KR100936238B1 - Lazy Replication System And Method For Balanced I/Os Between File Read/Write And Replication - Google Patents

Abstract

The present invention provides a delayed replication system and method for the balanced execution of file input and output and replication, the delayed replication system of the present invention, a plurality of data servers for storing a file; A metadata server for storing metadata; And a client which obtains information on the updatable file by referring to the metadata and updates the file information only in the data server storing the file among the data servers, wherein the metadata server is updated by the client. It is characterized by using a thread to instruct the data server storing the original file to transfer the original file to the data server does not have the original file.

File I / O, Delay, Replication, Balance

Description

Lazy Replication System And Method For Balanced I / Os Between File Read / Write And Replication}

The present invention relates to a delayed replication system and method for the balanced execution of file I / O and replication, and more particularly to a delayed replication system that provides the balance and efficiency between common file I / O operation and maintaining consistency between a plurality of files More specifically, in an environment where multiple storage devices are networked and multiple clients access storage devices at the same time, update them to maintain the same image when multiple identical files exist for backup or recovery purposes. And a delayed replication system and method that allows the copying operation to be performed in a balanced manner without disturbing the general file input / output request of the client.

The present invention is derived from the research conducted as part of the IT growth engine technology development project of the Ministry of Information and Communication and the Ministry of Information and Telecommunication Research and Development. [Task Management Number: 2007-S-016-01, Title: Development of a Low-cost Large-scale Global Internet Service Solution] ].

BACKGROUND A storage system that provides a user with an integrated storage space by connecting a plurality of computers using a network is used in various fields such as web search processing and super computing. Some of these systems create multiple identical files and store them separately on different storage devices in case of user-failed file failures, improving recoverability and improving performance when multiple clients want to read the same file. Is also intended.

There are two main types of file copy creation or update in such a system. The first is when some or all of the files have been updated, in which case the updated content must be propagated to all existing copies. The rest is when any one of the data servers that stored the file copy is no longer running due to a failure. In this case, an additional copy of the file must be created on the new data server.

General I / O operations and file replication operations use the same network, disk, and memory resources. Therefore, excessive use of resources in the file replication operation interferes with normal I / O operations. If the file A is stored in the data servers D1, D2, and D3, the update operation of A changes data stored in all of the data servers D1, D2, and D3. In this case, when the sequence is D1->D2-> D3, the first update operation to D1 is the recording of new data in the file system as a whole, but the subsequent recording to the data servers D2 and D3 is merely to increase the availability of the entire system. It's work. Therefore, if O _{B, which} is the creation operation of file B, is received during the update operation of file A, the importance of updating file A to data servers D2 and D3 is lower than that of newly creating file B. There is a possibility of delaying the creation of B.

US Patent US7065618B1, invented by Ghemawat et al. And filed on June 30, 2003 and filed on June 30, 2003, proposes a method for managing update operations of files stored on multiple computers. have. In this patent, the master server that manages metadata selects one chunk server from the chunk servers that store files, and issues a lease to the chunk server to issue a primary chunk. In the role of server, the rest of the chunk servers are secondary chunk servers, and whenever a file is updated, it is always under the control of the primary chunk server for that file. When the data is updated by the client, the data is immediately delivered to all chunk servers, and the primary chunk server decides whether to write the data to disk and informs the subchunk servers that all chunk servers simultaneously write the data to the disk. do. That is, a file write operation by the client immediately creates an additional load by the number of copies. Therefore, there is a problem in that it is difficult to secure resources required for input / output or update requests of other clients. In addition, if a problem occurs in the main chunk server during recording, there is a problem that the clients cannot access even if the images of the files stored by each chunk server do not match or match each other. There is a burden to classify and manage indexes.

The present invention is to solve the above problems, the object of the present invention is to update the first data among the multiple copies is the fastest and most efficient, and the remaining copies of each data server as long as it does not interfere with general input and output operations It is to provide a delayed replication system and method that can evenly distribute the load and improve the convenience of the client while ensuring the stability of the system by supporting the reversion by using the remaining copies for errors occurring during the minimum data update. .

