JP4245304B2 - Computer cluster system, file access method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a computer cluster system including a plurality of computers commonly connected via a network.
[0002]
[Prior art]
A computer cluster system in which a plurality of computers are connected via a communication device is used to supplement the functions of a single computer and achieve high performance and high reliability. By connecting computers in this way, processing can be distributed to individual computers to improve performance, and when a failure occurs in a part of the connected computers, on the computer that has failed by another computer High reliability can be realized by taking over the processing.
[0003]
One of the OS (Operating System) structures suitable for realizing such a computer cluster system is a microkernel system.
[0004]
As described in Japanese Patent Application Laid-Open No. 11-345134, the microkernel OS is separated from the kernel as a kernel part for performing process management and memory management, and a service for file system management and user management as a subsystem part. It has a structure.
[0005]
When a micro-kernel OS is used in a computer cluster system, the kernel is operated on all of a plurality of computers connected by a communication device, and the subsystem is only a part of the computer cluster system or a single computer. Make it work. When a general program (user program) that runs on a computer cluster system uses an OS service such as file access, it communicates with a subsystem that provides the service, and the subsystem cooperates with the kernel. And provide services.
[0006]
The advantage of the microkernel method is that it is possible to provide new services and upgrade the OS in the form of adding and changing subsystems instead of replacing the entire OS. Since the kernel to be operated can be made compact, it is possible to allocate a large amount of processing capacity of the computer to the operation of the user program.
[0007]
[Problems to be solved by the invention]
FIG. 12 is a diagram for explaining a computer cluster system realized by a conventional microkernel OS. In this system, a computer 900 that operates a user program 901 and a computer 910 that operates a subsystem 911 that manages a file system are commonly connected via a network 920. In order to simplify the description, a computer in which another subsystem operates is omitted.
[0008]
The computer 900 includes a user program 901, a kernel 902 that performs various hardware controls, a storage device 903 that stores data and the like, and a communication device 904 that communicates with other computers via a network. ing. The computer 910 has an equivalent configuration.
[0009]
The subsystem 911 operating on the computer 910 is a subsystem for managing the file system as described above, and a path management unit 915 for managing a path such as a directory structure and a storage device of which computer the file is in. A resource management unit 916 that manages the file, and a file management unit 917 that manages the physical location of the file in the storage device, and the subsystem 911 uses these to manage the file system.
[0010]
FIG. 13 is a diagram for explaining a read request and a data flow when the user program 901 reads a file in a conventional computer cluster system.
[0011]
(1) First, the user program 901 accesses the subsystem 911 that manages the file system and notifies the name of the file to be read. (2) The subsystem 911 searches in which space of which storage device of which computer the file with the name notified from the user program 901 is stored, and the kernel of the computer having the storage device in which the file exists Notify
[0012]
When notified to the kernel 902 (3) the kernel 902 notifies the storage device 903 of a read request. (4) The storage device 903 reads the contents of the file and notifies the kernel 902 of it. (5) The kernel 902 transfers the file contents to the subsystem 911. (6) The subsystem 911 transfers the contents of the file to the user program 901.
[0013]
On the other hand, when notified to the kernel 912 (3 ′), the kernel 912 notifies the storage device 913 of a read request. (4 ′) The storage device 913 reads the contents of the file and notifies the kernel 912 of it. (5 ′) The kernel 912 transfers the file contents to the subsystem 911. (6 ′) The subsystem 911 transfers the contents of the file to the user program 901.
[0014]
The write request and data flow at the time of writing the file are almost the same.
[0015]
As described above, in the conventional OS, both the logical structure and the physical structure of the file system are managed by the subsystem 911. Therefore, the reading and writing of the file is always performed once via the computer in which the subsystem is running. Therefore, especially when the computer that runs the user program and the computer that manages the storage device where the file exists are the same, it is efficient in order to send and receive data via a network that is slower than the data flow inside the computer. There was a problem.
[0016]
In particular, since (5) and (6) in FIG. 13 are data communications via a network, they are very inefficient compared to (4) or the like, which is communications within a computer.
[0017]
In addition, magnetic disks such as hard disks used as storage devices contain drive components inside, so they are very likely to fail compared to memory and processors that make up computers. Many methods are used to prevent file loss due to failure.
[0018]
Even in a conventional computer cluster system, data can be easily made redundant by having a file copy between storage devices of each computer. When the storage device fails, the process can be continued by using the replicated data. However, since the conventional computer cluster system reads and writes files via the subsystem, if data is made redundant, the overhead in making a copy of the file becomes very large and inefficient.
[0019]
Although only two computers have been described here, the opportunity to access a storage device connected to a computer different from the computer 910 on which the subsystem 911 operates increases as the number increases. In particular, in the case of having a copy of a file as described above, a case where both the copy source file and the copied file exist in a storage device connected to a computer different from the computer 910 on which the subsystem 911 operates. Occurs. Since the frequency increases in proportion to the number of computers, the efficiency also decreases in proportion to the number of computers.
[0020]
Therefore, an object of the present invention is to provide a computer cluster system that can efficiently read and write files.
