JP3614886B2 - File system - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to an improvement of a file system for storing data such as a file, and particularly relates to a file system having excellent reliability.
[0002]
[Prior art]
Most computers have an internal storage device and an external storage device. The internal storage device keeps information by continuous power supply from the outside. A typical internal storage device is a main memory composed of RAM. The external storage device has some kind of recording medium (for example, a magnetic disk or a magneto-optical disk) regardless of the principle or type. Regardless of whether the recording medium is detachable or not, a single external storage device is called a drive. A large number of blocks are usually formed on a drive medium. This block is a unit of access.
[0003]
For example, a standard floppy disk has two surfaces, each surface having 80 tracks arranged concentrically, and each track is divided into 16 sectors. Each sector or several consecutive sectors corresponds to the block. In the fixed disk device, the recording medium is a hard disk stacked in a cylindrical shape. For this reason, the fixed disk device has many surfaces. The term cylinder is used instead of truck.
[0004]
In most cases, a single operation that updates information on the disk updates multiple blocks on the media. The first reason is that the block size (eg 256 bytes) is usually much smaller than the amount of information in a single operation. For example, a document edited on a computer, which is only a few pages, is saved by spending dozens of blocks and sectors. The second reason is that data on a medium is generally accessed in file units, and file management information is divided (stored) like a FAT or a directory, which will be described later.
[0005]
The file system is a concept generically referring to an external storage device and a control procedure for accessing data in the external storage device in file units. The file system requires management information (for example, FAT (File Allocation Table) and directory). For this reason, when updating a file, it is necessary to update the FAT and the directory in addition to the data that is the contents of the file even if the file is extended by only one sector. As a result, often a single operation updates the contents of multiple blocks.
[0006]
In the file system, a buffer is usually used. The buffer is a predetermined area on the memory that temporarily stores data to be written to the medium. The buffer may be a main memory or may be provided on a drive driving sub-computer. For example, in a system in which a disk and a buffer are combined, the contents of the block to be changed are temporarily not stored in the disk but temporarily stored in the buffer when the data on the disk is changed. Block data written in the buffer (hereinafter referred to as “buffer block”) should be written to the medium at least before shutting down the system. If actual disk access is performed after such buffer blocks have accumulated to some extent, the number of disk accesses can be reduced and the processing can be made more efficient.
[0007]
If the buffer is full with data that has not yet been written to disk, and new data cannot be written to the buffer, the media is updated with the insufficient buffer blocks, and new data is written to the updated area of the buffer. .
[0008]
[Problems to be solved by the invention]
A file system using the above-mentioned buffer is common, but even when such a file system is used, in the past, as a hardware for realizing a single logical drive of a single computer, it is actually a single unit. Only drive of was connected. Computers with only a single drive often result in data loss due to drive failure. There are various drive failures such as power interruption, head crash, and control circuit failure. The most unfortunate consequence of a failure is that the failure occurs during data writing and the medium is in a so-called “write” state.
[0009]
Here, “write over” represents a state in which only a part of the plurality of blocks to be updated at the same time is updated, and the rest has old contents. For example, generally, data in a file and information on the FAT and directory of this file should not contradict each other. However, if a drive fails while any of these has not been updated, they will lose consistency with each other and file data will be lost or difficult to restore. As another example of “writing”, for example, in a state where only the FAT has been written to the disk and no directory has been written, inconsistency occurs in the management information itself. In addition, when a continuous block forms part of a single file, the content of one block is updated and the content of the next block remains old. is there. Naturally, the consistency between the old data and the new data is lost.
[0010]
Recovering file data that has lost consistency due to a mix of new and old content is much more difficult than recovering file data that remains entirely old. One cause of this difficulty is the difficulty in identifying the old and new parts. Another cause is destruction of data on the medium by a head that has malfunctioned (particularly when the drive is a magnetic disk). This is because the magnetic field for writing generated from the head is always improper from the moment of failure, or the head damages the magnetic surface. Furthermore, a situation may occur in which the entire file is lost due to the loss of the file management information.
[0011]
In particular, if the computer system file is unfortunately lost, the computer system cannot be started, and failure diagnosis and recovery itself are impossible.
[0012]
The tolerance for such data loss varies from field to field. For example, in the fields of form calculation, organization of experiment data, and document creation, it is convenient to redo old data. In such a field, fault tolerance is not so required. On the other hand, in fields such as large plant control, transportation system control, and bank account management, data is not allowed to disappear and cannot be redone. In the latter area, loss of data immediately leads to danger and disruption of rights and obligations, so fault tolerance is highly required.
[0013]
In recent years, with the downsizing of computer systems, various systems that were conventionally used only in fields where fault tolerance is not required are also used in fields where high fault tolerance is required. It has become like this. For example, a UNIX file system that is widely used in small computers is a file storage device that does not consider fault tolerance, but has come to be used in fields that require high fault tolerance. ing. For this reason, improvement in fault tolerance is an important issue.
[0014]
Typical techniques for preventing file loss and improving fault tolerance are mirroring and the distributed file system. Mirroring is a technology that connects multiple drives to a single computer and writes exactly the same data to each drive. The distributed file system is a system in which three or more drives are connected to a computer and files having the same contents are always stored on two or more drives. In this way, if files with the same contents are stored in multiple drives and the files are multiplexed, even if some of the drives fail, the files remaining on the remaining drives can be used to restore work and files. Is possible.
