JP2012190377A - Content decentralization and storage system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a system that makes a read without concentrating a load on a specific file server enabled when a content distribution server which distributes content pieces to a user terminal reads in a plurality of content pieces together before the distribution in a decentralization file system.SOLUTION: In a content decentralization and storage system including the decentralization file system which includes a plurality of file servers storing respective content pieces and the content distribution server which reads in N content pieces from the respective file servers together at a distribution request from the user terminal and distributes the content pieces to the user terminal, the decentralization file system has a decentralization file system client 21 which determines N file servers or more as write destinations of the respective content pieces from the plurality of file servers, and the decentralization file system client has a write destination determination part 52 which determines a write destination file server 22 so that the respective content pieces are decentralized in time-series order.

Description

  The present invention uses a distributed file system capable of providing virtually integrated storage of a plurality of computers scattered over a network in a video-on-demand (VoD) distribution server, and fragments one piece of content ( The present invention relates to a distributed content storage system that creates a piece of content and stores a series of file sets in separate file servers.

As this type of technology, as shown in Non-Patent Document 1 and Non-Patent Document 2, a distributed platform that functions as one distributed file system by combining disks of a plurality of machines has been proposed.
Gfarm shown in Non-Patent Document 1 is a scalable distributed file system platform that meets the demands of large-capacity, large-scale data processing on a wide area network, and is suitable for efficient file sharing on a wide area network A distributed platform.
On the other hand, Hadoop disclosed in Non-Patent Document 2 can process a large amount of data that cannot be saved on a single disk in parallel and can be processed at high speed and efficiently, and is relatively large and basically updated. It is a distributed platform suitable for I / O of files without a file.

  Conventionally, a monitoring system in a distributed file system composed of a plurality of servers regularly collects the status of each server and performs an integrated analysis. Information collected includes CPU usage rate, memory usage, disk usage, CPU temperature, network connection status, and the like. As an integrated analysis example, the degree of congestion of a file server can be understood by grasping the number of file servers having a CPU usage rate of 80% or more. Thereby, it is possible to obtain an index such as whether or not the number of file servers is sufficient with respect to the usage status of the system.

URL: http://datafarm.apgrid.org/indeX.ja.html URL: http://hadoop.apache.org/

In the distributed file system described above, the file server is selected when each piece of fragmented content is stored based on resource information such as the CPU usage rate and free disk capacity of each file server. ing.
For example, as shown in FIG. 7, the content distribution server distributes the content pieces in chronological order according to the reproduction request from the end user. However, instead of starting the reading of the target content piece at the delivery start time of each content piece, N pieces of the previous content pieces are read before the delivery time so that the delivery stream (video, audio) is not interrupted and stored in the memory. It is loaded in advance to prepare for the distribution start time.

  However, in the case of the above-described method, when N pieces of content pieces adjacent in time series are read together (in the example of FIG. 7, when pieces of content pieces 1, 2, and 3 are read together, Alternatively, when content pieces 4, 5 and 6 are read together), when the content pieces to be read together are stored in the same file server (2010srv) as shown in FIG. , 2 and 3 are read together, the content file 1 and content piece 3 are congested in the same file server (2010srv), so the data input / output load increases in the target file server (2010srv), There has been a problem that reading performance such as response speed to the content distribution server is lowered.

  The present invention has been proposed in view of the above circumstances. In a distributed file system including a plurality of file servers in which content pieces are stored, a content distribution server that distributes content pieces to user terminals has a plurality of It is an object of the present invention to provide a distributed content storage system that can read content pieces without having to concentrate the load on a specific file server.

To achieve the above object, the present invention (Claim 1) includes a content generation server that generates a plurality of content pieces by fragmenting content, and a plurality of file servers that are physically distributed to store the content pieces. In a system comprising: a distributed file system comprising: a content distribution server that reads N pieces of each piece of content from each file server in response to a distribution request from a user terminal and distributes the content pieces to the user terminal The following structure is included.
The distributed file system includes a distributed file system client that determines N or more file servers to which each piece of content is written from the plurality of file servers.
The distributed file system client includes a write destination determining unit that determines a write destination file server so that the content pieces are distributed in time series.

  2. The content distributed storage system according to claim 1, wherein the distributed file system is a metadata server for creating a list of file servers having resource information such as a CPU usage rate and a free disk capacity exceeding a predetermined threshold. The write destination determination unit selects a file server to which the content piece is to be written from a plurality of file servers based on the list acquired from the metadata server.

