JPH05508506A - Fault-tolerant network file system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Fault-tolerant network file system Background of the invention This invention provides a main file server and a backup file server that reflects this file server. It concerns a fault-tolerant network file system with a server. The master is broken When a backup is stored on a network in a way that is obvious to the user where the file is primarily stored, and assume the role of the Lord.

A circuit network typically has some connection points that communicate with each other by high-speed communication links. ing. Examples of connection points are single user personal computers and workstations. applications, dual-user computers, and image scanners, printers and displays Contains device peripherals.

Many networks have data and software used by multiple connection points in the network. has a file server attachment point that acts as a central storage facility for . A file server is a disk storage device for storing data and software. including a central processing unit (CPU) for controlling the server as well as the file server. There is. Files that reach a file server via a communication link typically first Stored in cache memory within the central processing unit and then transferred to disk for permanent storage. Copied to storage device.

Files in cache memory are copied to disk by the file server. If it does not work (becomes), the file will be irretrievably lost. Therefore , a requester connection point that sends some important files to a file server is commands the file server to immediately write the file to disk, thereby The specified file can reduce the hardness lost in case of such failure. Ru.

Many connection points rely on file servers, so if a file server fails, If so, many users will be affected. For example, even if all Even if the files are stored securely on the primary file server's disk, the primary file A temporary failure of a file server may result in heavy usage until the file server restores service. This will cause inconvenience to many users as their data files are unavailable. Become. Circuits M that require a high degree of availability are typically Use protective techniques. For example, to maintain a copy of the main file. A backup file server can be used for this purpose. In the event of a main failure, the main file An exact copy of the file can be accessed automatically by a backup file server. Wear.

One purpose of this invention is to create a backup file that quickly reflects each change in the main file. It is to provide a server. Therefore, when the main failure occurs, the backup file The server's files are the same as even the most recently added or modified files on the host. I will. A further object of this invention is to perform this reflection at high speed (with little computational cost). It is about achieving. Another purpose is to use the main file server in a way that is obvious to the user. The objective is to provide various elements for switching from a file server to a backup file server.

Summary of the invention This invention relates to an improved fault tolerant network file server system. this A circuit file server system has circuit network communication links connected to multiple connection points. It is growing. The primary file server and backup file server also receive files from the attachment point. Connected to a network communication link for storage.

In this improved file server system, the main file server is the main computer. a first network interface connected to the processing unit, the main storage disk, and the network communication link; a first independent interface connected to a backup file server; There is. The first independent interface transmits information to the backup file in response to commands from the main processor. can be transmitted to the file server.

A backup file server is a backup computer processing unit, backup storage disk, and network communication link. the back-up network interface connected to the link and the first independent interface. and a second independent interface connected to the second interface. The second independent interface is double Contains ported storage. This interface receives information from the main file server. This information is then stored in a dual-ported storage device. from backup processing equipment In response to the command, the second independent interface provides information from the dual ported storage device. supply.

In the adopted embodiment, the PCB2 independent interface is allocated to a back-up computer processing unit. This is where the dual-ported storage device stores the information it receives from the main file server. Contains a device for reporting that the One of the dual port storage devices Each portion is on a disk to store data to be replicated to backup storage disks. Stores a removal pointer that identifies the block.

The other part of the dual-ported storage device is connected from the primary file server to the backup file server. a first control message block for storing control messages to the A second system for storing control messages from the file server to the primary file server. It can also be configured as a message block.

The second independent interface is configured to remove data blocks from the data block portion. at least one device accessible to the primary and backup image processing devices for directing Contains two common registers. For example, the adopted embodiment includes a counting register. There is. The second interface is responsive to signals from the first independent interface. Change the contents of the number register to instruct data to be registered in the memory block. Contains equipment for The second independent interface is also a backup computer processing unit. Modifying the counting register in response to signals from and removing data from the storage block can be made to instruct.

Dual-ported storage devices use semaphores to arbitrate access to counting registers. Contains jista. This semaphore register has first and second operating states. There is. When read by the processing unit while in the first state, the semaphore register automatically enters the second state by supplying the register available code. In the second state When read by a processing unit, the semaphore register is marked as register unavailable. remains in the second state. Written by the processor when in the second state When the semaphore register is input, the semaphore register returns to the first state.

In the adopted embodiment, the primary file server further connects the first network interface to the primary file server. Reproduction of files received from the main storage disk and dual-port storage device. Improved Unix (lJnix) operating system for control Contains. Normal Yuex operating system is a reliable disk Writes the specified file in response to both a write command and an uncertain disk write command. Instructs the operating system to write to main storage disk. uncertain In response to a write command, the unreliable write procedure replicates the file to disk. writes to the main storage disk using an efficiency algorithm that temporarily postpones cormorant.