[0015]
However, the following problems remain in file multiplexing. The problem is that data can be “written” in a plurality of drives at the same time. This is because, conventionally, writing to a plurality of drives can occur simultaneously. This is because, conventionally, the unit of the buffer block for updating the drive block is performed regardless of the unit of operation for the file of the drive.
[0016]
For example, when operating two buffer blocks on two drives in a single operation, it often happens that only one block is updated on both drives and the other block is not updated. . A particularly unlucky condition is that a “write over” failure occurs on all drives with the same file. In this case, all of the plurality of files having the same contents are lost. For this reason, it is necessary to eliminate the risk of losing all files and improve the reliability of the file system.
[0017]
The present invention has been proposed in order to solve the above-described problems of the prior art, and an object of the present invention is to provide a highly reliable file system. A more specific object is to provide a file system in which inconsistencies between blocks due to “writing” are unlikely to occur. Another specific object is to provide a file system in which at least one of a plurality of files having the same contents is not lost even in the case of a drive failure.
[0018]
Another object of the present invention is to provide a file system that efficiently processes data. Another object of the present invention is to provide a file system having a simple configuration. Another object of the present invention is to provide a file system in which the influence of a failure is localized and hardly reaches other parts. Another object of the present invention is to provide a file system that is highly reliable and easy to introduce into a conventional file system.
[0019]
[Means for Solving the Problems]
In order to achieve the above object, the file system of claim 1 comprises:Each drive has a plurality of blocks as access units on the recording medium of each drive, and data to be written to each block in a buffer on the volatile memory. A buffer means for temporarily storing a buffer block; a judgment means for judging the necessity of updating the recording medium by the buffer block; and when the update is judged to be necessary, all the buffer blocks An update means for continuously updating and a serialization means for serializing the updates while avoiding temporal overlap of the updates in each drive are characterized.
[0020]
The file system according to claim 2, wherein when the serialization unit tries to perform the update in one drive, it checks whether another drive being updated exists, and another drive being updated exists. In this case, updating in the one drive is suppressed, and updating in the one drive is performed when there is no other drive being updated.
The file system according to claim 3 is configured such that, when there is a drive being updated, the serializing unit performs a check again after waiting for a predetermined time, and updates when the drive being updated does not exist. It is characterized by that.
In the file system according to claim 4, the serialization unit attempts to update by setting a flag indicating the state of each drive to “updating”, “under investigation”, and other states according to the state of the drive It is sometimes configured to check other drive flags.
The file system according to claim 5 is characterized in that the serializing means is provided individually for each drive.
[0021]
The file system according to claim 6 is characterized in that each of the serializing means is configured to perform the investigation by collecting information on the state of another drive via a communication line.
The file system according to claim 7, wherein the serializing unit transfers and copies the flag of each drive to all other drives as a copy, and the other drive refers to the copy flag to perform the investigation. It is characterized by being configured.
9. The file system according to claim 8, wherein the serializing means is configured to perform the serialization by limiting updating of each drive within a predetermined time cycle based on a predetermined clock. And
The file system according to claim 9 has a plurality of drives, and each file has a plurality of drives. It is recorded above.
[0022]
The file system according to claim 10 includes a plurality of non-volatile storage device drives, and each drive has a plurality of blocks as access units on a recording medium of each drive and a buffer on a volatile memory. Each time a single data write operation requiring a block update to the drive occurs, the buffer block, which is the data to be written to each block, is written to each block for each single data write operation. The buffer means for temporarily storing the buffer block in the buffer on the volatile memory, the judgment means for judging the necessity of updating the recording medium by the buffer block, and the judgment means judged that the update is necessary All buffer blocks subject to at least the single data write operation Therefore the updating means for updating successively a block of the drive, characterized by having a serialization means for serializing updates to avoid temporal overlap of the update in each drive.
[0023]
12. The file system according to claim 11, wherein the buffer means includes the identification information for associating the generated single data write operation with all the buffer blocks that are targets of the write operation, and the buffer block to the buffer. And the update means stores all buffer blocks corresponding to a single data write operation based on the identification information when the update is determined to be necessary by the determination means. The updating is performed as a unit.
The file system according to claim 12 is configured such that, when the same buffer block is included in a plurality of data write operation targets, the update unit continuously updates all the buffer blocks related to the plurality of data write operations. It is structured.
[0024]
The file system according to claim 13, wherein when the serialization means tries to perform the update in one drive, it checks whether another drive being updated exists, and another drive being updated exists. In this case, updating in the one drive is suppressed, and updating in the one drive is performed when there is no other drive being updated.
The file system according to claim 14 is configured such that the serializing means performs an investigation again after waiting for a predetermined time when there is a drive being updated, and updates when the drive being updated does not exist. It is characterized by that.
15. The file system according to claim 15, wherein the serialization means sets a flag indicating the state of each drive to “updating”, “investigation”, and other states according to the state of the drive, and tries to update it. It is sometimes configured to check other drive flags.
