JP2000242620A - Data duplication system and storage medium recording data duplication program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform application in the case of requesting more than a certain degree of the newness of obtained data. SOLUTION: This data duplication system is provided with plural duplication nodes for managing data, a client node for requesting the updating and read of the data and a front end node for mediating a request from the client node to the duplication node. The duplication nodes constitute one or more groups connected by a logical tree structure for indicating an updating route, and when an updating request is propagated, propagate the updating request to all the nodes other than the node which propagates the updating request. The front end node requests the updating to one duplication node (updating start node 3) in the respective groups for the updating request from the client node, mediates a read request to the plural duplication nodes (read nodes 4), calculates the data of how much recentness can be acquired or the like and returns the obtained data and a scale for indicating how recent the data are to the client node.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data replication system and a recording medium on which a data replication program is recorded, and more particularly to a decision support system used on a network and a directory service operating in cooperation with the decision support system. The present invention relates to a data duplication system and a recording medium that records a data duplication program.

[0002]

2. Description of the Related Art Data management of directory services, mobile terminal information, network management information, decision support systems and the like has become important due to the expansion of distributed environments and the number of nodes accompanying the recent development of networks. In these, strict consistency control of transactions and their constituent operations, such as online processing of banks, is less important, but the importance of response time and throughput for each transaction is higher.

For data having such characteristics, (1)
If the application that issues the transaction knows the processing order and dependencies required to maintain consistency, (2) a data error can be detected based on the transaction response and the operation result of the application using the response result. In such a case, (3) there is no collision between most transactions, and even if there is a collision, it is often possible to undo by undo. Therefore, Lazy replication using such characteristics and the data replication function using asynchronous update implemented in many current database systems improve response time, throughput, and control weak consistency. Has been done.

In a decision support system or the like, when data replication is performed between processes holding the same data, other databases have already been updated, but their own databases have not been updated yet. Data may be present. The number of data updated per unit time is the
s), which indicates the freshness of the data as viewed from the system. This shows the freshness of the acquired data from the client's point of view. In future decision support systems and the like, the newness of such actually acquired data is considered to be an important decision material.

[0005] From the above viewpoints, the following techniques are conventionally available. (1) Lazy replicatio
n) (Reference: R. Ladin, “Lazy replication: Exploiting th
e semantics of distributed services ”, Proceedings
of the Workshop on Management of Replicated Data, p
p.31-34, 1990, Literature: R. Ladin, “Providing high availa
bility using lazy replication ”, ACM Transaction Co
mputer Systems, Vol. 10, pp. 360-391, November 1992) This lazy replication provides casual operation (casual o
peration, forced operation
ration) and immediate operations (imme
This is a data duplication method that provides a date operation. The update request is propagated to the replication node asynchronously.

(2) Pbcast (Reference: K. Birman, “BUILDING SECURE ANDRELIABLE NETWORK”
APPLICATI0NS ”, MANNING Publisher, 1996) This peacast is a message propagation method, and the propagation method when it is used for propagation of an update request is as follows. fu
The number of replication nodes is randomly selected by the number determined by a constant called "nout", and an update request is transmitted. A node that has already received the same update request does not propagate the update request to other duplicate nodes.

[0007]

However, when such a conventional technique is used, the following problems occur. First, since the freshness of the data provided from the replica node is determined systematically, when a client wants to acquire data of a certain newness, it cannot acquire data satisfying the request. When the client or a person operating the client determines an operation based on the obtained data, the judgment support system or the like sometimes wants to make the freshness of the acquired data a factor of the judgment. However, conventionally, there is no method for providing the newness of the data provided from the replica node, and the client cannot know the newness of the acquired data.

The present invention is intended to solve such a problem, and records a data duplication system and a data duplication program which can be applied when it is required that acquired data has a certain degree of newness. It is an object of the present invention to provide a recording medium that has been used.

[0009]

In order to achieve the above object, a data replication system according to the present invention has a plurality of computers connected to a communication network, and these computers have a process for managing data. Are operating one or more times, and the contents of the data managed by each of the above processes are all the same, and the process of managing the data changes the data of its own process and receives other The present invention relates to a data replication system in which this update request is propagated and data is updated one after another.