It is another object of the present invention to provide a delayed replication system and method that can separate the copy operation from the recording operation for the file so that the input and output operations for the new file are not interrupted by the existing file replication operation.

Another object of the present invention is to adjust the priority of the replication operation according to the user's characteristics, such that the environment in which the stability by the replication should be guaranteed, the environment in which the response to the user is important, or files having two characteristics exist together. It is to provide a delayed replication system and method that can improve the overall performance of the system in both environments.

Another object of the present invention is to divide a copy into a file that can be updated and a read-only file, so that if a file error occurs during the update, it is possible to revert to a previous stable version file, which can improve the stability of the system. And providing a method.

The delay replication system of the present invention for achieving the above object comprises a plurality of data servers for storing a file; A metadata server for storing metadata; And a client which obtains information on the updatable file by referring to the metadata and updates the file information only in the data server storing the file among the data servers, wherein the metadata server is updated by the client. It is characterized by using a thread to instruct the data server storing the original file to transfer the original file to the data server does not have the original file.

The delayed replication method of the present invention includes selecting an updatable copy from metadata for a file to be updated, obtaining a data server address storing the copy, and then transmitting update data to the corresponding data server; Requesting a delayed replication from the metadata server as the start of replication is set; Searching for a data server having up-to-date data about files that need to be replicated and assigning a replication job to a job thread in charge of the data server; Selecting a data server having the least load and instructing the data server with the original file to transfer the file to the selected data server; And transmitting the file instructed to be replicated to a data server to be replicated.

According to the present invention having the above configuration, the first data update among the multiple copies proceeds most quickly and efficiently, and the remaining copies evenly distribute the load of each data server without interrupting general input / output operations, and minimize the data update. In the event of errors occurring during the rest of the copy, the rest of the copy can be used to guarantee the stability of the system while improving the convenience of the client.

The present invention can separate the copying operation from the recording operation for the file so that the input / output operation for the new file is not interrupted by the existing file replication operation.

The present invention adjusts the priority of the duplication operation according to the characteristics of the user, so that the system can be used both in an environment in which stability by replication is to be guaranteed, in an environment in which a response to the user is important, or in an environment in which files having both characteristics exist together. Can improve the overall performance.

According to the present invention, by dividing the copy into an updateable file and a read-only file, when an error occurs in the file during the update, it is possible to revert to the previous stable version file and improve the stability of the system.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

1 is a block diagram of a delayed replication system for balanced execution of file input and output and replication according to an embodiment of the present invention.

Referring to FIG. 1, the delay replication system of the present invention includes a metadata server 100, a client 200, a plurality of data servers 300-1 to 300-n, and a network 400. Equipped. In addition, the delay replication system of the present invention includes a plurality of clients not shown in addition to the client 200. Here, the metadata server 100, the client 200, and the plurality of data servers 300-1 to 300-n connect through the network 400. And, one file may be divided into several chunks or stored as one continuous file. The metadata server 100 is implemented as a separate independent server, but is not limited thereto and may be located on the same machine as the client 200 or the plurality of data servers 300-1 to 300-n.

FIG. 2 is an exemplary diagram for explaining a process of duplicating and storing a file in data servers by a delayed replication system according to the present invention, and illustrating a process of updating one file in an environment using an assumed delayed replication system. .

Referring to FIG. 2, when the client 200 stores the original file 210 in the delay replication system, the client 200 is duplicated and stored in the predetermined number of data servers 300-1 through 30-n. Here, the files 310-1 through 310-n stored in the data servers 300-1 through 30-n have the same image. At this time, the metadata 110 information about the storage location is stored in the metadata server 110. If an additional update to the file 210 occurs, the client 200 obtains information on the updatable file 310-1 by referring to the information of the metadata 110 to update the file 210. The file information is updated only in the data server 300-1 storing the file. In this state, the file 310-1 has the latest version of data, and the remaining files 310-2 to 310-n have backup data for the previous version.