[0021]
[Means for Solving the Problems]
In order to solve the above problems, a computer cluster system according to the present invention is a computer cluster system in which a plurality of computers each having a storage device and a communication device are commonly connected via a network. Exist in a distributed manner on the storage device of each computer, and some computers have a relationship between a path representing the logical structure of the file system and a global identifier uniquely assigned to the file in the entire computer cluster system. Each computer has a subsystem that has a path management unit for managing, and each computer has a kernel that controls the storage device that each computer has, and the kernel identifies which storage device the file is stored in A resource management unit for storing and managing a correspondence relationship between a local identifier and the global identifier; And a file management unit for managing the relationship between the address of the location where the file is stored as the local identifier within, the path management unit of the subsystems, operating on the computer cluster system User program If there is no global identifier corresponding to the file path notified by For each calculator The resource management unit newly stores a global identifier and local identifier pair corresponding to the file path, and the subsystem newly stores a global identifier and path pair corresponding to the file path, The user program acquires a global identifier corresponding to a path of the file.
[0022]
The computer cluster system of the present invention includes a storage device and a communication device. In a computer cluster system composed of a plurality of computers commonly connected via a network, the file system used in the entire computer cluster system is the same as that of each computer. An operating system that is distributed on a storage device and used in a computer cluster system has a relationship between a path that represents the logical structure of the file system and a global identifier that is uniquely assigned to the file throughout the computer cluster system. A local identifier that identifies a storage device in which the file is stored, and a global identifier that includes a subsystem having a path management unit for management and a kernel that controls a storage device included in each computer; For storing and managing the relationship A management unit, and a file management unit that manages the relationship between the address of the location where the file is stored in the storage device and the local identifier, and is operated in each computer. And operate on the computer cluster system in the path management unit of the subsystem. User program If there is no global identifier corresponding to the file path notified by For each calculator The resource management unit newly stores a global identifier and local identifier pair corresponding to the file path, and the subsystem newly stores a global identifier and path pair corresponding to the file path, The user program acquires a global identifier corresponding to a path of the file.
[0023]
The computer cluster system of the present invention includes a storage device and a communication device, and in a computer cluster system composed of a plurality of computers commonly connected via a network, the file system used in the entire computer cluster system is the same for each computer. Path management that exists in a distributed manner on a storage device and manages the relationship between the path that represents the logical structure of the file system and the global identifier that is uniquely assigned to the file in the entire computer cluster system. A local identifier that identifies which storage device the file is stored in, and a kernel that controls a storage device included in each computer is operated on each computer. A resource management unit for storing and managing a correspondence relationship with the global identifier; And a file management unit for managing the relationship between the address of the location where the file is stored as the local identifier within, the path management unit of the subsystems, operating on the computer cluster system User program If there is no global identifier corresponding to the file path notified by For each calculator The resource management unit newly stores a global identifier and local identifier pair corresponding to the file path, and the subsystem newly stores a global identifier and path pair corresponding to the file path, The user program acquires a global identifier corresponding to a path of the file.
[0024]
The file access method of the present invention is a file access method in a computer cluster system comprising a plurality of computers commonly provided with a storage device and a communication device and connected via a network, wherein the user program is a logic of the file system. An open step of notifying the subsystem that manages the physical structure of the file and obtaining the global identifier of the file; and the user program notifying the global identifier to the kernel operating on the same computer as the user program. A request step for requesting reading / writing, a file location search step in which the kernel converts the global identifier into a local identifier to check the location of a storage device in which a file to be read / written exists, and a result of the file location search step The computer on which the kernel is running When the computer to which the storage device in which the file exists belongs is different from the computer to which the file exists, a request transfer step for transferring the read / write request and the local identifier to the kernel operating in the computer to which the storage device in which the file exists belongs, and the file exists An access step in which a kernel operating on a computer to which the storage device belongs accesses the storage device indicated by the local identifier to execute a read / write request, and the kernel that has executed the access step sends the result of the access step to the user program A result reporting step of notifying, and in the open step, the user program notifies the subsystem of the file path when there is no global identifier corresponding to the path of the file in the path management unit of the subsystem. After that, the kernel A step of newly allocated and stores the set of the global identifier and the local identifier corresponding to the path of, Maintaining the same correspondence information between global identifiers and local identifiers of the kernel and kernels of other computers; and The subsystem includes a step of newly assigning and storing a global identifier corresponding to the path of the file and a path.
[0025]
The file access method of the present invention is a file access method in a computer cluster system comprising a plurality of computers commonly provided with a storage device and a communication device and connected via a network, wherein the user program is a logic of the file system. An open step of notifying the subsystem that manages the physical structure of the file and obtaining the global identifier of the file; and the user program notifying the global identifier to the kernel operating on the same computer as the user program. A request step for requesting read / write of the file, a result of the file location search step in which the kernel converts the global identifier into a local identifier to check the location of the storage device in which the file to be read / written exists, and the file location search step The computer on which the kernel is running When the computer to which the storage device in which the file exists belongs is different from the computer to which the file exists, a request transfer step for transferring the read / write request and the local identifier to the kernel operating in the computer to which the storage device in which the file exists belongs, and the file exists An access step in which a kernel operating on a computer to which the storage device belongs accesses the storage device indicated by the local identifier to execute a read / write request, and the kernel that has executed the access step sends the result of the access step to the user program A reporting result report step, and in the open step, when the path management unit of the subsystem does not have a global identifier corresponding to the path of the file, the user program sets the path of the file to the subsystem. After notification, the subsystem Assigning a global identifier corresponding to the path of Airu, and storing in association with the path, the steps of the kernel newly allocated and storing local identifier corresponding to the global identifier corresponding to the path of the file, Maintaining the same correspondence information between global identifiers and local identifiers of the kernel and kernels of other computers; and It is characterized by having.