[0025]
The file system according to a sixteenth aspect is characterized in that the serializing means is individually provided for each drive.
The file system according to claim 17 is characterized in that each of the serializing means is configured to perform the investigation by collecting information on the state of another drive via a communication line.
The file system according to claim 18, wherein the serializing means transfers and copies the flag of each drive as a copy to all other drives, and the other drive refers to the copy flag to perform the investigation. It is characterized by being configured.
20. The file system according to claim 19, wherein the serializing means is configured to perform the serialization by limiting updating of each drive within a predetermined time cycle based on a predetermined clock. And
The file system according to claim 20 has a plurality of drives, and each file is recorded on the plurality of drives.
[0026]
[Action]
Claims 1 and 10In the present invention, when updating the medium, all the buffer blocks are continuously updated. Therefore, the state of “writing” is limited to only during the update, and a block of the file is updated between the updates. No inconsistency occurs. For this reason, even if the drive fails at an instant other than during the update, the consistency between the blocks in the file is not lost. According to the first aspect of the present invention, since updating on the buffer is performed by all the buffer blocks, a complicated process for distinguishing the buffer blocks for each operation becomes unnecessary. In addition, since all the buffer blocks are updated at one time, if the update is performed when the CPU load is low, the possibility that the update is necessary when the CPU load is high can be reduced. Furthermore, since the serialization means shifts the update timing, the “writing” state does not occur simultaneously in a plurality of drives. For this reason, data of a plurality of drives are not lost at the same time, and data remains in at least one drive.
[0027]
According to the second and thirteenth aspects of the present invention, the serialization means investigates every update and detects other drives that are being updated, thereby avoiding the time overlap of the updates, so that reliable serialization can be performed. Further, the invention of claim 4 can be realized by giving the control for starting the update to the serialization means, and the hardware and procedure for performing the update itself can be used, so that the conventional equipment can be used effectively.
[0028]
According to the third and fourteenth aspects of the present invention, since the serialization means realizes serialization by waiting for a predetermined time, a supervisor for allocating update timing is unnecessary.
[0029]
According to the fourth and fifteenth inventions, even if the hardware of each drive is different, the serializing means can detect the drive being updated or investigated as long as it refers to the flags of other drives. Further, when each drive is provided with serialization means, the side to be updated refers to the flag. For this reason, it is not necessary to provide an upper supervisor that controls each drive.
[0030]
According to the fifth and sixteenth aspects of the present invention, since the serializing means is distributed to each drive, even if there is a failed drive, other drives are not affected. Further, when the serializing means is configured as a superior supervisor for controlling the drive, the processing in each drive is stopped due to the failure of the serializing means. It can be avoided.
[0031]
According to the sixth and 17th aspects of the present invention, if there is only a communication line, the serialization means uses this communication line to conduct an investigation, so that no flag or supervisor is required.
[0032]
In the inventions of claims 7 and 18, the serializing means of each drive does not need to refer to the state of another drive or the flag provided in the other drive every time it is investigated, so the processing at the time of update Is faster.
In the inventions of claims 8 and 19, the update opportunity in each drive is surely given in a predetermined cycle, and the case where the update in a specific drive is delayed due to the continuation of the update in another drive is eliminated.
[0033]
The inventions of claims 9 and 20 are so-called distributed file systems. In the present invention, which file normally exists for each drive is different. For this reason, the timing at which the buffer of each drive becomes full is dispersed, and the waiting time for drive update is reduced. That is, if the contents of all drives are the same as in mirroring, Since the timing of writing to be competed in time, a wasteful waiting time occurs, but the occurrence of such a waiting time is reduced in the present invention.
[0034]
Claim 11In this invention, since the correspondence relationship can be easily specified by recording the identification information such as the code number for each operation and the buffer block, the operation unit can be determined with a simple configuration without using a special data structure. Update can be realized.
[0035]
Claim 12In this invention, when a plurality of operations are targeted for the same block, the block is not updated only by the contents of some of the operations. There is no loss.
[0036]
【Example】
Next, embodiments of the present invention will be specifically described with reference to the drawings. In addition, the Example mentioned later is implement | achieved on a computer, and each function of an Example is implement | achieved when a predetermined procedure (program) controls this computer.
[0037]
Each “means” in the present specification is a conceptual one corresponding to each function of the embodiment, and does not necessarily correspond to a specific hardware or software routine on a one-to-one basis. The same hardware element constitutes different means depending on the case. For example, a computer can be one means when executing a certain instruction and another means when executing another instruction. One means may be realized by only one instruction, or may be realized by a large number of instructions.
[0038]
Therefore, in the present specification, the embodiment will be described below assuming a virtual circuit block (means) having each function of the embodiment. However, the use of a computer is an example, and all or part of the functions of the present invention may be realized on an electronic circuit such as a custom chip (dedicated integrated circuit) if possible.
[0039]
The computer used in the embodiments generally includes a CPU (Central Processing Unit) and a main storage device including a RAM (A Read / Write memory element as needed). The scale of the computer is arbitrary, and any scale such as a microcomputer, personal computer, small computer, workstation, mainframe, etc. may be used.