In response to a reliable write command, a reliable procedure can be performed directly without utilizing an efficiency algorithm. Write the file to disk immediately. This improved operating system is Directs specified files to dual-ported storage in response to both actual and uncertain commands. write the specified file to the primary disk using an unreliable write procedure. Write to storage device.

Other objects, features and advantages of the invention will be described in the following description of the selected embodiments, which are taken together with the drawings. It is clear from the explanation.

Brief description of the drawing Figure 1 is a configuration diagram of a computer circuit network with a main file server and a backup file server. It is.

Figure 2 shows a pair of interfaces for connecting the primary file server to the backup file server. FIG. 3 is a configuration diagram of a face plate.

FIG. 3 is a diagram of several registers on the interface board shown in FIG.

FIG. 4 is a diagram showing the organization of a dual-port storage device according to an adopted embodiment of the present invention. be.

Figures 5(a) and 5(b) show new files on the main file server in dual-port format. 2 is a flowchart of a method for reproducing to a storage device.

Figures 6(a) and 6(b) show data from a dual-ported storage device being transferred to a backup file server. 3 is a flowchart of a method for reproducing to a server;

In Figures 7 (a) to 7 (d), the backup file server temporarily becomes the main file server. 2 is a flowchart of the procedure to replace

Figures 8(a) and 8(b) show that the failed main file server is connected to the network after the failure is recovered. 2 is a flow diagram of the procedure for resuming the role.

Explanation of adopted examples Referring to FIG. 1, network 10 includes a plurality of connection points 12, each of which Perform specific tasks in the planning and production of publications such as newspapers and magazines can. For example, connection point 12(a) is used by journalists to include notes that should be included in a publication. connection point 12, which can be a computer workstation that makes it possible to draft things; (b) digitizes an image, for example a photograph, and prepares it at connection point 12(a); connection point 12 ( c) selects and distributes articles, images and advertisements prepared by other connection points of the network; A calculator workbook that allows users to allocate entire publications by It could be a work station.

The network can include other connection points for producing hard copies of publications. . For example, a printer connection point is a hard copy of an image called a ``proof''. Prepare copies. Other connection points are printing master plates used in mass printing of publications. Prepare.

Each connection point 12 communicates with other connection points by high speed communication links 14. for example , in a circuit network that follows the Ethernet protocol, the link 14 is a high-speed connection point that can spread messages to one or more other connection points. It is a serial communication channel. The connection point of poly() is the information used locally at the connection point. Contains a disk for storage of information. For example, workstation 12 (a ) and 12(c) typically perform these respective tasks, e.g. A disk for storing software used to perform and document allocation. Contains.

However, many attachment points are diskless and therefore do not store information and software. requires access to the central file server 15 to call the software.

Furthermore, even attachment points that have disks often have them stored on a central file server. memorize the data. Connection point to store data files on a central file server requires a common link that is accessible to many connection points in the network. It offers several advantages, including centralized control of the file system.

Failure of a file server attachment point occurs when a file server is required by a large number of attachment points. This is a disaster because it contains information that Therefore, the backup file server 16 The main file goes through a parallel port to maintain a copy of everything in the file. connected to the file server. In the event of a main failure, a backup is available to connect other connections in the network. Automatically assumes the role of the Lord in a way that is obvious to the point. More specifically, the other connection points are Continue to access the central file system without any special instructions from use be able to.

The primary file server receives messages (e.g. files) from communication link 14. a first network interface for storing information in the buffer storage device 21; I'm reading. The central processing unit CPU24 is the operating system software 2 5 through the first interface 28 by the parallel bus 22. , across the cable 19 to the second interface 3° in the backup file server 16. Transfer buffer contents. The second interface 3o temporarily stores buffer data. It includes a dual ported storage device 1f31 for storing.

After storing the data in the storage device 31, the CPU 24 stores the data in the backup file server 16. Instructs interface 3o to interrupt CPU 32, thereby causing storage 31 contains data to the CPU 32. In response to the interrupt, CP U32 stores the buffer 7 data contents of the dual port storage device 31 in a cache storage device. Move to 33. The operating system 35 later transfers the cache to the CPU 32. The data from the disk 33 is copied to the backup disk 34. Disk 34 is provided with a copy of each file arriving from the network.

After CPU 24 copies the buffer data to the dual-ported storage device, the operator processing system 35 moves this data to data cache storage 27. Command the CPU 24. The operating system 25 then caches the CPU 24. The contents of the flash storage device 27 are instructed to be copied to the main disk 26. Explained below As will be seen, the operating system retrieves files from cache 27. Three types of writing procedures are provided for reproduction onto the disc 26.