[0037]
DETAILED DESCRIPTION OF THE INVENTION
(System Overview / Configuration) A computer cluster system according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram for explaining a computer cluster system according to an embodiment of this invention.
[0038]
In the computer cluster system of this embodiment, two computers, a computer 100 that executes a user program 101 and a computer 110 that executes a subsystem 111, are connected via a network 120.
[0039]
The computer 100 includes a storage device 105 that is a magnetic disk for storing data, and a communication device 106 that communicates with other computers via a network. In the computer 100, the storage device 105, the communication device 106, the kernel 102 which is a program responsible for hardware control of a memory (not shown), and the services provided by the computer cluster system and the subsystem 108 are used. A user program 101 that is a program such as an application to be executed is executed.
[0040]
The kernel 102 has an interface (not shown) for controlling hardware, a resource management unit 103 that manages resources such as files and processes used in the entire computer system, and the physical data stored in the storage device 105. And a file management unit 104 that manages a specific location (that is, a physical structure of the file system), and these can be accessed from outside the kernel 102 (for example, the user program 101) using a system call.
[0041]
The configuration of the computer 110 is the same as that of the computer 100 and includes a storage device 115 and a communication device 116. Further, on the computer 110, there is a subsystem 111 which is a program responsible for hardware control and the like, and has a kernel 112 having the same function as the kernel 102 and a program that provides services to all user programs including the user program 101. It is running.
[0042]
The configuration of the kernel 112 is the same as that of the kernel 102. In addition to an interface (not shown) for controlling hardware, a resource management unit 113 for managing resources such as files and processes used in the entire computer system, and a storage device 115 And a file management unit 114 for managing the physical location of the data stored therein, which can be accessed from outside the kernel 112 (for example, the subsystem 111) using a system call.
[0043]
The subsystem 111 is a program that manages the logical structure of a file system such as a directory structure, and is a character string (for example, “/home/user1/movie/gogo.mpg”) used to identify a file called a path. Is provided with a path management unit 117 that provides a function of converting a file identifier into a file identifier. The subsystem 111 stores and manages a relationship between a GID and a path for a file used in the entire computer cluster system among global identifiers (hereinafter referred to as GIDs) assigned to all resources used in the entire computer cluster system.
[0044]
The resource managers 103 and 113 assign the above-mentioned GID and local identifier (hereinafter referred to as LID) that is a unique identifier within the computer in which the resource actually exists to resources used in the entire computer cluster system. In addition, the correspondence between GID and LID is stored and managed. The LID has a structure in which the identifier of the computer itself in which the resource exists and the resource identifier in the computer are combined. For example, the LID of a file stored in the storage device 105 of the computer 100 has a structure such as “(identifier representing the computer 100): (identifier representing the storage device 105): (identifier of the file unique within the computer 100)”. become.
[0045]
In the present embodiment, when a change occurs in the correspondence relationship between one GID and LID, the resource management units 103 and 113 notify the other to the change so that the information on the correspondence relationship held by each other remains the same. . However, the management of correspondence information between GID and LID is not limited to this. For example, the resource management unit 103 may perform centralized management, and the resource management unit 113 may inquire when necessary, and notify the resource management unit 103 of the change contents when a change is made. Alternatively, inquiries may be made when a mutual management method is required for each other, for example, when a GID having an unknown correspondence with an LID is received. It is desirable that the method of synchronizing or exchanging information on the correspondence relationship between GID and LID between resource management units is appropriately changed according to the scale, design, and purpose of the computer cluster system.
[0046]
The file management unit 104 manages the physical location of the file stored in the storage device 105 connected to the kernel 102, and the file management unit 114 manages the physical location of the file stored in the storage device 115 connected to the kernel 112. Manage whereabouts. The physical location is a block number or address which is a recording unit of the storage devices 105 and 115 and is assigned to the file in a number corresponding to the data size. The file management units 104 and 114 manage LIDs and addresses of all blocks assigned to the file indicated by the LID.
[0047]
The storage devices 105 and 115 are magnetic disks that store files. It should be noted that a flash memory or the like is not limited to a magnetic disk. The storage device 105 passes file data to the file management unit 104 in response to a read request from the file management unit 104 of the kernel 102, and receives data received from the file management unit 104 in response to a write request from the file management unit 104. Remember. The storage device 115 passes file data to the file management unit 114 in response to a read request from the file management unit 114 of the kernel 112, and receives data received from the file management unit 114 in response to a write request from the file management unit 114. Remember.
