KR101050155B1 - Data Distributed Storage Management System and Distributed Storage Management Method - Google Patents

Abstract

The present invention relates to a data distributed storage management system and a distributed storage management method. The present invention relates to a data distributed storage management system for distributing and managing a large amount of data in m data stores. The distributed database management system (distributed DBMS) and fragments, which are divided into L fragments composed of smaller relation tables by the vertical partition method, are divided into n distributed information (share) using the (k, n) threshold secret distribution method. And a storage agent for dividing, wherein m, L, k, and n are arbitrary natural numbers, and k is a natural number less than or equal to n.

Exclusive OR threshold secret distribution; Balanced Incomplete Block Design

Description

Data distributed storage management system and distributed storage management method {STORING AND MANAGING SYSTEM FOR SHARED DATA AND STORING AND MANAGING METHOD FOR THE SAME}

The present invention relates to a data distributed storage management system and a distributed storage management method. More specifically, the relational table is divided into a plurality of fragments by using a vertical partitioning method, and each fragment is divided into an exclusive logical sum (XOR) secret distribution method. The present invention relates to a data distributed storage management system and a distributed storage management method, which are distributed to a plurality of distributed information SHARE and distributed and stored in a plurality of databases DB by Balanced Incomplete Block Design (BIBD).

Secret distribution is one of the methods of storing data securely. A big feature of the secret distribution technique is that it prevents exposure of confidential information and prevents exposure from the inside. For example, if you forget your notebook PC somewhere, information will not leak unless data is stored. In other words, even if you try to take out personal information, if the data can not be copied, it is not exposed.

The simplest example of a secret distribution technique is a distributed management model over a USB interface. After encrypting not only personal data files but also programs and folders where system files are stored, the files are physically divided into local folders on the PC and USB memory.

In this manner, if neither a PC nor a USB memory is provided, files cannot be viewed, changed, copied, or deleted, so that important information can be prevented from being leaked even if the PC is stolen or lost.

In the case of a distributed database in which data is partitioned without being duplicated and arranged in each database server, when one of the database servers becomes inoperable due to a natural disaster or hardware failure, the database cannot function as a database. Therefore, this problem can be solved by replication. If all data is placed on all database servers, there is a risk that if any database is attacked, all data is stolen.

The present invention is to solve the problems and risks as described above, in the present invention, the data is distributed evenly across a plurality of data stores to safely store the data, and even in a state in which several database servers cannot be operated, the database server above the threshold Is to provide a highly confidential and reliable data distributed storage management system and distributed storage management method capable of restoring data if the system is operated.

The above object of the present invention is a data distributed storage management system for distributing and managing a large amount of data in m data stores, wherein the relationship table of data is composed of smaller relation tables by a vertical partition method. A distributed database management system (distributed DBMS) for dividing into fragments and a storage agent for dividing the fragment into n distributed information (shares) using a (k, n) threshold secret distribution scheme, wherein m, L, k, and n are arbitrary natural numbers, and k is a natural distributed number n or less, which can be achieved by a data distributed storage management system.

It is still another object of the present invention to provide a method for distributing and managing a large amount of data by distributing and managing a large amount of data in m data stores, wherein the relation table of data is divided into smaller relation tables by a vertical partition method. A first step of dividing the fragment into L fragments, a second step of dividing the fragment into n pieces of distributed information (share) using a (k, n) threshold secret dispersion method, and the m pieces of distributed information And a third step of distributing and storing the dog data stores according to the Balanced Incomplete Block Design, wherein m, L, k, and n are arbitrary natural numbers, and k is n. Achievable by the large-capacity data distributed storage management method characterized by the following.

The data distributed storage management system and distributed storage management method according to the present invention uses a threshold secret distribution method to evenly distribute data to a plurality of data stores to safely store data, and even if a few database servers are inoperable. If the database server is operating above the threshold, it can provide high confidentiality and reliability to restore data.

In addition, the present invention makes it possible to efficiently manage a data store by evenly distributing a plurality of distributed information to a plurality of data stores using a balanced incomplete block design method, and the time and access time for loading data into each data store. You can keep the same.

Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1. System Configuration

1 is a schematic diagram of a mass data distributed storage management system according to an embodiment of the present invention. The mass data distributed storage management system according to the present invention includes a plurality of database servers 100 connected over a network to store and store data in a plurality of data storages (DS), each of which is a database server 100. Is composed of a distributed database management system (distributed DBMS, distributed database management system, hereinafter referred to as 'distributed DBMS') and a storage agent (SA). The client is a user who inputs commands such as data search / input / delete / update to the database server 100. The data store is a physical device that stores distributed information, and may be generally configured as a hard disk, a floppy disk, a magnetic tape, a flash memory, or the like.

2 is a detailed configuration diagram of a database server 100 according to an embodiment of the present invention. The database server 100 includes a user authenticator, a privacy data manager, a distributed DBMS, and a storage agent 50, and the storage agent 50 includes a splitter and a reconstructor. ) And distributed data management tables.

The user authenticator is a module for preventing access by authorized or unauthorized users and authenticating only authorized users. Distributed DBMS processes queries of basic distributed DBMS functions and functions to split relation tables into fragments by using vertical partitioning. Vertical Fragmentation divides the attributes that make up the global table into several subsets. It can be divided into fragments, which are smaller relation table units depending on attributes that are more complex but more similar to horizontal partitions.

Distributed DBMSs exist on each database server and are primarily responsible for three functions.

(1) Partition function: The partition structure is determined by the vertical partition method based on the database statistics and the number of database servers.

(2) Distributed query function: In a distributed database, relations are divided into fragments, so each query is executed on the fragments. Distributed DBMSs create subqueries based on queries and partition structures sent from clients and work in parallel with other distributed DBMSs.

(3) Local processing function: The distributed DBMS requests the storage agent's fragment corresponding to the sub query and processes the sub query using the database engine installed in the system.

A storage agent resides on each database server, and all distributed DBMSs perform data distribution and restore using the (k, n) threshold secret distribution scheme, and manage fragments created from relations in the database.

The privacy data manager is a module that performs data privacy protection through authorization processing and authority management of data usage, and the disperser and decompressor are modules that inject and restore the fragmented fragments. The distributed management table is a module that allocates distributed information to a data storage. Each module shown in FIG. 2 is generally provided as a software module for performing a corresponding function, but may be provided as a hardware module when rapid data processing is required.

2. Overall Data Processing Flow

3 is an explanatory diagram showing a flow in which a large amount of data is processed as an embodiment of the present invention. In a distributed DBMS, a relation table is vertically divided into L fragments, which are small relation tables, according to high similarity. In this case, L is a natural number equal to or less than the total number of the database server 100. The storage agent writes each fragment into n distributions according to the exclusive OR (k, n) threshold secret distribution method. Where n is a natural number less than or equal to the total number of data stores, and threshold k is any natural number less than n. Of course, L, k, and n must be natural numbers greater than zero. A total of L × n distributed information is distributed and stored evenly in a plurality of data stores by a discrete structure called balanced incomplete block design (BIBD). The unbalanced block design approach has the advantage that the time to load data can be achieved relatively similarly in any fragment by storing distributed information in a single data store and distributing it evenly across multiple data stores.

The large-capacity data processing system of the present invention specifies a number of data stores (m) more than the number of distributed information (n) of each fragment and a data store for storing distributed information about each data into one distributed information management table. It is characterized by managing. Such a configuration has an advantage of improving the reliability of the distributed management system by providing redundancy when restoring distributed information.

In the present invention, each data stored in the data store is divided into a plurality of distributed information using a secret distribution method. Distribution information of each partitioned data is stored in different databases according to distribution information management tables.

When restoring the data, information on the data store in which the distributed information is stored is obtained by using the key of the data to be restored. The number of data stores required for the restoration of data from the data store in which the distributed information is stored is selected, and the distributed information is obtained from each of the selected data stores to restore the data.

As an example for the present invention, the following conditions are assumed for convenience of description.

o Total number of data: l

o Number of data stores: m (DS1, DS2, ..., DSm) (m> n)

o Number of distribution information of each data: n

o Threshold: k

The mass data storage and management method according to the present invention comprises an initialization step and a data acquisition step.

In the initialization step, each piece of data is divided into distributed pieces of information, and the pieces of information are stored in different data stores according to the pieces of distribution information management table. In the following, di is divided into n pieces of distributed information about data di and stored in different n data stores among m items according to the distributed information management table. Distributed information management table can be designed as shown in Table 1.