In the selected embodiment, the operating system 25 is installed on the backup file server 1. 6 to transfer files from buffer storage 21 to dual-ported storage 31 Yuex Operator modified to include code to allow replication to It is a ting system. The normal Yuex operating system is Contains file system code for maintaining a system of data files on the disk. I'm here.

This file system code copies files from cache 27 to disk. Contains several types of writing procedures for writing. locally on the main file server for generated files (i.e. files that do not arrive via communication link 14). For this reason, this file system code typically uses a delayed write procedure.

This delayed write procedure avoids unnecessary delays in writing data to disk. implement efficiency algorithms to manage For example, the delayed write procedure Search the cache for files assigned to the disk "cylinder".

It then directly stores the selected files for continuous transfer to disk. Commands the CPU to initialize the device controller (DMA). This initial setting After completion, the CPU returns to other tasks and the DMA0m transfers the selected data. Be careful about duplicating files to disk. Files are the same disk cylinder , so the time-consuming “cylindrical search” operation is performed by linking these files. This can be avoided by writing continuously.

For files arriving via link 14, a "synchronous" write procedure is generally used. used. This synchronous write procedure does not use the efficiency algorithm described above. Write files to disk quickly. Quickly retrieve files from volatile cache By removing the synchronous write procedure, the file Reduces the hardness at which file contents are lost. However, this procedure does not require efficient algorithms. by preventing the system from reducing the number of time-consuming cylinder search operations. and reduce disk performance.

Furthermore, the requester server can also operate in a synchronous manner, The file must be written to disk before you start running the application program that facilitated the writing. Wait for the file server to confirm that it has been loaded. Such “remote synchronization” ” writes remain unresponsive at the requester connection point until the data is safely stored on disk. Because the program cannot proceed, there is a high risk of failure in the file server. Provides a degree of safety. However, this procedure does not allow application programs to stand for long periods of time. Getting stuck waiting for the relatively slow disk write procedure to complete and receiving an acknowledgment message. The performance of the client connection point is reduced because it waits for messages to arrive over a relatively slow network. decrease.

Among the above types of disk writes, the relatively uncertain delayed disk write procedure is It is the least burden on system performance. However, this procedure risk loss of data. Volatile cache in case of file server failure Memory contents are destroyed. Therefore, delayed writes are generally performed on relatively unimportant data. synchronous and asynchronous writes are used for relatively important data. used.

Backup file server 16 can be used without the need for expensive synchronous disk writing equipment Provides a high level of safety against More specifically, each file is replicated to a dual-ported storage device 31 powered and controlled by the file server. Ru. Backup file servers are therefore independent of the main, so volatile cabling There is no need to immediately duplicate memory 27 to primary non-volatile disk 26. Therefore, The operating system 25 changes all synchronous disk writes to delayed writes. This is a modified version of the Yuex operating system. Do not write synchronously. This ensures that the efficiency algorithm optimizes disk performance by By coordinating all disk writes, file server performance is dramatically improved.

Furthermore, application programs that require many remote synchronous writes (e.g., The client connection point for processing (document processing applications) is a dual port type with relatively high file speed. As soon as it is written to storage, it receives an acknowledgment from the file server. That's it Moreover, the requester attachment point no longer has to wait for a slow disk write procedure; Its performance is dramatically improved.

Dual-ported storage operation Referring to FIG. 2, dual ported storage device 31 is configured to support both CPU 24 and CPU 32. It includes a 19-bit address bus ADDR which is accessible to both sides. memory device To access a given location in the device 31, the CPU 24 first The address of the location to be stored is loaded into a pair of address registers 40.42.

As shown in FIG. 3, register 42 contains the lower 16 bits of the address (i.e. , bits AO to A15), and register 40 contains the upper three bits of the address. It accommodates two bits (bits A16 to A18). CPU 24 is storage device 3 When performing a read or write cycle directed to 1 (a circular process), the register The contents of data 40.42 are automatically added to the address bus ADDR, thereby Points to a specific location within the storage device 31.

To load address register 40, CPU 24 assigns a A write cycle is performed to a predetermined address. Address within interface 28 Bus decoder 44 decodes the address from parallel bus 22. Predetermined register address When the decoder 44 recognizes the response, the decoder 44 requests that the CPU 24 access the register 40. A pair of encoding control signals rREGJ are asserted to indicate that the Interfuture A second address decoder 46 within the base 30 decodes this control signal and stores it in the register. Assert register enable signal R1 to select 40. writing is in progress A read/write control signal from bus 22 indicating that is sent to the interface 30. The decoder 46 receives the sent R/W signal and , provides a corresponding buffered signal “R/W1” to register 40. buff The supported read/write signal R/Wl indicates that a write is occurring. Therefore, the activation of the register enable signal R1 causes the register 40 to be connected to the data bus DB. These data are loaded into the register 42.