[0048]
The communication devices 106 and 116 communicate with each other via the network 120. The communication device 106 transmits data to the communication device 116 in response to a request from the kernel 102, and passes the data received from the communication device 116 to the kernel 102. The communication device 116 transmits data to the communication device 106 in response to a request from the kernel 112, and passes the data received from the communication device 106 to the kernel 112. As the communication devices 106 and 116, for example, Ethernet (registered trademark), asynchronous communication mode (ATM), token ring, or the like can be used.
[0049]
(File Opening) The operation of the user program 101 opening a file will be described below using the flowchart of FIG.
[0050]
The user program 101 notifies the resource management unit 103 of an open request including the path of the file to be opened (step 201). The resource management unit 103 transfers the open request to the subsystem 111 (step 202). The subsystem 111 searches the GID corresponding to the path included in the open request from the table stored in the subsystem 111 (step 203). Then, the process branches depending on whether it is found (step 204).
[0051]
If found, the subsystem 111 includes the GID in the request result and notifies the resource management unit 103 (step 205). The resource management unit 103 notifies the request result to the user program 101 (step 206). Then, the user program 101 obtains a GID (Step 207).
[0052]
On the other hand, if not found, the subsystem 111 requests the resource management unit 103 described later to assign a GID (step 208). The resource management unit 103 newly assigns a GID, which will be described later (step 209), and notifies the user program 101 of the assignment result (step 210). Then, the user program 101 obtains a GID (Step 207).
[0053]
Thereafter, the user program 101 designates a target file using this GID when reading and writing the file.
[0054]
FIG. 3 is a flowchart for explaining the above-described steps 208 and 209, that is, the GID assignment request by the subsystem 111 and the new assignment of GID by the resource management unit 103.
[0055]
If it is determined that the GID corresponding to the path does not exist, the subsystem 111 analyzes the open request from the user program 101, checks whether or not new file creation is permitted (step 221), and branches the subsequent processing (step 221). Step 222).
[0056]
If there is no permission for new creation, the open result is a failure, so the resource management unit 103 is notified of the open failure (step 227). Since the open result is unsuccessful, the resource management unit 103 does not assign a GID (step 228), and notifies the user program 101 to that effect in step 210 described above.
[0057]
If there is permission for new creation, the subsystem 111 requests the resource management unit 103 to assign a new GID (step 223). The resource management unit 103 generates and stores a new set of GID and LID (step 224). Then, the generated GID and path are notified to the subsystem 111 to store the correspondence, and the GID and LID are notified to the resource management unit 113 as described above to update the resource information (step 225). Then, the assignment result is successful (step 226), and in step 210 described above, the success and the GID are notified to the user program 101.
[0058]
Although the file open has been described here, in the case of a file deletion request, processing is performed as follows.
[0059]
(1) The user program 101 notifies the resource management unit 103 of a deletion request, and the resource management unit 103 transfers the deletion request to the subsystem 111. (2) The subsystem 111 deletes the GID stored in the path management unit 117 and the corresponding path, and notifies the resource management unit 103 that the GID has been deleted. (3) Upon receiving the notification from the subsystem 111, the resource management unit 103 deletes the corresponding GID and the corresponding LID from the storage, and notifies the file management unit 104 that the LID has been deleted. (4) Upon receiving the notification from the resource management unit 103, the file management unit 104 deletes the relationship between the corresponding LID and the addresses of all blocks assigned to the file indicated by the corresponding LID from the storage.
[0060]
The notification of the deletion result from the resource management unit 103 to the user program 101 may be performed in parallel with the processing of (3), or after the processing of (3) or after the processing of (4). I do not care. When notifying the deletion result after completion of the process (4), the resource management unit 103 is notified of the deletion processing result of the file management unit 104.
[0061]
(File Writing) The operation of the user program 101 writing data to a file will be described below using the flowchart of FIG.
[0062]
The user program 101 notifies the resource management unit 103 of a file write request, GID, and data (step 301). The resource management unit 103 searches for an LID corresponding to the GID. If the computer indicated by the found LID is different from the computer 100 (ie, indicating the computer 110), the write request, LID, and data are transferred to the resource management unit of the computer indicated by the LID (step 302).
[0063]
If the LID indicates the computer 110, the write request, LID, and data are transferred to the resource management unit 113 of the computer 110 in step 302. The resource management unit 113 performs a data writing process as described later (step 303), and the resource management unit 113 notifies the resource management unit 103 of the result (step 304). Then, the resource management unit 103 notifies the user program 101 of the result of the writing process (step 306).
[0064]
When the LID indicates the computer 100, there is no need for transfer. The resource management unit 103 performs a writing process to be described later (step 305) and executes step 306.
[0065]
FIG. 5 is a flowchart of file write processing in the resource management unit 103. The resource management unit 113 performs similar processing.
[0066]
The resource management unit 103 notifies the file management unit 104 of the write request, LID, and data (step 311). The file management unit 104 searches for the physical location of the file indicated by the LID in the storage device 105 (step 312), and the storage area allocated to the file indicated by the LID is sufficiently larger than the size of data to be written. It is checked whether it is secured (step 313). If there is not a sufficiently large storage area, it is secured (step 314).
[0067]
Then, the file management unit 104 writes the data into the storage device 105 (step 315). The file management unit 104 notifies the write result to the resource management unit 103 (step 316).