[0010] A plurality of replication nodes for managing the data, a client node for requesting updating and reading of the data, and a front-end mediating a request from the client node to the replication node.
And a node. The replica nodes constitute one or more groups connected by a logical tree structure indicating an update path, and when the update request is propagated, propagates the update request among the adjacent duplicate nodes on the tree structure. The update request is propagated to all nodes other than the node that has performed the update.

[0011] In response to the update request from the client node, the front-end node issues an update request to one copy node (hereinafter, referred to as an update start node) in each of the groups. In response to the read request, the read request is mediated to a plurality of the above-mentioned duplicate nodes (hereinafter, referred to as read nodes)
The latest data among the plurality of read data is returned to the client node to calculate how much recent data can be obtained, manage the calculation result, and respond to the request from the client node. In response, a read request is made by selecting a combination of read nodes satisfying this request, and the obtained data and a measure indicating how recent the data are are returned to the client node.

The scale indicating how recent the acquired data is the latest data is defined as the latest degree. When a predetermined probability p reflects the update from the present to the time T before, , Probability p may be defined as T.

Each of the duplicated nodes manages a statistical distribution of update delays when propagating through a logical link connected to the node, and manages a minimum subtree including the readout node (hereinafter referred to as a readout subtree). Of the read subtree, select a node closest to the update start node on the tree structure (hereinafter, referred to as a shortest node), and select a node from the update start node to the shortest node. In a subtree composed of a path and a read subtree, when the update start node is the root, a node that is a leaf of the read subtree uses the statistical distribution of update delay of a link connected to the node as a parent node. The replica node other than the leaf node receives the statistical distribution from all of the child nodes, and then uses this statistical distribution to After calculating the minimum statistical distribution of the update delay from the read node that is a descendant of the replica node to the replica node, the statistical distribution is transmitted to the parent node, and the shortest node transmits the calculated statistical distribution or the latest degree to the parent node. The front-end node that has transmitted to the front-end node and received the latest degree or the update delay statistical distribution for all update start nodes, based on the latest degree or statistical distribution, The latest degree may be calculated.

Further, when the shortest node is a read subtree, the read nodes in which the read nodes exist in the ancestors are deleted, and the read nodes in which the read nodes exist in the ancestors are deleted. A read subtree may be used.

On the other hand, in a recording medium on which a data duplication program according to the present invention is recorded, a plurality of computers are connected to a communication network, and one or more processes for managing data operate on these computers. The contents of the data managed by each process are all the same, and the process managing the data, upon receiving the data update request,
The present invention relates to a data replication system in which data of its own process is changed and this update request is propagated to other processes to update data one after another.

A plurality of replication nodes for managing the data, a client node for requesting updating and reading of the data, and a front-end for mediating a request from the client node to the replication node
And a node. The replica nodes constitute one or more groups connected by a logical tree structure indicating an update path, and when the update request is propagated, propagates the update request among the adjacent duplicate nodes on the tree structure. The update request is propagated to all nodes other than the node that has performed the update.

In response to the update request from the client node, the front-end node issues an update request to one of the duplicate nodes in each of the groups (hereinafter, referred to as an update start node). In response to the read request, the read request is mediated to a plurality of the above-mentioned duplicate nodes (hereinafter, referred to as read nodes),
The latest data among the plurality of read data is returned to the client node to calculate how much recent data can be obtained, manage the calculation result, and respond to the request from the client node. In response, a read request is made by selecting a combination of read nodes satisfying this request, and the obtained data and a measure indicating how recent the data are are returned to the client node.

The scale indicating how recent the acquired data is is defined as the latest degree. When the update from the present time to the time T before is reflected with a predetermined probability p. , Probability p may be defined as T.

Each of the replica nodes manages a statistical distribution of update delay when propagating through a logical link connected to the node, and manages a minimum subtree including the read node (hereinafter, a read subtree). Of the read subtree, select a node closest to the update start node on the tree structure (hereinafter, referred to as a shortest node), and select a node from the update start node to the shortest node. In a subtree composed of a path and a read subtree, when the update start node is the root, a node that is a leaf of the read subtree uses the statistical distribution of update delay of a link connected to the node as a parent node. The replica node other than the leaf node receives the statistical distribution from all of the child nodes, and then uses this statistical distribution to After calculating the minimum statistical distribution of the update delay from the read node that is a descendant of the replica node to the replica node, the statistical distribution is transmitted to the parent node, and the shortest node transmits the calculated statistical distribution or the latest degree to the parent node. The front-end node that has transmitted to the front-end node and received the latest degree or the update delay statistical distribution for all update start nodes, based on the latest degree or statistical distribution, The latest degree may be calculated.