[0040]
The computer hardware typically includes an input device such as a keyboard and a mouse, an external storage device such as a hard disk device, an output device such as a CRT display device and a printer printer, and necessary input / output control. Includes circuitry.
[0041]
However, the hardware configuration of the computer is free, and as long as the present invention can be implemented, some of the above components may be added, changed, or excluded. For example, the embodiment may be realized on a computer network in which a plurality of computers are connected. The type of CPU is arbitrary, and a plurality of CPUs may be used simultaneously, or a single CPU may be used for time sharing (time division) to perform a plurality of processes simultaneously.
[0042]
Other input devices (for example, a pointing device such as a touch panel, a light pen, and a trackball, an image input device such as a digitizer, an image reading device, and a video camera, a voice identification device, and various sensors) may be used. Other external storage devices (for example, floppy disk devices, RAM card devices, magnetic tape devices, optical disk devices, magneto-optical disk devices, bubble memory devices, flash memories, etc.) may be used. Further, other output devices (for example, a liquid crystal display device, a plasma display device, a video projector, an LED display device, a sound generation circuit, a voice synthesis circuit, etc.) may be used.
[0043]
Further, as a configuration of software for realizing the embodiment in the computer, typically, an aspect in which an application program for realizing each function of the embodiment is executed on an OS (operating system) is considered. It is done. Further, as a program mode for realizing the embodiment, typically, a machine language compiled (translated) from a high-level language or an assembler can be considered. However, the software configuration of the computer is also free, and the software configuration may be changed as long as the present invention can be implemented. For example, it is not always necessary to use the OS, the program expression format is also free, and an interpreter (sequential interpretation execution type) language such as BASIC may be used.
[0044]
The program can be stored in any manner, and may be stored in a ROM (read only memory), or stored in an external storage device such as a hard disk device, when the computer is started or processing is started. Sometimes it may be loaded into the main memory. Alternatively, the program may be divided into a plurality of parts and stored in an external storage device, and only necessary modules may be loaded (read) onto the main memory as needed according to the processing contents. Furthermore, each program part may be stored in a different manner.
[0045]
In addition, as long as each step of each procedure in the present embodiment is not contrary to its nature, the execution order may be changed, and a plurality of steps may be executed simultaneously, or may be executed in different orders for each execution. Such a change in the order can be realized by a menu-type interface method such as selecting a process executable by the user.
[0046]
Further, “input” in the present specification includes not only original information input but also other processing closely related to information input. Such processing is, for example, echo back of input content or correction / editing. In addition, “output” in the present specification includes not only original information output but also other processing closely related to information output. Such a process is, for example, an input of a range to be output or a screen scroll instruction. Note that input and output may be realized by an integrated operation by an interactive input / output procedure, and processing such as selection / designation / specification may be performed by such an integrated operation.
[0047]
Further, data (information) and data storage means in this specification may exist in any manner on the computer. For example, the location of the data on the hardware may be any part such as a main storage device, an external storage device, a CPU register, or a cache memory. The data holding mode is also free. For example, data is not only held in a file format, but may be realized by directly accessing a storage device such as a memory or a disk with a physical address. The data expression format is also free. For example, the unit of the code representing the character string may be a character unit or a word unit. In addition, it is sufficient that the data is held for a certain period of time as required, and it may disappear thereafter, and the length of the holding time is arbitrary. Information that is not changed for the time being, such as dictionary data, may be stored in the ROM.
[0048]
Further, in this specification, reference to specific information is confirming and does not deny the existence of information that is not mentioned. That is, in the operation of the present invention, general information necessary for the operation, for example, various pointers, counters, flags, parameters, buffers, and the like are appropriately used.
[0049]
Information required for processing by each part of the embodiment is obtained from other parts holding the information unless otherwise specified. Such acquisition of information can be realized, for example, by accessing a variable or memory storing the information. Note that the information can be erased / erased by changing the meaning of the information, such as setting a flag indicating erasure, without actually deleting the content of the information itself from the storage area.
[0050]
1. Configuration of the first embodiment
FIG. 1 is a block diagram showing the components of the first embodiment. In this case, the system shown in FIG. 1 connects a plurality of file storage devices 1S, 2S,..., NS (n is an arbitrary integer greater than or equal to 2) and the plurality of file storage devices 1S, 2S,. And a common system bus 10. Each of the file storage devices 1S, 2S,..., NS corresponds to a drive referred to in the claims.
[0051]
The file storage device 1S includes a flash serialization unit 1A (corresponding to the serialization unit described in the claims), a flash state flag 1B (corresponding to the flag in the claims), and a buffer management unit (buffer flash unit) 1C. , A disk buffer 1D, and a disk 1E. The disk 1E corresponds to a recording medium referred to in the claims. Further, the buffer management means 1C and the disk buffer 2D constitute buffer means referred to in the claims. The buffer management unit 1C also serves as a determination unit and an update unit in the claims. In this embodiment, “flash” is a process of updating a drive block by all blocks on the buffer.