Updates to the remaining files 310-2 to 310-n stored in the data servers 300-2 to 300-n may be closed or periodically or managed according to the setting of the client 200. The delay is executed by the metadata server 100 according to the direct instruction of.

The metadata server 100 transmits the original file 310 to the data servers 300-2 to 300-n to the data server 300-1 storing the original file 310-1 updated by the client 200. In this case, when multiple files need to be updated by using multiple threads at the same time, the corresponding work is performed simultaneously on each data server to improve work efficiency. That is, the files 310-2 to 310-n of the data servers 300-2 to 300-n are files transmitted from the data server 300-1.

FIG. 3 is a diagram illustrating the structure of metadata shown in FIG. 2.

Referring to FIG. 3, the metadata 110 is composed of as many Copy Informatios 110-1 to 110 -N as the number of current copies.

One copy information is composed of a version 111, a file status 112, and an address (DSIP) 113 of a data server. Here, version 111 is incremented when the file is updated by the client 200 to indicate which has the latest data between each copy. The copy state 112 includes information on whether the corresponding copy is an updatable copy, validity information of data, validity information of data servers 300-1 to 300-n, and the like. The data server address 113 is used in the update process from the client 200 to the corresponding data server.

FIG. 4 is a flowchart illustrating a process of performing a file update operation by a client illustrated in FIG. 2, and illustrates a process of performing a file update operation to one data server before delay copying starts.

Referring to FIG. 4, the client 200 queries the metadata server 100 for metadata 110 about a file to be updated (S101). In this case, the client 200 selects a copy that can update the copy status information 112 from the copy information 110-1 to 110-N on the metadata 110, and selects the address of the data server storing the copy. 113) (S102).

The client 200 transmits an update request and update data to the corresponding data server (S103). The client 200 determines whether an error has occurred during the transmission of the update request and the update data (S104). If the error has not occurred as a result of the determination, the client 200 determines that the file update is successful and ends the file update.

If an error occurs as a result of the determination, the client 200 reports the error of the data server to the metadata server 100 (S105). Accordingly, the metadata server 100 updates the metadata 110 by reselecting an updateable copy from the copy information 110-1 to 110 -N by using the reported error information. Therefore, other clients can perform the update operation on the normal data server.

FIG. 5 is a flowchart illustrating a process of requesting a delay copy by a client illustrated in FIG. 2.

Referring to FIG. 5, when the client 200 finishes all file operations and closes a file, it is determined whether a replication start is set in advance (S201). If the determination result is not set, the file closing process is terminated.

If it is determined that the replication start is set, the client 200 requests the metadata server 100 to start the replication daemon dedicated to the replication job (S202).

After the replication request is made, the client 200 immediately performs all operations or determines whether to wait for the end of the replication operation for the file (S203). End the process.

If the determination result in the determination process (S203) can wait for the completion of the replication job, the client 200 periodically inquires the metadata server 100 whether the replication operation for the file is finished (S204). . At this time, the metadata server 100 checks the metadata 110 of the corresponding file, checks whether the current copy number reaches a predetermined number, and informs the client 200 whether to perform a replication operation.

Then, the client 200 determines whether the copying operation is finished (S205), and if the determination result is finished, the file closing process is terminated.

As a result of the determination in the determination process (S205), if the duplication operation is not finished, the client 200 waits for a predetermined time (S206), and then proceeds to the inquiry process (S204).

The metadata server 100 improves processing efficiency by collecting and processing replication tasks for all update data in units of entire delayed replication systems rather than in units of files. Here, the update process includes collecting information on a file that needs to be updated in the entire delayed replication system as shown in FIG. 6 and performing an update on the corresponding file as shown in FIG. 7.