  According to a third aspect of the present invention, in the distributed content storage system according to the second aspect, the list is created by arranging file servers in order of uniquely identifiable values.

  According to a fourth aspect of the present invention, in the distributed content storage system according to the third aspect, the order of values that can uniquely identify the file server in the list is the host name order, the IP address order, or the MAC address order of each file server. It is a feature.

  According to a fifth aspect of the present invention, in the content distributed storage system according to the second aspect, the list is created by arranging file servers in an arbitrary order designated by an operator.

  A content distribution storage system according to claim 1 or claim 2 is an environment in which the number of file servers equal to or greater than the multiplier of the number of simultaneous reads from the content distribution server and the total number of writes in the file server is installed. In the list, a copy of the content piece is stored in a file server having a rank obtained by adding the number of simultaneous readings to the rank of the file server storing the content piece.

  According to the present invention, the distributed file system client determines the write destination file server so that each piece of content is distributed in chronological order by the write destination determination unit, so that the content distribution server responds to the distribution request from the user terminal. When N pieces of content are read together from each file server, the read pieces of content are stored in N or more file servers so as to be distributed in chronological order, so that the read destination file servers are duplicated. Without having to concentrate the load on a specific file server.

  In addition, the write destination determination unit determines the write destination of the content piece from among the plurality of file servers based on the list of file servers having resource information such as CPU usage rate and disk free space equal to or greater than a predetermined threshold. The piece of content can be securely stored in the file server.

It is a whole structure model figure which shows an example of embodiment of the content distribution storage system of this invention. It is a model figure which shows the relationship between each piece of content in a distributed file system, and the file server (host name order) of a write-destination. It is a sequence diagram which shows from a content production | generation server, a distributed file system, a content delivery server, and the content production | generation between user terminals to distribution. It is a block diagram which shows the structure of the distributed file system client in a distributed file system. It is a flowchart figure which shows the process in the write-destination determination part of a distributed file system client. It is a model figure which shows the relationship between each content piece (copy existence) in a distributed file system, and the file server of a write destination (in order of host name). It is a sequence diagram explaining the timing of reading and delivery of a content piece in the prior art. It is a model figure which shows the relationship between each piece of content and the file server of a write destination in the conventional distributed file system.

An example of an embodiment of a content distribution storage system of the present invention will be described with reference to the drawings. FIG. 1 is an overall configuration diagram of a content distribution storage system.
The content distributed storage system includes a content generation server 10, a distributed file system 20, a content distribution server 30, and a user terminal 40, which are connected to each other via a network.

  The content generation server 10 is connected to the external storage 11 that provides content, and generates a plurality of content pieces 15 by fragmenting the content provided from the external storage 11 in units of time designated by the operator. The process of dividing and writing 15 into the distributed file system 20 is performed.

The distributed file system 20 includes a distributed file system client 21 that writes and reads content pieces 15, a plurality of file servers 22 that store each content piece in time series, and storage location information (meta information) of each content piece. ) Is recorded. When the distributed file system client 21 writes the content piece 15, the distributed file system client 21 selects the file server 22 that is the storage destination of each content piece.
The metadata server 23 records meta information indicating in which file server 22 each piece of content is stored.

The file servers 22 are physically distributed in order to store the content pieces 15 in time series.
In response to a distribution request from the user terminal 40, the content distribution server 30 reads N pieces of each content piece from each file server 22 and distributes them sequentially to the user terminal 40 via the network. Since the file server 22 stores N pieces of content pieces 15 in a time-sequentially physically distributed manner, the file server 22 includes at least N file servers 22 that can be selected as write destinations. ing.

Next, content piece writing and distribution processing in the content distributed storage system will be described with reference to FIGS.
First, a case where content pieces are written will be described.
The content generation server 10 generates a plurality of content pieces 15 by fragmenting the content provided in time units designated by the operator, and performs a process of dividing and writing each content piece 15 into the distributed file system 20. (A in FIG. 3).

  In the distributed file system 20, the distributed file system client 21 receives a write request for the content pieces 15 from the content generation server 10, and selects the file server 22 to which the content pieces 15 that are sequentially passed from the content generation server 10 in time series are written. To do. The file server 22 is selected by inquiring a list of write destinations with respect to the metadata server 23, and extracting (response) a plurality of file servers 22 having sufficient resource information such as CPU usage rate and free disk capacity. Create a list in which the host names of each file server 22 are arranged in the order of the host names (in the example shown in FIG. 2, the host names are in the order of 2010srv, srv_A, srv_B, workhost). The content piece 15 is determined and written.