[0068]
As described above, when the user program 101 writes data to a file, the subsystem 111 has always received and written data from the user program 101 in the past, but in the present embodiment, it is connected to the computer 100 on which the user program 101 operates. When the file to be written exists in the storage device 105, the file can be written without transmitting data via the network, which is efficient.
[0069]
(File Read) The operation of the user program 101 reading data from a file will be described below using the flowchart of FIG.
[0070]
The user program 101 notifies the resource management unit 103 of the read request and GID (step 401). The resource management unit 103 searches the LID from the GID. If the computer indicated by the found LID is different from the computer 100 (ie, indicating the computer 110), the read request and the LID are transferred to the resource management unit of the computer indicated by the LID (step 402).
[0071]
If the LID indicates the computer 110, the read request and the LID are transferred to the resource management unit 113 of the computer 110 in step 402. The resource management unit 113 performs a read process as described later (step 403), and the resource management unit 113 notifies the resource management unit 103 of the read result and data (step 404). Then, the resource management unit 103 notifies the read result and data to the user program 101 (step 406).
[0072]
When the LID indicates the computer 100, there is no need for transfer. The resource management unit 103 performs a read process described later (step 405) and executes step 406.
[0073]
FIG. 7 is a flowchart of file read processing in the resource management unit 103. The reading process in the resource management unit 113 is performed in the same manner.
[0074]
The resource management unit 103 notifies the file management unit 104 of the read request and LID (step 411). The file management unit 104 searches the physical location of the file indicated by the LID in the storage device 105 (step 412). The file management unit 104 reads data from the storage device 105 (step 413). The file management unit 104 notifies the resource management unit 103 of the data and the read result (step 414).
[0075]
As described above, when the user program 101 reads data from a file, conventionally, the subsystem 111 always reads the data from the storage device 105 and transfers the data to the user program 101. However, in this embodiment, the user program 101 If the file to be read exists in the storage device 105 connected to the computer 100 that operates, the file can be read without receiving data via the network, which is efficient.
[0076]
(Configuration of Resource Management Unit) FIG. 8 is a diagram for explaining the configuration of the resource management unit. The resource management unit 601 includes an ID conversion unit 602 that converts GID and LID, and a resource management table 603 that stores a correspondence relationship between GID and LID.
[0077]
The ID conversion unit 602 receives and processes a request notified from a user program or a subsystem and a request transferred from a resource management unit on another computer. If the ID included in the request is a GID, the GID is converted to an LID with reference to the resource management table 603. If the LID obtained by the conversion is an LID related to another computer, the request is transferred to the corresponding computer. Further, when a file is newly opened, a new GID and LID are generated and stored in the resource management table 603. When a file is deleted, the GID and LID are deleted.
[0078]
As described above, in the case of the present embodiment, when a change occurs in the resource management table 603, the change contents are notified to the resource management unit of another computer, and the same contents of the resource management table 603 are maintained.
[0079]
The resource management table 603 manages the LID corresponding to the resource represented by GID. The GID is numbered differently depending on the resource type (process, file, etc.). For example, in FIG. 8, GID starting with VP indicates a process, and GID starting with VF indicates a file. For example, it can be seen that the resource whose GID is VP0001 is P0001 of a process operating on the computer 100.
[0080]
(Configuration of File Management Unit) FIG. 9 is a diagram for explaining the configuration of the file management unit. The file management unit 701 includes a block position conversion unit 702 that converts the LID into a position on the storage device, and a file management table 703 that stores the correspondence between the LID and the position on the storage device.
[0081]
The block position conversion unit 702 uses the file management table 703 to obtain the LID included in the write or read request received from the resource management unit and the position information of the data in the file (information on the number of bytes from the beginning of the file). Convert to a location on the storage device. If the block size of the storage device is 64 KB, it can be seen that the 130 KB data from the beginning of the file with LID F0003 is stored in block 8 from the file management table 703.
[0082]
Note that the file management unit need not have the configuration shown in FIG. 8 as long as the location on the storage device can be known from the LID and the location information of the data in the file.
[0083]
(Configuration of Path Management Unit) FIG. 10 is a diagram for explaining the configuration of the path management unit of the subsystem. The path management unit 801 includes a path conversion unit 802 that provides a function of converting a path, which is a function of the file system, to a GID, and a path management table 803 that stores the correspondence between the path and the GID.
[0084]
The path conversion unit 802 receives a file open request from the resource management unit, converts the path included in the request into a GID using the path management table 803, and notifies the resource management unit of the result. Conversely, when the GID and path of the file generated by the resource management unit are received and stored in the path management table, or when a notification that the GID has been deleted is received from the resource management unit, The path management table is managed, such as deleting a GID and a path associated with the GID.
[0085]
For example, when the path is “/home/doc/document.txt”, the path management table 803 is searched in the order of “home”, “doc”, “document.txt” to obtain the GID. Get.
[0086]
First, data having no parent GID and having a name “home” is retrieved from data stored in the path management table 803 to obtain data 804.
[0087]
Next, using the GID “VF0001” included in the data 804, this time, the data having the parent GID “VF0001” and the name “doc” is retrieved to obtain the data 805.