When the shortest node is a read subtree, the read nodes whose read nodes exist in the ancestors are deleted, and the read nodes are completed when the read nodes whose read nodes exist in the ancestors disappear. A read subtree may be used.

With this configuration, the client node can acquire data satisfying the required freshness from the front-end node. Also,
The acquired data can be calculated by the front-end node, and the freshness can be confirmed from the latest degree provided to the client.

[0022]

Next, one embodiment of the present invention will be described with reference to the drawings.

FIG. 1 shows a network according to the present invention and a computer connected to the network. As shown in FIG. 1, a computer 1 is connected to a network 2 via an interface, and stores a central processing unit (CPU) for performing various controls such as data duplication, data, a communication control program according to the present invention, and the like. It has a random access memory (RAM) and a hard disk drive (HDD) to hold. The data and the like are distributed and stored in a plurality of computers 1, and various processes described below operate on each computer 1.

FIG. 2 schematically shows a plurality of processes distributed on a network. As shown in the figure, each process is classified into three types according to its operation. A replication node, which is a process for managing data, a client node, which is a process for requesting data update and data read, and a request (update request) to a plurality of replication nodes via a request from the client node , A read request, etc.).

A plurality of replication nodes constitute one group by gathering a plurality of nodes, and one or more groups exist on the network. FIG. 2 shows a case where there are three groups. The duplicate nodes included in each group are connected by a logical tree structure indicating an update path. The replicated node that has received the update request propagates the update request to all nodes other than the node that issued the update request among the adjacent nodes on the tree structure.

In response to an update request from a client node, the front-end node issues an update request to one copy node (hereinafter, referred to as an update start node 3) in each group, and responds to a plurality of read requests with a plurality of update nodes. A read request is issued to the replica node, and the time stamp (information attached at the end of the update) attached to the data read from the replicated node (hereinafter referred to as read node 4) that has performed a plurality of reads is added. Based on this, reply the latest data to the client node. Further, it calculates how recent data can be acquired, manages the result as information, and responds to a request from a client node by selecting a combination of read nodes 4 satisfying the request and issuing a read request. Then, the client node returns the acquired data and a measure indicating how recent the acquired data is (hereinafter, referred to as the latest degree) to the client node.

The front-end node manages the latest information when a read operation is performed on a plurality of read nodes 4, and manages a set of nodes satisfying the latest request from the client node. To perform a read operation. Further, the front-end node cooperates with the duplication nodes, and based on the statistical distribution of the partial update delays managed by each duplication node, updates the update delay generated by the route propagating from the update start node 3. Calculate the statistical distribution. Further, the front-end node collects the statistical distribution or the latest of all the update start nodes, and sets the maximum value of the obtained latest as the latest in the group (replica node). Based on the statistical distribution of update delays of nodes that start multiple updates, calculate the ratio of the number of update requests started by each node to the number of update requests of all update start nodes, and take a weighted average of the update delay distribution To calculate the latest degree.

Next, the details of this embodiment will be described.

[Definition of Latest Level] The latest level is a scale indicating how recent acquired data is. That is, when the update from the present to the time T before is reflected by the probability p, the latest degree of the probability p is defined as T.

[Operation of Each Node in Update and Read Operations] The operation of each node is as follows. (1) Client node "read request" ... Specifies the data object to be read and the required latest degree to the front-end node. "Update request" ... A data object to be updated and a value of the updated data object are specified for the front-end node.

(2) Duplicate node "Data update operation": Receives an update request from the front-end node or another duplicate node, updates the data of its own node, and then updates the update request among the adjacent duplicate nodes. Is transmitted to a node other than the replica node that has transmitted the update request. “Read operation”… Responds to a read operation from the front end node to the front end node with data managed by the own node and a time stamp assigned to the data. "Latestness calculation" ... The latestness is calculated in cooperation with the front-end node. Details will be described later.

(3) Front-end node "Data update operation" ... An update request from a client node is transmitted to one replication node included in each of the grouped replication nodes. "Read operation" ... manages, as information, a set of a plurality of read nodes and the latest degree when reading from the set. A read node set that satisfies the latest degree specified in the read request from the client node is selected,
A read request is transmitted to each of the read nodes included in this set, and the data with the latest time stamp given to the received data, the time stamp of the data,
Responds to the client node with the latestness. "Latestness calculation" ... The latestness is calculated in cooperation with the replica node. Details will be described later.