  In the distributed file system client 21, a plurality of file servers 22 extracted in the list are selected with resource information larger than a preset threshold value. Note that in order to store N pieces of content 15 separately on physically separate file servers 22, there are N or more file servers 22 extracted from the list (those with sufficient resource information). is necessary.

  Accordingly, N pieces of content 15 that are close in time series in a series of file sets of one content are distributed and stored in physically separate file servers 22. Therefore, for the content distribution server 30 that reads these pieces of content 15, since the files to be read are physically distributed and stored, each of them can be read in parallel while distributing the load on the file server 22. The data can be loaded onto the memory at high speed and distribution can be started.

  In the example of FIG. 2, the writing destination of the content pieces is determined in the order of host names (ascending order) of each file server 22, but in order of host name (descending order), IP address order, and MAC address order of each server. For example, the write destination may be determined in the order of values that can uniquely identify the file server 22 or in any order designated by the operator in the setting file.

Next, a case where content distribution is performed will be described.
When the content distribution server 30 receives a distribution request for the desired content from the user terminal 40, the content distribution server 30 sends the first content piece among the N pieces of content 15 constituting the desired content to the distributed file system 20. Is issued (B in FIG. 3).
The distributed file system client 21 that has received the content piece read command makes an inquiry about the read destination to the metadata server 23, makes a read request to the corresponding file server 22 from the response from the metadata server 23, and The file server 22 that has received the read request sends (responds) the data of the first content piece 15 to the distributed file system client 21, and the distributed file system client 21 that receives this sends a read response to the content distribution server 30. .

By performing the same processing sequentially, the distributed file system client 21 reads the second content piece to the Nth content piece from the content distribution server 30 (reads N pieces of each content piece from each file server 22 at a time. ), The first to Nth content pieces are sequentially delivered to the user terminal 40.
Then, at the time of distributing the first to Nth content pieces, the content distribution server 30 issues a command to read the (N + 1) th content piece to the distributed file system 20, and from the distributed file system 20 based on this command. Read responses are sequentially made to the content distribution server 30.

  Next, the configuration of the distributed file system client 21 will be described with reference to FIG. The distributed file system client 21 includes modules of a user access control unit 51, a write destination determination unit 52, a write destination information collection unit 53, and a file system access unit 54.

The user access control unit 51 sequentially receives write requests for content pieces from an external application (content generation server) and transmits the requests to the write destination determination unit 52. The write result is returned to the external application.
The write destination determination unit 52 can uniquely identify the file server 22 such as the host name order of the file server 22, the IP address order, and the MAC address order of each server from the file server list that can be selected as a write target. In this order, the write destination is determined (therefore, the file server 22 to be written last time is stored and the next file server 22 is selected). The determined write destination file server 22 is transmitted to the file system access unit 54.

The write destination information collection unit 53 includes a writable file server 22 created on the basis of resource information such as a free disk capacity and a CPU usage rate of each file server 22 periodically collected from the file server 22 by the metadata server 23. A file server list is acquired and passed to the write destination determination unit 52.
The file system access unit 54 transfers write data to the target file server 22 according to the write destination file server information passed from the write destination determination unit 52 and receives a response.

  Next, the processing procedure in the write destination determination unit 52 of the distributed file system client 21 will be described with reference to the flowchart of FIG.

First, the write destination determination unit 52 receives a write request from the user access control unit 51 (step 61).
Next, from the write request received from the user access control unit 51, a file name, a file open option (in what mode the file is opened), an access mode at the time of file creation (write permission, read permission, write and read permission) Any required parameter is extracted (step 62).
The write destination determining unit 52 acquires the created file server list from the metadata server 23 (step 63).

If acquisition of the file server list is successful (step 64), the presence / absence of the previous write destination is checked (step 65). The presence / absence of the previous write destination is checked by checking whether the previous write destination file server stored in the memory of the write destination determination unit 52 exists in the file server list received from the metadata server 23.
If there is a previous write destination file server (step 66), the next host name to be the current write destination (the write destination obtained by adding +1 in the list order to the previous write destination) is selected (step 67).

A write request is issued to the selected write destination file server (step 68).
A file write response is received from the write destination file server (step 69).
The write destination is recorded in the memory (step 70).

  If there is no previous write destination file server (step 66), it is determined whether or not it is the first write (step 71), and if it is the first time, it is determined as the first host in the file server list (step 72). Since this is the first writing, the top of the file server list sorted by host name is selected.