Key Data store ID ₁ DS ₁ , DS ₂ , DS ₄ , ..., DS _{m -1} ID ₁ DS ₂ , DS ₃ , DS ₅ , ..., DS _{m -3} ID ₁ DS ₁ , DS ₃ , DS ₅ , ..., DS _{m -2} ... ... ID ₁ DS ₁ , DS ₃ , DS ₄ , ..., DS _{m -1} ... ... ID ₁ DS ₁ , DS ₃ , DS ₄ , ..., DS _m

를 얻는다.N distribution information for di, with data di, threshold k, and the total number of variance information n as input to the information dispersion function Share.

Get

를 m개의 데이터 저장소 가운데 분산 정보 관리 테이블에 의해 각각 저장한다. n개의 분산 정보는 이하와 같이 데이터 저장소에 저장된다.N distributions of information available

Are stored by distributed information management table among m data stores. The n pieces of distributed information are stored in the data store as follows.

는 데이터 저장소 DS₁ 에 저장된다.Distributed information

Is stored in the data store DS ₁ .

는 데이터 저장소 DS₃ 에 저장된다.Distributed information

It is stored in the data storage DS _3.

는 데이터 저장소 DS₄ 에 저장된다.Distributed information

Is stored in the data store DS ₄ .

...

는 데이터 저장소 DS_{m -1} 에 저장된다.Distributed information

Is stored in the data store DS _{m -1} .

o Data Acquisition Step

Explain when a client wants to retrieve data di. First, the client uses the key IDi of the data di to obtain the information of the data store where the distributed information of di is stored in the distributed information management table.

And select k from the n data stores where the distribution information of data di is stored. The distributed information of the data di is obtained from these databases, respectively, and the data di is restored from the distributed information obtained using the information decoding function Merge.

를 선택한다. 다만, k개의 데이터 저장소

는 데이터 di 의 분산 정보가 저장되고 있는 데이터 저장소 (DS₁, DS₃, DS₄, ..., DS_{m -1})에 포함된다.-The user refers to the distribution information management table and acquires the information of the data storage in which the distribution information of the data di is stored. And k datastores out of m datastores

Select. However, k data stores

Is included in the data storage (DS ₁ , DS ₃ , DS ₄ , ..., DS _{m -1} ) in which distributed information of data di is stored.

를 취득한다.The user obtains distribution information of data di from each of k selected data stores. First, distributed information of data di from data store DS _j1

Get.

를 Query 한다.. Step 2-1: The client distributes data di of information di to data store DS _j1

Query.

를 클라이언트에게 보낸다.. step2-2: Data store DB _j1 receives Query from client

Send to the client.

를 취득한다.. step2-3: The client is distributed information of data di from data store DS _j1

Get.

를 취득한다.Distribution information for data d2 from data store DS _j2

Get.

를 Query 한다.. step2-1: Client distributes data di to data store DS _j2

Query.

를 사용자에게 보낸다.. step2-2: DS _j2 data store inquired from the client distributed information of the data di

Send to the user.

를 취득한다.. step2-3: The user is distributed information of data di from database DS _j2

Get.

In this way, distribution information of data di is obtained from k selected data stores.

, 임계치 k, 분산 정보의 총 수 n, 이들 3개를 함수 Merge의 입력으로 해서, 데이터

를 얻는다.-K pieces of distributed information obtained

, The threshold k, the total number of variance information n, and these three are input to the function Merge, and the data

Get

3. Description of basic operation of the system

Describe the behavior of distributed DBMSs and storage agents in creating and retrieving tables in the system.

- relation( relation ) Filling Procedure

The relation creation and function processing flow of the distributed database will be described with reference to FIG. 4.

① Create a fragment

A distributed DBMS creates L fragments using vertical partitioning from the relationship data. Register the partition structure, relation name, etc. as data shared by all distributed DBMSs.

② Fragment transmission

The distributed DBMS specifies the number of shares n and the number k required for restoration, and sends a batch request and each fragment to any storage agent.

③ fragmentation

The storage agent receiving the fragment creates n pieces of distributed information (Share) from the fragment by using the (k, n) threshold secret distribution method.

④ Distributed information share

The storage agent that created the distributed information (Share) randomly selects n storage agents and sends the distributed information (Share) to them. The header information of the distribution information (Share) is transmitted to the storage agent that does not transmit the distribution information (Share).