[0088]
Similarly, data 805 is used to search for data whose parent GID is “VF0002” and whose name is “document.txt” to obtain data 806, and GID “VF0003]” of “document.txt” Is included in the result and notified to the resource management unit.
[0089]
If conversion is not possible due to circumstances such as the file does not exist, the open request is analyzed. If permission to create a new file is granted, the GID is requested to the resource management unit. If permission is not granted An error is returned to the resource manager.
[0090]
Note that the path management unit may not have the configuration of FIG. 9 as long as it has a function of converting a path to GID.
(Data Flow) FIG. 11 is a diagram for explaining the data flow when the user program 1001 reads data from the storage device. Although a communication device is omitted in the figure, communication between the computers 1000 and 1010 is performed via a network using the communication device.
[0091]
When the user program 1001 reads a file from the storage device 1003, the read request is transmitted (1) from the user program 1001 to the kernel 1002 (2) from the kernel 1002 to the storage device 1003. The read data is (3) transmitted from the storage device 1003 to the kernel 1002 and (4) transmitted from the kernel 1002 to the user program 1001.
[0092]
All communications from (1) to (4) are performed inside the computer. Since the communication inside the computer is much faster than the communication via the network, it can be understood that the time until the reading is completed is short and the processing efficiency is good. The same applies to writing.
[0093]
On the other hand, when the user program 1001 reads a file from the storage device 1013, the read request is transmitted from the user program 1001 to the kernel 1002 (2 ') and transferred from the kernel 1002 to the kernel 1012 (3') from the kernel 1012. It is transmitted to the storage device 1013. The read data is transmitted (4 ′) from the storage device 1013 to the kernel 1012 (5 ′) transferred from the kernel 1012 to the kernel 1002, and (6 ′) transmitted from the kernel 1002 to the user program 1001. The same applies to writing.
[0094]
When the computer on which the user program operates and the computer to which the storage device storing the file is different, the read / write efficiency is the same as before. However, when reading and writing from a storage device connected to the same computer, file contents are not communicated over the network as in the conventional case, so overhead is drastically reduced and efficiency is improved. The efficiency of.
[0095]
Although this embodiment has been described with only two computers, it is also possible to configure a computer cluster system using three or more computers. A plurality of user programs may be executed on different computers in the computer cluster system, and a part of the user programs may be executed on a computer on which the subsystem is executed. A plurality of subsystems may be executed on different computers in the computer cluster system. Furthermore, the functions of the subsystem may be divided and executed by being distributed over different computers in the computer cluster system.
[0096]
In this embodiment, the GID for the new file is assigned by the resource management unit 103 or 113, but may be assigned by the subsystem 111. The case where the resource management unit 103 is notified of the opening of the user program 101 file will be described as an example. If the file does not exist, the subsystem 111 allocates a GID, stores it in association with the path, and notifies the resource management unit 103 of it. . Then, the resource management unit 103 newly allocates and stores an LID corresponding to the GID.
[0097]
Up to this point, basic file read / write has been explained so far. The following is the operation when data redundancy is provided by creating a file copy on a storage device of a different computer. explain.
[0098]
In the case of a file having redundancy, a plurality of LIDs “computer 100: storage device 105: F0004” “computer 110: storage device 115: F0004” are associated with one GID, such as GID “VF0004” in FIG. Exists.
[0099]
In the file writing operation, all LIDs corresponding to GID are extracted in step 302. Then, the write request, the data, and the LID “computer 110: F0004” are transferred to the resource management unit 113 of the computer 110 and the resource management unit 113 performs the write process. In parallel with this, the resource management unit 103 also performs write processing. The resource management unit 103 executes step 306 after both writings are completed, and notifies the user program 101 of the writing result. It should be noted that step 306 may be executed at the stage where one writing is completed depending on the operation mode and policy of the computer cluster system.
[0100]
The file read operation is performed in the same manner as when one of a plurality of computers is selected and no redundancy is provided. When selecting a computer, it is most efficient to select a computer executing the user program that issued the request.
[0101]
When a failure occurs in the storage device, the resource management unit selects an LID of another storage device instead of the failed storage device.
[0102]
For example, when a failure occurs in the storage device 105, the resource management unit 103 that has received a read request for the file with the GID “VF0004” from the user program 101 changes the storage device 105 when converting from GID to LID. The “computer 100: storage device 105: F0004” is not selected, but “computer 110: storage device 115: F0004” is selected.
[0103]
As a result, the write operation executes steps 401, 402, 403, 404, and 406 in FIG. 6, and the write operation executes steps 301, 302, 303, 304, and 306 in FIG.
[0104]
In any case, the user program 101 may request reading and writing as in the case where there is no failure.
[0105]
In the computer cluster system according to the present embodiment, when the computer that is executing the user program or subsystem that requested reading / writing of the file is the same as the computer to which the storage device storing the file to be read / written is connected. Since the file can be read and written efficiently, the average file access efficiency of the entire system is improved. In addition, by executing reading and writing of files on the kernel side, it becomes difficult for communication to be concentrated on the computer on which the subsystem operates, so that the capacity, robustness and reliability of the entire system can be improved. In addition, even when the file is given redundancy, the file can be read and written efficiently as in the case where the file is not given redundancy.