[0052]
Among these, the flash serializing means 1A responds to the flush start request from the buffer management means 1C and sets the flash status flag 1B to “investigation” and checks the flash status flags of other file storage devices. Then, the flash serialization means 1A sets the flash status flag 1B to “update (execution)” when there is no “under investigation” or “running” flag in another file storage device, and requests A flash start permission is returned to the original buffer management means 1C. Further, the flash status flag 1B is a flag indicating the flash status of the file storage device 1S. As described above, the flash serialization unit 1A performs the flash serialization unit 1A during the investigation by the flash serialization unit 1A or the flash management unit 1C. “Under investigation” or “Updating (executing)”, otherwise “Normal”.
[0053]
On the other hand, the buffer management means 1C sends a flush start request to the flash serialization means 1A in response to an external operation request for the file storage device 1S. Then, as described above, when the buffer management means 1C receives the flush start permission from the flash serialization means 1A, in response to this, all the data in the disk buffer 1D that will need to be updated in the future is used. 1E is updated.
[0054]
Next, the file storage device 2S includes a flash serialization unit 2A, a flash state flag 2B, a buffer management unit (buffer flash unit) 2C, a disk buffer 2D, and a disk 2E. These components 2A to 2E are also configured in the same manner as the components 1A to 1E of the file storage device 1S. Similarly, the nth file storage device nS has a flash serialization means nA, a flash status flag nB, a buffer management means (buffer flush means) nC, a disk buffer nD, and a disk nE, and these components nA ... To nE are configured in exactly the same way as the components 1A to 1E of the file storage device 1S.
[0055]
2 is a conceptual diagram showing an example of a specific configuration of the system shown in FIG. 1. The file storage device 1S includes a CPU (1F), a ROM (1G), a RAM (1H), a SCSI (1I), and a disk. 1E and a local bus 1J. In the ROM (1G) and RAM (1H), buffer management means 1C, flash serialization means 1A, flash status flag 1B, and disk buffer 1D are configured. Similarly, the file storage device 2S includes a CPU (2F), a ROM (2G), a RAM (2H), a SCSI standard interface circuit (2I), a disk 2E, and a local bus 2J. In the RAM (2H), buffer management means 2C, flash serialization means 2A, flash status flag 2B, and disk buffer 2D are configured.
[0056]
The first embodiment is a so-called distributed file system in which there are three or more drives and each file is recorded on two or more drives. In the present invention, which file is usually assigned to each drive. Is different. For this reason, the timing at which the buffer of each drive becomes full is dispersed, and the waiting time for drive update is reduced. When serialization is performed in the same mirroring where the contents of all drives are the same as in mirroring, a need for writing to a plurality of drives always occurs at the same time, resulting in unnecessary waiting time.
[0057]
In the first embodiment, each drive is a separate and independent disk device, and each medium is a recording disk of each disk device. For this reason, in the first embodiment, the possibility of losing a plurality of files due to a failure of a single drive can be eliminated as compared with a partition in which a single disk is regarded as a plurality of logical drives.
[0058]
Conversely, each drive may be each partition provided in a single disk device. In this way, since a plurality of drives are unnecessary, an excellent file system can be realized at a low cost with a simple hardware configuration.
[0059]
2. Operation and effect of the first embodiment
The operation of the distributed file system of the present embodiment having the above configuration will be described below with reference to the flowcharts of FIGS. In this case, FIG. 3 is a flowchart showing a processing procedure when the buffer management means receives a block write request (update request), and FIG. 4 shows a processing procedure when the flash serialization means receives a flush start request. It is a flowchart to show.
[0060]
Here, as an example, an example will be described in which an operation request for updating a block as shown in FIG. 5 is given to the file storage device 1S shown in FIG. For convenience of explanation, in the initial state, all the disks 1E to nE of all the file storage devices 1S to nS are consistent in data, and all the flash state flags of all the file storage devices 1S to nS are used. The flash state flags 1B to nB are assumed to be “normal”. Further, it is assumed that the entire disk storage 1D of the file storage device 1S is “free” as shown in FIG.
[0061]
2-1. Processing when all other file storage devices are "normal"
First, the buffer management means 1C responds to the request of the operation 1 that requires the update of the blocks 1, 2, and 3 shown in FIG. 5 with the number of write blocks in the disk buffer 1D as shown in FIG. It is determined whether or not there is an “empty” area for three blocks or an area having no update necessity (step 301). In this case, as shown in FIG. 6, since there are three blocks of “free” areas in the disk buffer 1D, each of the three blocks of “free” areas has the blocks of write blocks 1, 2, and 3. The address and data contents are stored, and the necessity for updating is set to “present” (step 306). As a result, the contents of the disk buffer 1D are as shown in FIG. In the middle of this processing, the disk 1E is not updated, so the data consistency in the disk 1E is still maintained. The data content of each block in the disk buffer corresponds to the buffer block referred to in the claims.
[0062]
Next, the buffer management means 1C responds to the request of the operation 2 that requires updating of the blocks 4, 5, and 6 shown in FIG. 5, and the number of write blocks in the disk buffer 1D as shown in FIG. It is determined whether or not there is an “empty” area for 3 blocks or an area having no update necessity (step 301). In this case, as shown in FIG. 7, since there are three blocks of “free” areas in the disk buffer 1D, each of the three blocks of “free” areas has blocks of write blocks 4, 5, and 6. The address and data contents are stored, and the necessity for updating is set to “present” (step 306). As a result, the contents of the disk buffer 1D are as shown in FIG. In the middle of this processing, the disk 1E is not updated, so the data consistency in the disk 1E is still maintained.