[Summary of Latest Level Calculation] Next, the calculation of the latest level will be described. The outline of the calculation of the latest level is as follows. 1. An update delay statistical distribution is assigned to each link. 2. The read subtree is constructed and the shortest node is set. 3. Nodes irrelevant to the improvement of the latest degree are deleted from the read subtree. 4. Calculate the statistical distribution of update delays. 5. Increase the efficiency of calculating the statistical distribution of update delays. These details will be described.

[1. Assignment of Update Delay Statistical Distribution to Each Link] When an update request propagates along a route, the causes of delay until the update propagates to each replica node are classified into the following four factors. “D _in ”: delay time from when an update request arrives at a node to when it is queued and processed. “D _proc ”: processing time of the update request in the node “d _out ”: the delay time from the time of entering the queue to the time of sending the update request as an update request message to the next node “d _trans ” …… Delay time to reach the next node after propagating on the communication link

Now, consider a tree structure composed of a route from the update start node to the readout node with the update start node as a root. When an update request propagates through this tree structure, an update request that has arrived at a certain replicated node is propagated to its child nodes. Then, factors of the update delay of the plurality of child nodes that have received the update request include those that are independent and those that are not independent among the child nodes. Thus, in the present embodiment, when calculating the statistical distribution of a plurality of read nodes, the statistical distribution of delay is assigned to each link. To a link incident on a certain replicated node, among the delay factors, a factor that is not independent of a sibling node is assigned. For example, when each replica node transmits an update request to a child node at the same time and when transmission timing to each child node is not simultaneous, the following is used as the delay time δ of each link.

Δ = d _trans + d _in ^(b) + d _proc ^(b) + d _out ^(b) (1)

Δ = d _out ^(a) + d _trans + d _in ^(b) + d _proc ^(b) (2)

In the above description, node a is a parent and its child is a node b, and d _* ^(a) and d _* ^(b) indicate delay times at the nodes a and b, respectively. Since these delays δ depend on the propagation direction, they need to be separately managed according to the propagation direction.

[2. Construction of Read Subtree and Setting of Shortest Node] The front-end node constructs a minimum tree including the read node (hereinafter, referred to as a read subtree). The setting of the shortest node is performed by setting a node that satisfies one of the following conditions among the nodes included in the read subtree as the shortest node. Condition 1 The own node is the update start node. Condition 2 On the path from the own node to the update start node, the adjacent node is not a read subtree.

[3. Delete nodes that are not relevant to
Exclusion] Now, the update message propagating on the tree
Triggers a data update event in the code. This update
During the occurrence of the vent, the connection
There will be a partial order relation based on them. Two update a
Vent e_iAnd e _jThe relationship of the occurrence timing
e_iAfter e_jThe relationship that occurs

[0041]

(Equation 1)

Is defined as Assuming that the parent of the update event is e _p and the child is e _c between the nodes having a parent-child relationship,

[0043]

(Equation 2)

Is as follows. Obviously, this relation can be said to be a transitive relation. Therefore, there is a certain node n _i and node n _j, the node n _i is assumed to be an ancestor of node n _j,

[0045]

(Equation 3)

Is as follows. In general, let E _i and E _{j denote the} events whose update delays for a certain node n _i and a node n _j are d _i and d _j , respectively.

[0047]

(Equation 4)

Is as follows. Now, since node n _i is an ancestor of node n _j and when d _i <d _j , P (E _i | E _j ) = 1, so

[0049]

(Equation 5)

Thus, the node n _j does not contribute to the improvement of the latest level. When d _i ≧ d _j , P (E _i | E _j ) = 0, and

[0051]

(Equation 6)

Is as follows. Therefore, in the calculation of the update delay statistical distribution at a plurality of read nodes, when the descendants of the read node are included in the asynchronous update tree, all the descendants can be deleted and calculated.