⑤ Send batch confirmation of distributed information (Share)

The storage agent receiving the distribution information (Share) or the header information executes the batch and sends a batch confirmation to the sender storage agent.

⑥ Confirmation of batch termination

When the batch confirmation is received from all storage agents that have distributed the share information and the header information, the batch completion confirmation is transmitted to the distributed DBMS requesting the batch.

⑦ End of relation creation

The distributed DBMS receives the end of batch confirmation from all storage agents that have sent fragments and ends the job.

Data retrieval procedure

The data search function processing flow of the distributed database will be described with reference to FIG. 5.

① Search request: Client sends search request query to arbitrary distributed DBMS.

② Subquery: A distributed DBMS that receives a query creates a subquery considering the attributes and partition structure that the query accesses and sends it to each distributed DBMS.

3. Fragment Request: The distributed DBMS that receives the subquery requests to any storage agent to restore the fragments needed to execute the subquery.

Restoration of fragments: A storage agent that receives an intention from a distributed DBMS sends a survey of distribution information to another storage agent. The storage agent, which has received the distributed information survey, returns a presence report if the shared information exists on the local disk. The storage agent receiving the presence report requests the distributed information (Share) from k storage agents that store the distributed information (Share). When k pieces of distributed information are collected, they are recovered to obtain the original fragments, and the fragments are sent to the distributed DBMS.

5. Subquery Execution: The distributed DBMS that acquired the fragment executes the subquery and sends the search results to the distributed DBMS that sent the subquery.

⑥ Merge the search results: After receiving the search results of all subqueries, the distributed DBMS merges the search results and creates a search result for the query sent from the client.

⑦ Send search result: Distributed DBMS sends search result to client.

Data update procedure

The data update function processing flow of the distributed database will be described with reference to FIG. 6.

Update request: The client sends an update request query to a distributed DBMS on any database node.

② Subquery: The distributed DBMS that receives the query creates a subquery considering the attributes updated by the query and the partition structure and sends it to each distributed DBMS.

File request: The distributed DBMS that receives the subquery requests the arbitrary storage agent to restore the fragment that executes the subquery.

④ File Restoration: The storage agent that receives a request from the distributed DBMS sends a share survey to other storage agents. The storage agent that receives the share investigation returns a presence report if the share exists in the locker disk. The storage agent that receives the existence report requests the distributed information (Share) from k storage agents that have the distributed information (Share). When k pieces of distributed information are collected, they are restored to obtain the original fragments. The fragment is then sent to the distributed DBMS.

5. Subquery Execution: Each distributed DBMS that has obtained a fragment executes a subquery. Sends an updated fragment and update request to any storage agent.

(6) Distribution of fragments: The storage agent receiving the updated fragment creates n pieces of distribution information (Share) from the fragments.

(7) Distribution information (Share) transmission: The storage agent which created the distribution information (Share) selects n storage agents arbitrarily and sends the distribution information (Share) to them. The header of the distribution information (Share) is transmitted to the storage agent that has not sent the distribution information (Share).

(8) Update distributed information (Share) and send update confirmation: The storage agent receiving the distributed information (Share) or its header updates it and sends an update confirmation to the sender's storage agent.

9) Update end confirmation: Upon receiving update confirmation from distributed storage (Share) and all storage agents that sent the header, it sends a batch, termination, and confirmation to the distributed DBMS requesting the update.

Fragment Update End: The distributed DBMS receives the update end acknowledgment from all storage agents that sent the fragment and sends the update result to the distributed DBMS that sent the subquery.

Update end: The distributed DBMS sends an update end confirmation to the client after receiving the update results in all subqueries.

4. Secret distribution using exclusive OR

The (3, n) threshold secret variance method will be described as an example for the exclusive OR (XOR) (k, n) threshold secret variance method. If less than three pieces of distributed information are exposed, no secret information is leaked, and the data lengths of the secret information and the distributed information are the same, and the distributed recovery process of the secret information can be realized at high speed by using the XOR operation process.