When there is no previous write destination and the write is not the first time (step 71), it is assumed that the file server that was the previous write destination has been deleted or the file server that was the previous write destination has failed Yes. In this case, a file server close to the previous writing destination is searched (step 73).
That is, the file server name next to the previous write destination host name is determined as the current write destination in the dictionary order from the file server list. After the write destination is determined, a file write request is issued (step 68), a file write response is received (step 69), and the write destination is recorded (step 70).

FIG. 6 shows another embodiment of content piece writing and distribution processing in the content distributed storage system, and parts having the same configuration as in FIG. 2 are given the same reference numerals.
In the content distributed storage system of this example, a content generation server, a content distribution server, and a distributed file system (distributed file system client, metadata server, multiple file servers) are operated as in FIG. The number of simultaneous readings from the content distribution server is “3”.

  Then, the distributed file system client that has received the content piece write request from the content generation server, as in the example of FIG. 2, sets the file server to which the content pieces sequentially passed in time series from the content generation server to each file server. In addition to the fact that there is room in resource information such as CPU utilization and free disk capacity (from those above a certain threshold), each file server is selected in the order of values that can be uniquely identified.

  The distributed file system of the content distributed storage system of this example has a function of creating a copy of the content piece (original) written in the file server first in another file server. That is, the number of file servers equal to or greater than the multiplier (in this case, 6) of the number of simultaneous readings from the content distribution server (for example, “3”) and the number of copies (for example, the number of copies including the original is “2”) In this environment, the duplicates are stored in the file server having the rank obtained by adding the number of simultaneous readings (“3” in the example of FIG. 6) to the rank of the file server that stored the content pieces 15 (original) (ranked rank). To do. According to the example of FIG. 6, a copy of the first content piece 15 (original) stored in the file server (2010srv) is written as a copy to the file server (srv_B) after the third (rank 4) in ascending order.

  With this algorithm, when the content distribution server 30 reads the three content pieces 15 that are adjacent in time series at the same time, the read destination file server does not overlap even if the original or duplicate is read.

  In addition, when a failure occurs in the file server 22 to which the content piece 15 has been written once, the operator prepares an alternative machine, whereby the meta data of the content piece 15 held by the metadata server 23 in the distributed file system 20 is prepared. If the content piece is rearranged from the information to the prepared alternative machine, the failure can be dealt with.

  According to each embodiment described above, when the content distribution server 30 reads N pieces of each content piece 15 together, N pieces of content 15 that are close in time series are stored in physically separate file servers 22. Therefore, in the process of sequentially reading the content pieces 15, it is possible to prevent a state in which a plurality of content pieces are stored in one file server and to read in parallel without concentrating the load on the file server. It becomes possible to perform seamless distribution under load.

  DESCRIPTION OF SYMBOLS 10 ... Content generation server, 11 ... External storage, 15 ... Content piece 20 ... Distributed file system, 21 ... Distributed file system client, 22 ... File server, 23 ... Metadata server, 30 ... Content delivery server, 40 ... User terminal.

Claims (6)

Content generation server that generates a plurality of pieces of content by fragmenting content, a distributed file system that includes a plurality of physically distributed file servers for storing each piece of content, and distribution from user terminals In a system including a content distribution server that reads N pieces of each piece of content from each file server in response to a request and distributes the pieces to the user terminal,
The distributed file system has a distributed file system client that determines N or more file servers to which each piece of content is written from the plurality of file servers,
The distributed file system client includes a write destination determination unit that determines a write destination file server so that the content pieces are distributed in time series. The distributed file system has a metadata server for creating a list of file servers having resource information such as a CPU usage rate and a disk free capacity equal to or greater than a predetermined threshold,
The content distribution storage system according to claim 1, wherein the write destination determination unit selects a file server to which a piece of content is written from a plurality of file servers based on the list acquired from the metadata server.   The content distribution storage system according to claim 2, wherein the list is created by arranging file servers in order of uniquely identifiable values.   4. The distributed content storage system according to claim 3, wherein the order of values that can uniquely identify the file server in the list is the host name order, IP address order, or MAC address order of each file server.   The content distribution storage system according to claim 2, wherein the list is created by arranging file servers in an arbitrary order designated by an operator.   A file server that stores pieces of content in the list in an environment in which a number of file servers equal to or greater than the multiplier of the number of simultaneous reads from the content distribution server and the total number of writes on the file server (the number of copies including the original) are installed The content distribution storage system according to claim 1 or 2, wherein a copy of the content piece is stored in a file server having a rank obtained by adding the number of simultaneous readings to the rank.

Images