[4. Calculation of Update Delay Statistical Distribution] When the above-mentioned nodes irrelevant to the latest degree are deleted, an update tree composed of a path from the update start node to the shortest node and a read subtree starting from the update start node shown in FIG. In the above, all the read nodes are leaves. Where node v
_When the update generated from _r , propagates to the read node, the probability density function is obtained as the statistical distribution of the minimum value of the propagation time. Node v _n sent from among the update delay until propagated to the read node descendants of node v _n,
The minimum update delay probability density function is _denoted by △ _n, and △ _i1i2 …
_{Find ij} . _v i1i2 ... _{i j} and _v i1i2 ... If the probability density function of the link assigned delay in the _{j + 1} and δ _i1i2 ... _{ij + 1,} consisting of the links and nodes _v i1i2 ... _{j + 1} and its progeny The probability density function of the update delay in the subtree is

[0054]

(Equation 7)

Is as follows. The nodes v _i1i2 ... _Ij have the above-mentioned probability density function for each link toward their children. At this time, the probability density function of the update delay in the update tree △ _i1i2 …
_ij (t) is

[0056]

(Equation 8)

Is represented by Where △ ^(k) = Δ _i1i2 … _ij
^(k) . By repeating the equations (3) and (4), the probability density function of the whole in the same tree structure can be obtained.

[5. Efficiency of Update Delay Distribution Calculation in Updates from Plural Nodes] Focusing on the update start node and the read node, the update path is a path common to all read nodes, that is, the LCA (Least) of the read node from the update start node. Common Ancestor) and a path from the LCA to each node. Now, when the update is started from a plurality of nodes, the nodes having the same LCA have the same delay probability density function lower than the LCA. Therefore, in the present embodiment, by dividing the update path into the above-described two paths and calculating the probability density function, in the case where the update is started from a plurality of nodes, the latter one of the nodes having the same LCA is used. The calculation for the route can be shared.

The LCA of the read node for a certain update start node u is the one closest to the update start node u among the nodes included in the minimum tree including the read node. In this way, by sharing the calculation of the update delay probability density function in the minimum tree, the delay probability density function when reading from a plurality of nodes can be made more efficient.

[Details of Algorithm] Next, an algorithm for programming the above procedure will be described. The algorithm has the following steps. Step 1 Construct a read subtree. Step 2 Set the shortest node. Step 3 The node corresponding to the leaf of the read subtree is
A probability density function is obtained using equation (3), and the obtained result is transmitted to an adjacent node.

Step 4 For each adjacent node s on the read subtree, if each node has received the probability density function from the other adjacent node, the node obtains the probability density function using equation (4). To the node s. Step 5 Except for the shortest node,
Transmit the probability density function to all neighboring nodes and end.
After receiving the probability density functions from all adjacent nodes, the shortest node first calculates the delay probability density function of the shortest node and its descendants using Expression (4). Next, a probability density function of the update delay from the update start node is obtained by using Expression (3) between the calculated result and the update delay probability density function from the update start node to the shortest node. Finally, the obtained result is transmitted to the front-end node, and the processing ends.

The operation of each node is as follows. R: set of read nodes A: adjacent (connected via a direct link)
A set of duplicate nodes S: a set of duplicate nodes adjacent on the read subtree (the initial value is an empty set) N: a set of duplicate nodes that must send a message nexthop (node): a logical connection between duplicate nodes This function calculates an adjacent duplicate node to which a message is to be transmitted when transmitting to a node using a proper link. send (x ₁ , x ₂ ,..., x _k ) _ij : x ₁ , x _{2 ,.}
x _k means that node i sends to node j. receive (x ₁ , x ₂ ,..., x _k ) _ij : x ₁ ,
It means that node j receives x ₂ ,..., x _k from node i.

FIG. 4 illustrates the above algorithm, and shows the operation of the front-end node and the replica node. (1) Operation of Front End Node For any node rk included in the set R, the instruction “s
end ”. “Id” indicates the node number of the front-end node.

(2) Operation of Duplicate Node Four States Out_of_subtree, in_su
btree, subtree_done, calcul
ation_done and four state transitions receive (“makeReadSubtre
e ", R, N) _{j, i} , become_lca, star
t_pdf_calculation, receive
(“PDF”, P _j ) _{j, i} ).

[0065] Incidentally, MakeReadSubtree is a string for instructing the start of reading partial tree construction, PDF is a character string representing that P _i is the probability density function of the delay. _Pi is the probability density function of the delay. The state transition of the above algorithm is illustrated in FIG.