The (2, n) threshold secret distribution scheme, which can be distributed and restored at high speed using an exclusive logical sum, is ideal in which no secret information is leaked from a single distributed information and the ratio of the data length of the distributed information to the data length of the secret information is 1. It's a secret dispersion method. Also, (3,5) and (3,7) threshold secret variance methods have been proposed as a threshold secret variance method using an exclusive logical sum other than 2. Like the (2, n) threshold secret dispersion method, this method is also an ideal secret dispersion method in which decompression and restoration are possible at a high speed and information ratio is 1.

Hereinafter, the distribution / restoration algorithm of the exclusive logical sum (3, n) threshold secret dispersion method will be described.

Distributed algorithm

The order of dispersing secret information K using the (3, n) threshold method is shown below.

(1) The distributor D divides the secret information K into n-1 partial secret information Kq (1 ≦ q ≦ n-1).

(2) D independently generates both n-1 random numbers Rm and n random numbers Ti.

(3) D generates partial dispersion information S (i, m).

R_m

T_{i +m} S _{(i, m)} = K _{i -m}

R _m

T _{i + m}

(0 ≤ i ≤ n-1, 0 ≤ m ≤ n-2)

(4) D connects S _{(i, 0)} , ..., S _{(i, n-2)} to generate distributed information Si and distributes it to the manager Pi.

Table 2 shows a configuration table of the partial dispersion information generated here. In addition, the information rate is 1.

m = 0 m = 1 ... m = n-2 P0 S (0, m)

R₀

T₀ K ₀

R ₀

T ₀ K _{n -1}

R ₁

T ₁ ... K- _(n-2)

R _{n -2}

T _{n -2} P1 S (1, m) K ₁

R ₀

T ₁ K ₀

R ₁

T ₂ ... K _{1- (n-2)}

R _{n -2}

T _{1 + (n-2)} ... ... ... ... ... ... Pn-1 S (n-1, m) K _{n -1}

R ₀

T _{n -1} K _{n -2}

R ₁

T ₀ ... K ₁

R _{n -2}

T _{(n + 1) + (n = 2)}

Reconstruction algorithm

If three managers Pi, Pj, and Pk are gathered, they can be restored. Each dispersion information is referred to as Si, Sj, and Sk. (0 ≤ i <j <k ≤ n-1), and x and y are defined as follows.

x = j-i, y = k-j, where (x, y ∈ N, 1 ≤ x, y ≤ n-2)

(1) Each dispersion information is divided into partial dispersion information. That is, the following 3 (n-1) pieces of partial dispersion information can be obtained.

(2) The following U _{(i, k, m)} and U _{(j, k, m)} are required for all m (0 ≦ m ≦ n-2).

S_(k,m) U _{(i, k, m)} = S _{(i, m)}

S _{(k, m)}

S_(k,m) U _{(j, k, m)} = S _{(j, m)}

S _{(k, m)}

(3) The minimum natural number r that satisfies the following equation is required.

r (x + y) = x

(4) The following substeps are performed for all m (0 ≦ m ≦ n-2).

4-i) It is verified whether or not s that satisfies the following expression exists in the range of 0 ≤ s ≤ r-1 for the variable s (s) N ₀ ).

m + s (x + y) = n-1

If s that satisfies the above expression does not exist, the process of 4-ii-a) is executed. If present, the process of 4-ii-b) is executed.

4-ii-a) The following A _{(i, k, m)} is required.

4-ii-b) the following A _{(i, k, m)} is required.

(5) The minimum natural number v that satisfies the following equation is required.

vy = -x

(6) The following substeps are executed for all m (0 ≤ m ≤ n-2).

6-i) Variable t (t∈N ₀ For t), it is verified whether or not t satisfying the following expression exists in the range of 0 ≤ t ≤ v-1.

m + ty = n-1

If there is no. That satisfies the above expression, the processing of 6-ii-a) is executed. If present, the process of 6-ii-b) is executed.

6-ii-a) or less, A _{(j, k, m)} .

6-ii-b) A or less _{(j, k, m)} is required.

(7) The following W _{(i, j, k, m)} is required for all m (0 ≦ m ≦ n-2).

A_(j,k,m) W _{(i, j, k, m)} = A _{(i, k, m)}

A _{(j, k, m)}

(8) Restore all Kq (1 ≦ q ≦ n−1).