[0106]
【The invention's effect】
As described above, the computer cluster system of the present invention can improve the read / write efficiency by dividing the file system functions and implementing path management in the subsystem and resource management and file management in the kernel.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a configuration of a computer cluster system according to an embodiment of this invention.
FIG. 2 is a flowchart for explaining a file open operation in the computer cluster system according to the embodiment of this invention.
FIG. 3 is a flowchart for explaining a GID assignment operation in steps 208 and 209 in FIG. 2;
FIG. 4 is a flowchart for explaining a file writing operation in the computer cluster system according to the embodiment of this invention.
FIG. 5 is a flowchart for explaining a file writing operation in a resource management unit.
FIG. 6 is a flowchart for explaining a file read operation in the computer cluster system according to the embodiment of this invention.
FIG. 7 is a flowchart for explaining a file read operation in the resource management unit.
FIG. 8 is a diagram illustrating a configuration of a resource management unit.
FIG. 9 is a diagram illustrating a configuration of a file management unit.
FIG. 10 is a diagram illustrating the configuration of a subsystem.
FIG. 11 is a diagram illustrating an example of a data flow when reading a file in the computer cluster system according to the embodiment of this invention.
FIG. 12 is a diagram for explaining the configuration of a conventional computer cluster system.
FIG. 13 is a diagram for explaining an example of a data flow when reading a file in a conventional computer cluster system.
[Explanation of symbols]
100, 110 computer
101 User program
102, 112 kernel
103, 113 Resource Management Department
104, 114 File management unit
105, 115 Storage device
106, 116 Communication device
111 subsystem
117 Path management department
120 network
601 Resource Management Department
602 ID converter
603 Resource management table
701 File Management Department
702 Block position conversion unit
703 File management table
801 Subsystem (path management unit)
802 Path conversion unit
803 Path management table
900, 910 calculator
901 User program
902, 912 kernel
903, 913 storage device
904, 914 communication device
915 Path Management Department
916 Resource Management Department
917 File Management Department
920 network

Claims (16)

In a computer cluster system in which a plurality of computers each having a storage device and a communication device are connected in common via a network,
The file system of the entire computer cluster system is distributed on the storage device of each computer,
Some computers have a subsystem having a path management unit that manages a relationship between a path representing the logical structure of the file system and a global identifier that is uniquely assigned to the file in the entire computer cluster system.
Each computer has a kernel that controls the storage device that each computer has,
The kernel stores and manages a correspondence relationship between a local identifier for identifying a storage device in which a file is stored and the global identifier, and an address of a location where the file is stored in the storage device. A file management unit for managing a relationship with the local identifier,
When the path management unit of the subsystem does not have a global identifier corresponding to the file path notified by the user program operating on the computer cluster system, the resource management unit of each computer corresponds to the file path. A new set of global identifiers and local identifiers to be stored, the subsystem newly stores a set of global identifiers and paths corresponding to the path of the file, and the user program corresponds to the path of the file A computer cluster system characterized in that a global identifier is obtained. In a computer cluster system comprising a plurality of computers that are commonly connected via a network, including a storage device and a communication device,
The file system used in the entire computer cluster system is distributed on the storage device of each computer,
The operating system used in the computer cluster system is
A subsystem having a path management unit that manages a relationship between a path representing a logical structure of the file system and a global identifier uniquely assigned to the file in the entire computer cluster system;
A kernel for controlling the storage device of each computer,
The kernel stores and manages a correspondence relationship between a local identifier for identifying a storage device in which a file is stored and the global identifier, an address of a location where the file is stored in the storage device, and the A file management unit that manages the relationship with the local identifier, and operates on each computer,
The subsystem operates on some computers,
When the path management unit of the subsystem does not have a global identifier corresponding to the file path notified by the user program operating on the computer cluster system, the resource management unit of each computer corresponds to the file path. A new set of global identifiers and local identifiers to be stored, the subsystem newly stores a set of global identifiers and paths corresponding to the path of the file, and the user program corresponds to the path of the file A computer cluster system characterized in that a global identifier is obtained. In a computer cluster system comprising a plurality of computers that are commonly connected via a network, including a storage device and a communication device,
The file system used in the entire computer cluster system is distributed on the storage device of each computer,
In some computers, a subsystem having a path management unit that manages a relationship between a path representing the logical structure of the file system and a global identifier uniquely assigned to the file in the entire computer cluster system is operated.