6 is a flowchart illustrating a process in which the metadata server illustrated in FIG. 2 collects file information requiring a duplication operation.

Referring to FIG. 6, the metadata server 100 sequentially searches for files requiring replication by searching the metadata 110 for all files (S301). At this time, the metadata server 100 determines whether there are no more files that need to be duplicated through the search (S302), and if there is no file that requires the duplication, the file duplication process is terminated. At this time, the metadata server 100 creates and uses an index for a file that needs to be updated in advance every error of the data server, or differentially prioritizes replication according to the number of remaining copies of each file. I can regulate it.

As a result of the determination, if there is a file requiring duplication, the metadata server 100 determines whether the number of copies is smaller than a preset reference value (S303). If it is determined that the number of copies is not small, the metadata server 100 proceeds to the search process (S301).

As a result of the determination, if the number of copies is small, the metadata server 100 searches for a data server having the latest data on the file for a file that requires replication (S304). In this process, the metadata server 100 refers to the version information 111 of the metadata 110.

Subsequently, the metadata server 100 checks whether a work thread that is in charge of the data server exists (S305), and if there is a work thread as a result of the check, allocates a replication work to a work thread in charge of the data server (S306). ), The process proceeds to S301.

If there is no task thread as a result of the check, the metadata server 100 generates a new task thread in charge of the corresponding data server (S307), and proceeds to the allocation process (S306).

FIG. 7 is a flowchart illustrating a process of performing a delay replication operation by the metadata server illustrated in FIG. 2.

Referring to FIG. 7, when the metadata server 100 starts a file replication thread, the thread determines whether a new file replication job is allocated to the thread (S401), and if it is not assigned, the new replication job is allocated. Wait for it. Here, the thread waits for the allocation of a new job may be implemented by implementing a queue for a job event, waiting for the implemented queue to be filled, or operating by interpreting a signal from the outside.

If it is determined that a new replication job has been allocated, the thread sets a lock on the corresponding file to prevent file changes during replication (S402). In this process, the thread may allow the file to change during replication, detect the file's change after the replication, and retry copying the file when the file changes.

After the file lock is set, the thread selects the data server having the least load among the data servers 300-1 to 300-n (S403). In this process, a thread can use a variety of aggregated state information, such as storage capacity, CPU and network, disk utilization, and physical distances taking into account the topology of the rack or switch. Alternatively, a thread may be implemented to perform a simple and fast random selection method or a sequential selection method.

Subsequently, the thread instructs the data server having the original file to transfer the file to the data server selected in the selection process (S403) (S404). At this time, the thread checks whether an error occurs during the file transfer (S405), and if the check is normally completed without error, the thread reports the completion of replication to the metadata server 100 to update the copy information of the metadata. To be made (S406). Here, the error of the data server with the original file can be confirmed when there is no response for a certain period of time or by inspection of network activity, and the error of the replication target data server is reported by the data server with the original file.

After the copy information is updated in this way, the thread releases the lock on the original file to allow access to another request (S407), and then determines whether the performed replication operation is the last replication operation (S408). If the determination result is not the last replication job, the thread proceeds to step S401 to determine whether the job is allocated.

If the determination result in the determination process (S408) is the last replication job, the metadata server 100 terminates the operation of the file replication thread. On the other hand, when the thread requests a task for the last duplicate request, the thread indicates that there are no more requested tasks. At this time, when there is no further request work, the metadata search for all files on the delayed replication system is finished.

If there is an error as a result of the check in the checking process (S405), the thread cleans up the file interrupted during transmission in the data server to be replicated (S409), and determines whether an error has occurred in the data server where the original file is located (S410). If there is no error as a result of the determination, the thread proceeds to the server selection process (S403).

If there is an error as a result of the determination in the determination process (S410), since the thread can no longer perform a replication request for the corresponding file, the replication operation is terminated and the lock release process is performed (S407).

8 is a flowchart illustrating a process of performing a copy operation by the data server illustrated in FIG. 2.