[Method of calculating latest degree] Next, a method of calculating the latest degree will be described. According to the method described above, the front-end node calculates the probability density function expression of the order statistic of the update delay of the data obtained from the plurality of read nodes,
It can be calculated for all update start nodes.
In the following, these statistical distributions are expressed as ordinal statistics, and a method of calculating the stochastic freshness of data when a new read is performed will be described. Here, the order statistics are obtained by arranging arbitrary samples in ascending order and ranking them.

Now, the order statistics of the n old samples are represented by X ₍₁₎ , X
₍₂₎ , X ₍₃₎ ,..., X _(n) , m new samples Y ₍₁₎ , Y
₍₂₎ , Y ₍₃₎ ,..., Y _(m) were taken and all the samples were ranked. Possible combinations for taking new specimens are:
_{n + m} C _n and the probability of each occurrence is 1 /
Is a _{(n +} m C _n). Therefore, the probability density function calculated by the above method is expressed as an order statistic, and the calculation of the confidence interval when reading is performed from a plurality of read nodes is as follows. X ₍₁₎ ¹ , X ₍₂₎ ¹ , X ₍₃₎ ¹ ,..., X _(n1) ^1, and X _(n1) ^{1 represent} the order statistics based on the probability density function of the update delay to two different read nodes. ₍₁₎ ² , X _{( 2)} ² , X ₍₃₎ ² ,..., X _(n2) ² . X _(i) ¹
The new sample for Y ₍₁₎ ¹ and the new sample for X _(i) ² are Y
₍₁₎ ² .

[0068]

(Equation 9)

The probability P is X _(j) ² ≦ X _(l) ¹ ≦ X
_{(j + 1)} ²

Lj / ｛(n₁+1) (n_Two+1)｝ ≦ P ≦ l (j + 1) / ｛(n ₁ +1) (n_Two+1)｝ (5)

Here, n ₁ and n ₂ are X _(i) ¹ and X
_(j) means the number of ² ; Therefore, the confidence interval over the probability P is

P ≦ lj / {(n ₁ +1) (n ₂ +1)} (6)

The minimum X _(l) ¹ that satisfies the above condition. This represents the probability density function obtained by the above-described method as an order statistic, and matches the confidence interval when one new sample is taken based on this. Therefore, in the present embodiment, first, the latest degree from a single update start node is represented as a probability density function obtained by the above method as an order statistic, and when one new sample is taken based on this, As a confidence interval with probability P,
Calculate at each shortest node. Next, the front-end node that has received the latest degrees from the plurality of update start nodes returns the largest latest degree to the client node as its worst value.

[0074]

As described above, according to the present invention, when a client wants to acquire data of a certain newness, the acquired data and a scale indicating how recent the data are are provided. It is possible to obtain data that supplements the requirements. Further, the present invention can be applied to a system that determines an operation based on obtained data, that is, a determination support system or the like.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a network and a computer according to the present invention.

FIG. 2 is an explanatory diagram showing a relationship between each node when receiving an update request and a read request.

FIG. 3 is an explanatory diagram showing an update tree composed of a path from the update start node to the shortest node, and a read subtree, with the update start node as a root.

FIG. 4 is an explanatory diagram describing a program for realizing operations of a front-end node and a replication node.

FIG. 5 is a state transition diagram showing the state transition according to FIG. 4;

[Explanation of symbols]

1 ... computer, 2 ... network, 3 ... update start node,
4: Read node.

Claims (8)