[0063]
Subsequently, the buffer management means 1C responds to the request of the operation 3 that requires the update of the blocks 7, 8, and 9 shown in FIG. 5, and the number of write blocks in the disk buffer 1D as shown in FIG. It is determined whether or not there is an “empty” area for 3 blocks or an area having no update necessity (step 301). In this case, as shown in FIG. 8, since there is no “empty” area for 3 blocks or “no” update necessity in the disk buffer 1D, the buffer management means 1C uses the flash serialization means 1A for flashing. A start request is sent (step 302), and the flash start permission from the flash serialization means 1A is awaited (step 303).
[0064]
Further, in response to the flush start request from the buffer management unit 1C, the flash serialization unit 1A sets the flash status flag 1B to “investigation” (step 401) as shown in FIG. The flash status flags 2B,..., NB of 2S,..., NS are examined (step 402), and it is determined whether there is a file storage device that is “under investigation” or “updated (execution)” (step 403). ). At this time, if any of the other file storage devices 2S,..., NS is not “investigating” or “updating (executing)”, the flash serialization means 1A sets the flash status flag 1B to “updated (executing) Middle ”), a flash start permission is returned to the requesting buffer management means 1C (step 405), and a flush end notification from the buffer management means 1C is awaited (step 406).
[0065]
In the process so far, the disk 1E of the file storage device 1S has not been updated at all from the initial state, so the consistency of data on the disk 1E is still maintained.
[0066]
Further, in response to the flash start permission from the flash serializing unit 1A, the buffer management unit 1C performs the flush process of the disk buffer 1D with all the data in all the areas in the disk buffer 1D that have the update necessity “present”. (Step 304). That is, here, the disk 1E is updated with the block addresses and data contents of the blocks 1, 2, 3, 4, 5, and 6 stored in the area where there is an update necessity “present”, and these 6 blocks are updated. Change the area to "None" for disk update necessity.
[0067]
In the midst of this flash processing, when only a part of the blocks 1, 2, 3 or only a part of the blocks 4, 5, 6 of the disk 1E is updated, a plurality of blocks that must be updated by one operation. Since the data on the disk 1E is not consistent at this point in time because all of the data are not updated, the target blocks 1, 2, 3, 4, 5, and 5 of the disk 1E are represented by all the data of these 6 blocks. When the update of 6 is completed, the consistency of the data on the disk 1E is taken again. After this flush processing, the buffer management means 1C sends a flush end notification to the flash serialization means 1A (step 305). When the flash serializer 1A receives the flush end notification from the buffer manager 1C, it returns the flash status flag 1B to “normal” as shown in FIG. 4 (step 407).
[0068]
The buffer management means 1C continues to store the block addresses and data contents of the write blocks 7, 8, and 9 in each of the three blocks of the “vacant” area of the disk buffer 1D or the “necessity of updating” area. The update necessity is set to “present” (step 306). As a result, the contents of the disk buffer 1D are as shown in FIG. However, since the disk 1E is not updated in this case, the data consistency on the disk 1E is still maintained.
[0069]
As described above, in the file storage device 1S of the present embodiment, the consistency of data on the disk 1E is lost when the disk buffer 1D is being flushed, that is, the flash status flag 1B is “updated ( It is only when “Running”).
[0070]
2-2. Processing when any of the other file storage devices is "Investigating" or "Updating (executing)"
On the other hand, for example, while the file storage device 2S is performing the flash process, the flash status flag 2B of the file storage device 2S is “updating (running)”. In this state, when the flash serializing unit 1A of the file storage device 1S receives a flush start request, the flash serializing unit 1A sets the flash status flag 1B to “under investigation” as shown in FIG. (Step 401), the flash status flags 2B,..., NB of the other file storage devices 2S,..., NS are examined (step 402), and there is a file storage device that is “under investigation” or “updated (execution)”. Then, it is determined (step 403). Subsequently, the flash serializing unit 1A sets the flash state flag 1B to “normal” and waits for a predetermined time (step 404).
[0071]
Then, after a certain time, the flash serializing means 1A again sets the flash status flag 1B to “investigation” (step 401), as shown in FIG. 4, and flushes the other file storage devices 2S,. The status flags 2B,..., NB are examined (step 402), and it is determined whether there is a file storage device that is “under investigation” or “updated (execution)” (step 403). In this case, if the file storage device 2S is “updating (running)”, the flash serialization unit 1A sets the flash state flag 1B to “normal” and again waits for a fixed time.
[0072]
On the other hand, while the flash serialization means 1A is on standby, the flash processing ends in the file storage device 2A, the flash status flag 2B returns to “normal”, and “in search” or “update in progress (execution) If it is determined that there is no file storage device “medium)”, the flash serialization unit 1A sets the flash status flag 1B to “updating (during execution)” and permits the requesting buffer management unit 1C to start flashing. Is returned (step 405), and a flush end notification from the buffer management means 1C is awaited (step 406). Finally, the flash serialization means 1A returns the flash status flag 1B to “normal” when it receives the flash end notification from the buffer management means 1C (step 407).