On each computer, run the kernel that controls the storage devices of each computer,
The kernel stores and manages a correspondence relationship between a local identifier for identifying a storage device in which a file is stored and the global identifier, and an address of a location where the file is stored in the storage device. A file management unit for managing a relationship with the local identifier,
When the path management unit of the subsystem does not have a global identifier corresponding to the file path notified by the user program operating on the computer cluster system, the resource management unit of each computer corresponds to the file path. A new set of global identifiers and local identifiers to be stored, the subsystem newly stores a set of global identifiers and paths corresponding to the path of the file, and the user program corresponds to the path of the file A computer cluster system characterized in that a global identifier is obtained. If the path management unit of the subsystem does not have a global identifier corresponding to the path of the file notified by the user program operating on the computer cluster system, the resource management unit can determine the global identifier corresponding to the path of the file. 4. The computer cluster system according to claim 1, wherein a set of a local identifier is newly allocated. If the path management unit of the subsystem does not have a global identifier corresponding to the file path notified by the user program operating on the computer cluster system, the subsystem assigns a global identifier corresponding to the file path. 4. The resource management unit newly assigns and stores a local identifier corresponding to a global identifier corresponding to the path of the file, and stores the information in association with a path. The computer cluster system described in the section.   When a change occurs in the correspondence between the global identifier and the local identifier of the resource management unit, the correspondence information between the global identifier and the local identifier between the resource management unit and the resource management unit of another computer is kept the same. The computer cluster system according to claim 1, wherein: The resource management unit
7. The computer cluster system according to claim 1, wherein a global identifier is received from the kernel and a local identifier corresponding to the global identifier is output. The resource management unit
8. The computer cluster system according to claim 1, wherein information on a correspondence relationship between the local identifier and the global identifier is acquired from a resource management unit of another computer constituting the computer cluster system. The resource management unit
9. The computer cluster system according to claim 1, wherein information on a correspondence relationship between the local identifier and the global identifier is notified to a resource management unit of another computer constituting the computer cluster system. The kernel is
A file path received from a user program operating on a computer cluster system and a file open request are notified to the subsystem to obtain a global identifier corresponding to the path,
10. The computer cluster system according to claim 1, wherein a global identifier obtained is returned to the user program. The kernel is
Notifying the resource management unit of the global identifier received from the user program and a file read request to convert the global identifier into the local identifier,
The local identifier obtained by conversion is converted into an address in a storage device by the file management unit,
Access the storage device using the address obtained by conversion, read the data,
11. The computer cluster system according to claim 1, wherein the read data is returned to the user program. The kernel is
The global identifier and data to be written to the file and a file write request are received from the user program
Converting the global identifier into the local identifier in the resource management unit;
The file management unit converts the local identifier into an address in a storage device,
12. The computer cluster system according to claim 1, wherein data is written by accessing the storage device. The kernel is
When a file write request is received from a program running on a computer cluster system,
Write files to multiple storage devices connected to different computers in the computer cluster system,
13. The computer cluster system according to claim 1, wherein when any of the plurality of storage devices fails, file access is performed using a file of another storage device. The subsystem is
Receiving a path representing the logical structure of the file system from the kernel;
14. The computer cluster system according to claim 1, wherein a global identifier corresponding to the path is output. A file access method in a computer cluster system including a storage device and a communication device and comprising a plurality of computers commonly connected via a network,
An open step in which the user program notifies the subsystem that manages the logical structure of the file system of the file path and obtains the global identifier of the file;
A request step in which a user program notifies the kernel operating on the same computer as the user program to request reading and writing of the file by notifying the global identifier;
A file location search step in which the kernel converts the global identifier into a local identifier and checks the location of a storage device in which a file to be read and written exists;
As a result of the file location search step, when the computer in which the kernel is operating differs from the computer to which the storage device in which the file exists belongs, the kernel requests a read / write request to the kernel operating in the computer in which the storage device in which the file exists And a request transfer step for transferring the local identifier;
An access step in which a kernel operating on a computer to which the storage device in which the file exists belongs accesses the storage device indicated by the local identifier and executes a read / write request;
A kernel that has executed the access step has a result reporting step of notifying the user program of the result of the access step;
In the open step, when there is no global identifier corresponding to the file path in the path management unit of the subsystem, the user program notifies the subsystem of the file path,
The kernel newly assigning and storing a global identifier and local identifier pair corresponding to the path of the file;
A step of maintaining the information of the correspondence relationship between the global identifier and the local identifier of the kernel of the kernel and other computers on the same,
A file access method comprising the step of: the subsystem performing a new assignment of a global identifier and a path corresponding to the path of the file and storing them. A file access method in a computer cluster system including a storage device and a communication device and comprising a plurality of computers commonly connected via a network,
An open step in which the user program notifies the subsystem that manages the logical structure of the file system of the file path and obtains the global identifier of the file;
A request step in which a user program notifies the kernel operating on the same computer as the user program to request reading and writing of the file by notifying the global identifier;
A file location search step in which the kernel converts the global identifier into a local identifier and checks the location of a storage device in which a file to be read and written exists;
As a result of the file location search step, when the computer in which the kernel is operating differs from the computer to which the storage device in which the file exists belongs, the kernel requests a read / write request to the kernel operating in the computer in which the storage device in which the file exists And a request transfer step for transferring the local identifier;
An access step in which a kernel operating on a computer to which the storage device in which the file exists belongs accesses the storage device indicated by the local identifier and executes a read / write request;
A kernel that has executed the access step has a result reporting step of notifying the user program of the result of the access step;
In the open step, when there is no global identifier corresponding to the file path in the path management unit of the subsystem, the user program notifies the subsystem of the file path,
The subsystem assigns a global identifier corresponding to the path of the file and stores it in association with the path;
The kernel newly assigning and storing a local identifier corresponding to a global identifier corresponding to the path of the file;
Maintaining the same correspondence information between global identifiers and local identifiers of the kernel and kernels of other computers; and
A file access method comprising:

Images