Referring to FIG. 8, after one data server having an original file starts file duplication, a file instructed for duplication is transferred to a data server that is a duplication target (S501). At this time, one data server determines whether the file transfer succeeded without error (S502). In this process, one data server determines that the transmission failed when a failure response or a success response does not come within a certain time.

If the determination is successful, one data server terminates the file duplication process.

If it is determined that the transmission fails, one data server reports the transmission failure to the metadata server 100 to update the corresponding metadata (S503).

Although the technical spirit of the present invention has been described in detail according to the above preferred embodiment, it should be noted that the above embodiment is for the purpose of description and not of limitation. In addition, those skilled in the art will understand that various embodiments are possible within the scope of the technical idea of the present invention.

1 is a block diagram of a delayed replication system for balanced performance of file input and output and replication according to an embodiment of the present invention.

2 is an exemplary diagram for explaining a process of replicating and storing a file in data servers by a delayed replication system according to the present invention.

FIG. 3 is a diagram showing the structure of metadata shown in FIG. 2; FIG.

4 is a flowchart illustrating a process of a file update operation performed by a client illustrated in FIG. 2.

5 is a flowchart illustrating a process of requesting a delayed replication by a client illustrated in FIG. 2.

FIG. 6 is a flowchart illustrating a process in which the metadata server illustrated in FIG. 2 collects file information requiring a copy operation.

7 is a flowchart illustrating a process of performing a delayed replication operation by the metadata server illustrated in FIG. 2.

8 is a flowchart illustrating a process of performing a copy operation by the data server illustrated in FIG. 2.

Claims (19)