[0073]
Thus, in the first embodiment, when the flash processing is to be performed by one file storage device 1S, the operation of the flash serialization means 1A causes other file storage devices to perform “inspection” for the flash processing. In the case of “medium” or “updating (executing)” of the flash processing, execution of the flash processing by the buffer management means 1C is stopped for a certain period of time, and “inspection” or “updating (executing)” For the first time when the file storage device runs out, the buffer management means 1C is returned to the flash start permission, and the disk buffer 1D is flushed. Therefore, two or more file storage devices do not perform flash processing at the same time.
[0074]
As described above, in the first embodiment, the consistency of the data on the disk 1E is lost when the disk buffer 1D is being flushed, that is, the flash status flag 1B is “updated (during execution). ) ”, And since two or more file storage devices do not perform flash processing at the same time, two or more file storage devices simultaneously lose data consistency on the disk. There is no end. Therefore, even if a plurality or all of the file storage devices are down due to a failure at the same time, the data of the files multiplexed and stored in the plurality of file storage devices is at least one file. It is surely stored in the storage device.
[0075]
In particular, in the first embodiment, the serialization means detects another drive that is being updated for each update, thereby avoiding the time overlap of the updates, so that reliable serialization can be performed. Also, in the first embodiment, a highly reliable file system can be realized by giving the control of update start to the serialization means, and the hardware and procedure for performing the update itself can be used, so conventional equipment can be used. Can be used effectively. That is, when applying the present invention to a conventional file system, if only the basic control procedure for accessing the disk, such as hardware and BIOS (Basic Input / Output System) is changed, the logical control procedure of the file system can be changed. There is no need to change (FIG. 10).
[0076]
3. Second embodiment
In the second embodiment, the buffer means is configured to record the operation and each buffer block targeted by each operation using the identification information, and the update means is configured to perform a single operation based on the identification information. The updating is performed in units of all buffer blocks corresponding to the above operation. In the second embodiment, when the same buffer block is included in a plurality of operations, the updating unit is configured to continuously update all the buffer blocks related to the plurality of operations. . For example, every time the operation content is recorded in the buffer, the operation number and the target buffer block number are recorded in the comparison table (FIG. 11). Also, an operation number for each buffer block is recorded in a predetermined area of the disk buffer. The operation number for each buffer block is recorded (FIG. 12). For example, in FIGS. 11 and 12, blocks 1 and 2 are operated by operation 1, blocks 3 and 4 are operated by operation 2, and blocks 3 and 5 are operated by operation 3. In this case, the update is performed integrally with the buffer blocks 1 and 2, and the buffer blocks 3, 4 and 5 related to the operations 2 and 3 are performed together. This is because the state in which only the buffer blocks 3 and 4 related to the operation 2 are updated and the buffer block 5 is not updated is “write” in view of the operation 3.
[0077]
In the second embodiment, since the correspondence relationship can be easily specified by recording identification information such as a code number for each operation and the buffer block, a simple data structure is not used. The operation unit can be updated with the configuration.
[0078]
Further, in the second embodiment, when a plurality of operations target the same block, the block is not updated only by the contents of some of the operations. Will not lose its integrity.
[0079]
4). Other examples
The present invention is not limited to the above-described embodiment. For example, the specific configuration and the flow of processing of the buffer flush unit and the flash serialization unit can be freely changed. Further, the components of each unit other than these means and the configuration of the entire file storage device can be freely changed in the same manner, and the number of file storage devices can be freely selected.
[0080]
For example, performing buffer flush is not limited to the case where there is not enough room to write new data. When a buffer flush instruction is given by a user or a program, an instruction for removing the medium from the drive is given. When you want to stop the system, you can decide freely.
[0081]
Further, for example, each of the serializing means may be configured to perform the investigation by collecting information on the state of another drive via a communication line (Claim 6). In such an embodiment, as long as there is a communication line, the serialization means uses this communication line to perform an investigation, so that no flag or supervisor is required.
[0082]
Further, for example, the serialization means is configured to transfer and copy the flag of each drive to all other drives as a copy, and to perform the investigation by referring to this flag in the other drive. (Claim 7). In such an embodiment, the serialization means of each drive does not need to refer to the state of another drive or a flag provided in another drive every time an investigation is performed, so that the processing at the time of updating is fast. It becomes. The communication and flag transfer as described above can also be performed using a LAN such as Ethernet and MAP.
[0083]
Further, for example, the serialization unit may be configured to perform the serialization by limiting the update of each drive within a predetermined clock cycle based on a predetermined clock. In such an embodiment, an update opportunity in each drive is surely given in a predetermined cycle, and there is no case where an update in a specific drive is delayed by a series of updates in other drives.
[0084]
【The invention's effect】
As described above, according to the present invention, since the consistency between blocks in a file is not lost even if a drive fails at a moment other than during update, a highly reliable file system can be provided.
[Brief description of the drawings]
FIG. 1 is a block diagram showing components of a first embodiment of a file system of the present invention.
FIG. 2 is a conceptual diagram showing an example of a specific configuration of the system of FIG.