8-i) when m (0 ≤ m ≤ n-2) is present,

i + y-m = 0

M satisfying the above expression is m '. Increasing the value of w (initial value 0) by one or two, the following processing is performed until m '= -2wx = n-1. If m '-2wx = n-1 is stopped, 8-ii) is discontinued. (w ∈ N ₀ , 0≤w≤n-2)

...

_...

_...

...

8-ii) when m (0 ≤ m ≤ n-2) is present,

i + 2x + y-m = 0

M satisfying the above formula is m ''. Increasing the value of w (initial value 0) by one or two, the following processing is performed until m ''-2wx = n-1 is established. m ''-2wx = n-1 ends. (w ∈ N ₀ , (0 ≤ w ≤ n-2)

...

...

(9) Restore all original secret information K by connecting all partial secret information.

7 will be described with respect to the flow of data retrieval using a large data storage management system according to the present invention.

① The distributed DBMS receives the query sent from the client.

② Write a subquery from the received query and send the subquery to be processed to another DB server.

③ Subqueries are sent to the storage agent to request the fragments to be processed by the DB server.

(4) The storage agent creates a fragment from the data storage using an exclusive logical sum (k, n) threshold secret distribution method and delivers the fragment to the distributed DBMS.

The distributed DBMS integrates the results of searching with subqueries from fragments and the processing results of other DBs.

⑥ Deliver the integrated result to the client.

As a database system according to the present invention, it is assumed that a relation database is generally widely used. In principle, it is also applicable to other database systems such as XML DB and object DB.

Although specific embodiments of the present invention have been described and illustrated above, it will be apparent that various modifications may be made by those skilled in the art without departing from the technical spirit of the present invention. Such modified embodiments should not be understood individually from the spirit and scope of the present invention, but should fall within the claims appended to the present invention.

1 is a schematic diagram of a mass data distributed storage management system of one embodiment according to the present invention;

2 is a detailed configuration diagram of a database server 100 according to an embodiment of the present invention.

3 is an explanatory diagram showing a flow in which a large amount of data is processed as an embodiment of the present invention.

Explanatory drawing explaining relationship creation and function processing flow of a distributed database using FIG.

Explanatory drawing explaining the data search function process flow of a distributed database using FIG.

Explanatory drawing explaining the data update function processing flow of a distributed database using FIG.

7 is a flow chart for retrieving data utilizing a mass data storage management system in accordance with the present invention.

***** Brief description of the main symbols on the drawing *****

50: storage agent 100: database server

Claims (11)

A large data distributed storage management system that distributes and manages large amounts of data in m data stores. A distributed database management system (distributed DBMS) for dividing the relation table of data into L fragments composed of smaller relation tables by a vertical partition method; And A storage agent for dividing each fragment into n pieces of share information using a (k, n) threshold secret distribution scheme; Characterized in that it comprises a, M, L, k, n are any natural numbers, k is a natural number of n or less, And a data store in which the n distributed information is distributed, is stored in a distributed information management table, and the distributed information management table is stored in only one selected of m data stores. The method of claim 1, The storage agent distributes and stores the n pieces of distributed information in the m data stores according to a balanced incomplete block design method. 3. The method of claim 2, And the number m of the data stores is greater than the number n of the distributed information. delete The method of claim 1, An authenticator that prevents access to authorized or unauthorized clients and authenticates only authorized clients, and the privacy data manager further includes a privacy data manager that protects data privacy through authorization processing and authority management of data usage. Distributed Storage Management System. The method of claim 1, The storage agent includes a splitter for injecting the divided fragments and a reconstructor for restoring the divided fragments. The method of claim 1, And the storage agent uses an exclusive logical sum (k, n) threshold secret distribution scheme. As a method of distributed storage management of large data that distributes and stores large amounts of data in m data stores, A first step of dividing the relation table of data into L fragments composed of smaller relation tables by a vertical partition method; A second step of dividing each fragment into n pieces of share information using a (k, n) threshold secret distribution scheme; And A third step of distributing and storing the n distributed information in the m plurality of data stores according to a balanced incomplete block design method; It characterized in that the m, L, k, n is any natural number, k is less than n, And a data store in which the n distributed information is distributed, is stored in a distributed information management table, and the distributed information management table is stored in only one selected of m data stores. The method of claim 8, And the number m of the data stores is greater than the number n of the distributed information. delete The method of claim 8, And the second step is performed by an exclusive logical sum (k, n) threshold secret distribution method.

Images