Any one of a plurality of data servers for storing the original file; At least one other data server, each storing a copy of the original file among the plurality of data servers; A metadata server for storing metadata of the original file and each copy; And A client for updating the original file by obtaining information about the original file by referring to the metadata; After the update of the original file is completed, the metadata server transmits the updated original file to other data servers storing a copy of the original file to a data server storing the original file updated by the client. Instruct them to replicate, Lazy replication system for replicating the data server with the least load among the other data server first. The method of claim 1, And the update unit of the original file is smaller than the delay copy unit of the original file. The method of claim 1, And the files stored in the other data servers have the same image. The method of claim 1, wherein the copy of the original file stored on the other data servers, And a delayed replication system that is older than the updated original file. The method of claim 1, Updating the files of the other data servers is delayed and performed according to the setting of the client. The method of claim 1, Updating the files of the other data servers is delayed and executed by the metadata server according to an administrator's instruction. The method of claim 1, wherein the metadata, And as many copies of the copy of the original file as possible. The method of claim 7, wherein the copy information, A version that is incremented when the original file is updated by the client to indicate which has the latest data between each copy; A copy status including information on whether the copy is an updatable copy, validity information of data, and validity information of data servers; And Data server address used in the client update process of the original file Delayed replication system comprising a. In a delayed replication method in which a client updates original files stored in any one of a plurality of data servers using metadata stored in a metadata server, and then copies the updated original files to other data servers. Searching the one data server with reference to metadata of the original file and a copy of the original file and updating the original file of the one data server; When the update of the original file is completed, searching for another data server storing a copy of the original file, and transmitting and replicating the updated original file, The delayed replication method of replicating the data server with the least load among the other data servers first. The method of claim 9, wherein the updating is, The client querying the address of the one data server; Receiving the one data server address from the metadata server; Transmitting an update request and update data of the original file to the one data server; Determining, by the client, whether an error occurred while transmitting the update request and the update data; If it is determined that no error has occurred, the client determining that the update of the original file is successful and ending the update; And If the determination result is an error, the client reports to the metadata server that an error occurred in the update of any one data server. Delayed replication method comprising a. The method of claim 10, If the metadata server is notified that the error has occurred, recovering the original file using a copy of the original file stored in at least one of the other data servers. Delayed replication method further comprising. The method of claim 9, wherein the replicating step, Determining whether the client presets the start of replication after finishing all file operations and closing the file; If it is determined that duplication is not set, the client terminating the file closing process; If it is determined that replication start has been established, the client requesting a metadata server to start a replication daemon dedicated to replication; Determining whether the client can perform all operations immediately after requesting to start the replication daemon or wait for the completion of the replication operation for the file; If the client cannot wait for the end of the duplication operation, the client terminating the file closing process; If the client can wait for the completion of the duplication, the client periodically inquiring the metadata server whether the duplication of the file is finished; Determining, by the client, whether a copy job is finished; If it is determined that the copying operation is finished, the client terminating the file closing process; And If it is determined that the copying job is not finished, the client waits for a predetermined time and then proceeds to inquiring whether to perform the copying job; And Storing a copy of the original file on the other data servers as the copy operation ends. Delayed replication method comprising a. The method of claim 12, wherein in the step of inquiring whether to perform the replication operation, The metadata server checks the metadata to determine whether the number of copies of the original file has reached a preset number, and informs the client whether or not to perform a copy operation. The method of claim 9, wherein the replicating step, The metadata server sequentially searching for files requiring duplication by searching metadata for all files; Determining, by the metadata server, that the file no longer needs a copy by searching; If there is no file requiring duplication, the metadata server terminating the duplication process; Determining, by the metadata server, if there is a file requiring duplication, the number of copies is less than a predetermined reference value; If it is determined that the number of copies is not small, proceeding to the searching step; If it is determined that the number of copies is small, the metadata server searching for the one data server in which the original file is stored; The metadata server checking whether there is a working thread that is responsible for any one of the data servers; Test results further comprising: If the working threads, the metadata server assigns the task to the task thread replication responsible for a group of any one data server; And If there is no task thread as a result of the check, the metadata server creates a new task thread that is in charge of the one data server and then proceeds to the step of allocating the replication task. Delayed replication method comprising a. 15. The method of claim 14, wherein in the file replication termination step, The metadata server creates an index for a file that needs to be updated in advance whenever an error occurs in the plurality of data servers, or adjusts the priority of replication differentially according to the number of remaining copies per file. Way. 15. The method of claim 14, wherein retrieving any one data server: And the metadata server refers to version information of the metadata. The method of claim 9, wherein the replicating step, The metadata server starting a file replication thread; The thread determining whether a new file duplication task has been allocated to the thread; If it is determined that no new file duplication task has been allocated, the thread waiting for a new duplication task to be allocated; If it is determined that a new copy job is allocated, the thread locks the updated original file to prevent file changes during copying; The thread selecting one of the plurality of other data servers with the least load among the plurality of data servers; The thread instructing the one data server to transfer a file to the selected other data server; The thread checking whether an error occurred during the transfer of a file; If the duplication is completed normally without error, the thread reports completion of duplication to the metadata server to update the copy information of the metadata; The thread releasing a lock on the original file; Determining, by the thread, whether the replication operation performed is the last replication operation; If it is not determined that the final copy job, the method proceeds to the step of determining whether to assign the job; If the determination result is the last replication job, the metadata server terminating the operation of the file replication thread; If there is an error in the check result, the thread cleans up a file interrupted during transmission at any one of the data servers and then determines whether an error has occurred; If there is no error as a result of the determination, the method proceeds to selecting the other data servers one by one; And If it is determined that there is an error, the procedure ends the replication operation and proceeds to the unlocking stage. Delayed replication method comprising a. The method of claim 9, wherein the replicating step, Duplicating the original file of the one data server and transmitting it to the other data servers; Determining whether the transmission succeeded without error; If the transmission is successful, terminating the file duplication process; And If it is determined that the transmission failed, reporting the transmission failure to the metadata server; Delayed replication method comprising a. 19. The method of claim 18, wherein in the determining step, And determining that the transmission failed when a failure response comes from at least one of the other data servers, or a success response does not come within a predetermined time.

Images