster growth means to determine based on; network system. The computer that formed the cluster disappears when the execution of the job by the cluster is completed, and belongs to the additional ease information recorded in the second table and the disappeared cluster. Cluster extinction means for returning the clustering availability information recorded in the first table for all computers to the state before cluster formation;
The network system according to claim 1. The additional ease information is a numerical value obtained by quantifying the ease of addition, and the cluster growth means facilitates adding the candidate computer to the cluster as the numerical value increases.
The network system according to claim 1. The cluster growth means holds the numerical value at a constant value regardless of addition of candidate computers.
The network system according to claim 3. The cluster growth means is variable after recording the numerical value in the second table;
The network system according to claim 3. The cluster growth means changes the numerical value when the candidate computer is added to the cluster.
The network system according to claim 5. A network system in which a plurality of management computers each having one or a plurality of computers that can be clustered with other computers can freely join and leave;
A first table that records clusterability information indicating whether each computer is in a clusterable state;
A second table recording addability information indicating how easily a cluster already formed by one or more computers can add another computer;
At least one of the management computers includes a cluster including a computer under its own control and a computer under the control of another management computer including a computer in which the clusterability information in the first table indicates a clusterable state. Forming and updating the clustering availability information for all the computers included in the formed cluster with information representing a non-clusterable state, and a cluster forming means,
The management computer forming the cluster determines whether to add the candidate computer to the cluster when there is a candidate computer to be added based on the additional ease information recorded in the second table. Having cluster growth means to
Network system. The management computer that formed the cluster lost the cluster when the job execution by the cluster was completed, and belonged to the additional ease information recorded in the second table and the deleted cluster Cluster erasure means for returning the clusterability information recorded in the first table for all computers to the state before the cluster formation,
The network system according to claim 7. Any of the management computers has the second table of the maximum number of computers under its control,
The network system according to claim 8. The second table includes a master table that is generated for the first computer when the management computer that owns it forms a first cluster that includes the first computer under its control, and a management computer that owns the master table. Including at least one of the slave tables generated for the second computer when monitoring and controlling the operation of the second computer under its control, which is added to the second cluster formed by another management computer,
The additional ease information is recorded in the master table,
The management computer having the master table behaves as a master management computer that mainly performs information processing related to formation of the first cluster, change of the number of computers in the first cluster, and disappearance of the first cluster, and the slave table , The management computer that acts as a slave management computer for the second cluster,
The network system according to claim 9. The master management computer has search means for searching for candidate computers to be added to the first cluster by inquiring whether any of the management computers is in a clusterable state.
The network system according to claim 10. Any of the management computers determines whether to add a candidate computer belonging to the management computer to the first cluster based on the additional ease information about the first cluster formed by the master management computer.
The network system according to claim 10. The additional ease information is a numerical value quantifying the ease of adding a candidate computer to the first cluster,
The cluster growth means of the master management computer makes it easier to add the candidate computer to the first cluster as the numerical value increases.
The network system according to claim 10. The cluster growth means holds the numerical value at a constant value regardless of addition of candidate computers.
The network system according to claim 13. The cluster growth means makes the value variable after recording the numerical value in the second table.
The network system according to claim 13. The cluster growth means varies the numerical value when the candidate computer is added to the first cluster.
The network system according to claim 15. A network system in which a plurality of computers each capable of clustering with other computers can freely join and leave;
For each cluster already formed by one or more computers, including a table listing identification information of other clusters associated with the computers belonging to the cluster;
The computer that formed the cluster has the identification information listed in the table as to whether or not to add the candidate computer to another cluster associated with its own cluster when there is a candidate computer to be added. Having a cluster growth means to determine based on
Network system. The table is preferentially listed so that the candidate computer is added to a related cluster having a higher relevance to the processing executed by the own cluster.
The network system according to claim 17. Including a first table that records clusterability information indicating whether each computer state is clusterable.