[Claims] A plurality of computers are connected to a communication network, and one or more processes for managing data operate on these computers, and the contents of data managed by each of the processes are all the same, The process of managing the data, when receiving a data update request, changes the data of its own process and propagates this update request to other processes, and in a data replication system in which data is updated one after another, , A client node that requests updating and reading of the data, and a front-end node that mediates a request from the client node to the replication node. The replica nodes constitute one or more groups connected by a logical tree structure indicating an update path. When the update request is propagated, the update request is propagated to all nodes other than the node that has propagated the update request among the replica nodes adjacent on the tree structure, and (b) the front-end node Performs an update request to one copy node (hereinafter referred to as an update start node) in each of the groups in response to an update request from the client node; Mediates a read request to the duplicate node (hereinafter, referred to as a read node), returns the latest data among the plurality of read data to the client node, and determines how recent data can be obtained. Calculate and manage the calculation result,
In response to a request from a node, a read request is made by selecting a combination of read nodes that satisfy this request, and the client node determines the acquired data and a measure indicating how recent the data is. A data replication system characterized by responding to 2. The scale according to claim 1, wherein the scale indicating how recent the acquired data is is defined as the latest degree, and reflects a update from the present to a time T before the current time with a predetermined probability p. Wherein the latest degree of the probability p is defined as T. 3. The method according to claim 2, wherein each of the replica nodes manages a statistical distribution of update delays when propagating a logical link connected to the node, and includes a minimum subtree including the read node. Hereinafter, a read subtree is formed, and among nodes included in the read subtree, a node closest to the update start node on the tree structure (hereinafter, referred to as a subtree).
When a subtree composed of a path from the update start node to the shortest node and the read subtree is rooted at the update start node, a node serving as a leaf of the read subtree is Transmitting the update delay statistical distribution of the link connected to the node to the parent node, and the replica node other than the leaf node receives the statistical distribution from all of the child nodes, and then uses this statistical distribution. After calculating the minimum statistical distribution of the update delay from the duplication node to the read node that is a descendant of the duplication node, this statistical distribution is transmitted to the parent node, the shortest node is calculated statistical distribution or The latest degree is transmitted to the front-end node, and the latest degree or the update delay statistical distribution for all update start nodes is received. Ntoendo node,
A data replication system for calculating the latest degree in the same tree structure based on the latest degree or statistical distribution. 4. The read-out node according to claim 3, wherein when the shortest node is a read subtree, read nodes in which read nodes exist in ancestors are deleted, and read nodes in which read nodes exist in ancestors disappear. A data duplication system using the completed read subtree. 5. A plurality of computers are connected to a communication network, and one or more processes for managing data operate on these computers, and the contents of data managed by each of the processes are all the same, The process of managing the data, when receiving a data update request, changes the data of its own process and propagates this update request to other processes, and in a data replication system in which data is updated one after another, , A client node that requests updating and reading of the data, and a front-end node that mediates a request from the client node to the replication node. The replica nodes constitute one or more groups connected by a logical tree structure indicating an update path. When the update request is propagated, the update request is propagated to all nodes other than the node that has propagated the update request among the replica nodes adjacent on the tree structure, and (b) the front-end node Performs an update request to one copy node (hereinafter referred to as an update start node) in each of the groups in response to an update request from the client node; Mediates a read request to the duplicate node (hereinafter, referred to as a read node), returns the latest data among the plurality of read data to the client node, and determines how recent data can be obtained. Calculate and manage the calculation result,
In response to a request from a node, a read request is made by selecting a combination of read nodes that satisfy this request, and the client node determines the acquired data and a measure indicating how recent the data is. A recording medium on which a data duplication program is recorded, wherein a program for responding to the data is recorded. 6. The method according to claim 5, wherein the scale indicating how recent the acquired data is is defined as the latest degree, and reflects the update from the present to a time T before the current time with a predetermined probability p. A data duplication program, wherein the latest degree of the probability p is defined as T. 7. The method according to claim 6, wherein each of the replica nodes manages a statistical distribution of update delays when propagating a logical link connected to the node, and includes a minimum subtree including the read node. Hereinafter, a read subtree is formed, and among nodes included in the read subtree, a node closest to the update start node on the tree structure (hereinafter, referred to as a subtree).
When a subtree composed of a path from the update start node to the shortest node and the read subtree is rooted at the update start node, a node serving as a leaf of the read subtree is Transmitting the update delay statistical distribution of the link connected to the node to the parent node, and the replica node other than the leaf node receives the statistical distribution from all of the child nodes, and then uses this statistical distribution. After calculating the minimum statistical distribution of the update delay from the duplication node to the read node that is a descendant of the duplication node, this statistical distribution is transmitted to the parent node, the shortest node is calculated statistical distribution or The latest degree is transmitted to the front-end node, and the latest degree or the update delay statistical distribution for all update start nodes is received. Ntoendo nodes,
A recording medium on which a data duplication program is recorded, wherein the latest degree in the same tree structure is calculated based on the latest degree or statistical distribution. 8. The method according to claim 7, wherein when the shortest node is a read subtree, a read node having a read node existing in an ancestor is deleted, and a read node having a read node existing in an ancestor disappears. A recording medium storing a data duplication program, wherein the read subtree is used.