FIG. 3 is a flowchart showing a processing procedure when the buffer management unit of FIG. 1 receives a block write request (update request);
FIG. 4 is a flowchart showing a processing procedure when the flash serialization unit of FIG. 1 receives a flash start request;
5 is a table showing operation numbers and update target block addresses of operation requests for the file storage device 1S of FIG.
6 is a table showing an initial state of the disk buffer 1D of FIG.
7 is a table showing a state after the operation 1 of the disk buffer 1D of FIG.
8 is a table showing a state after the operation 2 of the disk buffer 1D in FIG.
FIG. 9 is a table showing a state after the operation 3 of the disk buffer 1D in FIG.
FIG. 10 is a conceptual diagram showing a software configuration of a distributed file system
FIG. 11 is an operation history table according to the second embodiment.
FIG. 12 shows the contents of a disk buffer in the second embodiment.
[Explanation of symbols]
1S, 2S ... File storage device
1A, 2A ... Flash serialization means
1B, 2B ... Flash status flag
1C, 2C... Buffer management means
1D, 2D ... disk buffer
1E, 2E ... disc
10 ... System bus

Claims (20)

A plurality of nonvolatile storage drives;
Each drive has a
A plurality of blocks as access units on the recording medium of each drive;
Buffer means for temporarily storing a buffer block which is data to be written to each block in a buffer on a volatile memory;
Determining means for determining the necessity of updating the recording medium by the buffer block;
Update means for continuously updating the blocks of the drive with all buffer blocks when it is determined that the update is necessary;
A file system comprising serialization means for serializing updates while avoiding temporal overlap of the updates in each drive. The serialization means investigates whether there is another drive being updated when attempting to perform the update in one drive, and if there is another drive being updated, in the one drive 2. The file system according to claim 1, wherein updating is inhibited, and updating in the one drive is performed when there is no other drive being updated. The serialization unit is configured to perform an investigation again after waiting for a predetermined time when a drive being updated exists, and to update when a drive being updated does not exist. 2. The file system according to 2. The serializing means sets a flag indicating the state of each drive to “updating”, “under investigation”, and other states according to the state of the drive, and sets other drive flags when attempting to update. The file system according to claim 2, wherein the file system is configured to investigate. 2. The file system according to claim 1, wherein the serializing means is provided individually for each drive. 5. The file system according to claim 4, wherein each of the serializing units is configured to perform the investigation by collecting information related to a state of another drive via a communication line. The serialization means is configured to transfer / copy the flag of each drive to all other drives as a copy, and the other drive refers to the copy flag to perform the investigation. The file system according to claim 4 . 2. The file according to claim 1, wherein the serialization unit is configured to perform the serialization by limiting the update of each drive within a predetermined time cycle based on a predetermined clock. system. 2. The file system according to claim 1, comprising a plurality of drives, wherein each file is recorded on the plurality of drives. A plurality of nonvolatile storage drives;
Each drive is
A plurality of blocks as access units on the recording medium of each drive;
Each time a single data write operation occurs , a buffer block, which is data to be written to each block in a buffer on volatile memory, is requested for each block to be updated to the drive. Buffer means for temporarily storing a buffer block, which is data to be written in each block, in a buffer on a volatile memory;
Determining means for determining the necessity of updating the recording medium by the buffer block;
Updating means for continuously updating blocks of the drive by at least all buffer blocks to be subjected to the single data write operation when it is determined by the determining means that updating is necessary;
A file system comprising serialization means for serializing updates while avoiding temporal overlap of the updates in each drive. The buffer means temporarily stores the buffer block in the buffer together with identification information for associating the generated single data write operation with all the buffer blocks that are targets of the write operation. Yes,
The update unit is configured to perform the update in units of all buffer blocks corresponding to a single data write operation based on the identification information when the determination unit determines that the update is necessary. The file system according to claim 10, wherein: It said updating means, if it contains the same buffer block subject to a plurality of data write operations, characterized in that it is configured to continuously update all the buffer block according to the plurality of data write operations The file system according to claim 11. The serialization means investigates whether there is another drive being updated when attempting to perform the update in one drive, and if there is another drive being updated, in the one drive 11. The file system according to claim 10 , wherein the file system is configured to suppress the update and perform update in the one drive when there is no other drive being updated. The serialization unit is configured to perform an investigation again after waiting for a predetermined time when there is a drive being updated, and perform an update when there is no longer a drive being updated. 13. The file system according to 13. The serializing means sets a flag indicating the status of each drive to “updating”, “investigating” and other states according to the status of the drive, and sets the flags of other drives when attempting to update. The file system of claim 13, wherein the file system is configured to investigate. 11. The file system according to claim 10 , wherein the serializing means is provided individually for each drive. 16. The file system according to claim 15, wherein each of the serializing units is configured to perform the investigation by collecting information on the state of another drive via a communication line. The serialization means is configured to transfer / copy the flag of each drive to all other drives as a copy, and the other drive refers to the copy flag to perform the investigation. The file system according to claim 15 . 11. The file according to claim 10 , wherein the serialization means is configured to perform the serialization by limiting the update of each drive within a predetermined time cycle based on a predetermined clock. system. 11. The file system according to claim 10 , comprising a plurality of drives, wherein each file is recorded on the plurality of drives.

Images