Any one of the computers forms a cluster including self and other computers whose clusterability information in the first table indicates a clusterable state, and for all the computers included in the formed cluster. The cluster forming means for updating the clustering availability information to information indicating a non-clusterable state,
The network system according to claim 17. Each is a management computer that owns one or more computers that can be clustered with other computers;
Network connection means for connecting to a computer network that can freely join and leave with other management computers having similar functions;
A first table that records whether or not each computer can be clustered, and how much other clusters are added to the cluster already formed by one or more computers Table management means enabling access to a second table for recording additional ease information indicating whether or not
A cluster including a computer in which the clustering availability information in the first table represents a clusterable state among the computers under the control of the own computer and other management computers is formed and formed. Cluster forming means for updating the clusterability information on all computers included in the cluster to information indicating a non-clusterable state, and further recording the additional ease information on the cluster in the second table When;
And a cluster growth means for determining, based on the additional ease information recorded in the second table, whether to add the candidate computer to the cluster when there is a candidate computer to be added. Computer. When the execution of the job by the formed cluster is completed, the cluster is deleted, and the additional ease information recorded in the second table and the first computer for all computers belonging to the deleted cluster. Further comprising cluster erasure means for returning the clustering availability information recorded in the table to the state before the cluster formation,
The management computer according to claim 20. A cluster management method in a network system in which a plurality of computers each capable of clustering with other computers can freely join and leave;
Each computer records in the first table clusterability information indicating whether its own state is a state in which clustering is possible;
One of the computers forms a cluster including itself and another computer whose clusterability information in the first table indicates a clusterable state, and for all computers included in the formed cluster. Updating the clusterability information to information indicating a non-clusterable state, and further recording in the second table additional ease information indicating how easy the cluster can add another computer;
Determining whether to add the candidate computer to the cluster based on the additional ease information recorded in the second table when there is a candidate computer as an addition candidate. . When the computer that formed the cluster ends execution of the job by the formed cluster, the computer disappears and belongs to the additional ease information recorded in the second table and the disappeared cluster. Returning the clusterability information recorded in the first table for all computers to a state before cluster formation.
The cluster management method according to claim 22. A cluster management method in a network system in which a plurality of computers each capable of clustering with other computers can freely join and leave;
A computer that has already formed a cluster with one or a plurality of computers lists the identification information of other clusters with which the computers belonging to the cluster are associated in a predetermined table;
The identification information recorded in the table indicates whether or not the computer forming the cluster adds the candidate computer to another cluster associated with the cluster when there is a candidate computer to be added. Determining based on: a cluster management method. Each computer can be read and executed by any computer in the network system in which a plurality of computers that can be clustered with other computers can freely join and leave, respectively.
A first table that records whether or not each computer can be clustered, and how much other clusters are added to the cluster already formed by one or more computers Table management means enabling access to a second table for recording additional ease information indicating whether it is easy to do;
The clustering availability information of the first table forms a cluster including self and other computers indicating a clusterable state, and the cluster of the first table for all computers included in the formed cluster Cluster forming means for updating the information indicating whether or not to enable clustering to information indicating a non-clusterable state, and further recording the additional ease information for the cluster in the second table;
For functioning as cluster growth means for determining whether to add a candidate computer to the cluster based on the additional ease information recorded in the second table when there is a candidate computer to be added. Computer program. The management computer is read and executed by a management computer that owns one or more computers each capable of clustering with another computer;
Network connection means for connecting to a computer network that can freely join and leave with other management computers having similar functions;
A first table that records whether or not each computer can be clustered, and how much other clusters are added to the cluster already formed by one or more computers Table management means enabling access to a second table for recording additional ease information indicating whether it is easy to do;
A cluster including a computer in which the clustering availability information in the first table represents a clusterable state among the computers under the control of the own computer and other management computers is formed and formed. Cluster forming means for updating the clusterability information on all computers included in the cluster to information indicating a non-clusterable state, and further recording the additional ease information on the cluster in the second table ;
For functioning as cluster growth means for determining whether to add a candidate computer to the cluster based on the additional ease information recorded in the second table when there is a candidate computer to be added. Computer program. A plurality of computers each capable of clustering with other computers are read and executed by any computer in the network system that can freely join and leave;
Cluster management means for listing, in a predetermined table, identification information of other clusters associated with computers belonging to the cluster when a cluster is already formed by one or a plurality of computers;
Cluster growth means for determining, based on the identification information listed in the table, whether or not to add a candidate computer to another cluster associated with its own cluster when there is a candidate computer to be added A computer program for functioning as;

Images