Data on bus DB is provided by CPU 24 through transceiver 50.52. be provided. More specifically, address decoder 44 reads/writes from bus 22. Monitor control signals. Authorizes write cycles to locations on interface 30. When decoder 44 recognizes that transceiver 50 transfers data from bus 22 to cable 1 9 allows for delivery. Similarly, decoder 46 30, the transceiver 52 transmits the data from the cable 19. can be sent to the data bus DB.

In the same manner, CPU 24 writes to the address assigned to register 42. Load register 42 by doing: In response to this integration cycle, The decoder 46 loads the data from the bus DB into a second register 42. Assert enable signal R2.

Once registers 40 and 42 are initialized with appropriate addresses, CPU 24 Place! By performing a write cycle to a predetermined address assigned to and writes the data to the addressed location in storage device 31. In response to this , decoder 46 asserts the storage address on storage address bus ADDR. Assert memory register signal R3, which instructs registers 40.42 to Ru.

The memory register signal R3 is further supplied to a multiplexer 54, which In response, Kusa sends a memory access signal rMEM-3electJ to the storage device 31. Add. The buffered read/write control signal R/Wl is also sent to multiplexer 5. 4 and this multiplexer responsively outputs the memory read/write control signal r Assert MEM-R/WJ. MEM-R/W indicates that the current cycle is a write memory selection signal rMEM-3electJ. transfers data DO to D16 (see Figure 3) from the data bus DB to the address bus AD. Commands storage device 31 to load to a given location on DR. therefore , the predetermined address assigned to the storage device 31 is stored in the address register 40.4. 16-bit training whose contents are determined by the contents of the memory location indicated by 2. It functions as a register 43.

Referring to FIG. 3, register 40 is a 16-bit register. however, Only five bits in this register are used. as explained before The lower three bits store the higher address bits (A16 to A18). There is. Bit 9 is the write increment bit rW11j. this bit is set Then, the address in registers 40.42 is Incremented each time you write to a location. Therefore, the CPU 24 Load the block of memory locations at 1 into register 40.42 and write increment You can write to this block by setting the bit.

CPU 24 then simply iteratively copies the data to the address provided to storage device 31. Write. Register 4.0.42 increments the storage address with each write; to contiguous locations in storage. Similarly, the CPU 24 The read increment bit R11 is set in the register 40 and the read increment bit R11 is read from the storage device 31. Blocks of storage can be read by successive reads.

CPU 32 reads and writes data from storage device 31 in a similar manner. example For example, to read a block of data from storage device 31, CPU 32 Load the address of the first location to be stored into a pair of address registers 56 and 58. and sets the read increment bit RI2 in register 56. it will be remembered next Data is read from a predetermined address given to the device 31. address decoder 60 decodes the address from the bus 62 and recognizes the address of the storage device 31, Address register 5 asserts the storage address on address bus ADDR. 6. Assert memory access signal R7 commanding 58. Access signal R7 is also added to multiplexer 54, which multiplexer outputs the memory access signal. It responds by asserting rMEM-3e l ec tJ. Accessories In response to the bus signal MEM-9e]ect, the storage device 31 The data DO~D15 from the storage location identified by the address of Supply to B. Therefore, from the perspective of the CPU 32, the The given address entered is determined by the contents of the memory location pointed to by the address register. It functions as a 16-bit register 43 whose contents are determined.

Read size from CPU 32 directed to storage location at interface 30 When the decoder 60 recognizes the bus 62, the decoder 60 asserts the contents of the data bus DB on the bus 62. asserts an "enable" signal that causes data transceiver 64 to This supplies the desired data to the CPU. read increment bit set Then register 56. '58 automatically increments storage addresses. Therefore, C The PU 32 decides to read again from the address assigned to the storage device 31. Therefore, the next location in the storage device 31 is read.

Referring to Figures 4.5(a), 5(b) and 6, the main file server 15 The software procedure for transferring data from to backup file server 16 is more detailed. It is described in

Configuring dual-ported storage devices FIG. 4 illustrates the storage map for the dual-ported storage device 31 of the selected embodiment. Ru. The storage device 31 includes controls for controlling access to registers, which will be explained below. The semaphore block at the beginning of the storage device containing the semaphore used when There are 70 (ie, first 512 locations). After this semaphore block 24 and 24 in sending data and control messages as described below. Control register block 7 containing various registers used by CPU 32 There are 2.

Following the control register block 72 are a pair of control message blocks 74 and 76. be. The first control message block 74 is for sending control messages to the CPU 32. It is used by CPU 24 for this purpose. Each message is defined by three words 75 be done. To send the message, the CPU 24 sets the corresponding three word codes. the first container 78 in the page block 74. Subsequent messages are adjacent written to the container.

The CPU 32 retrieves messages on a first-in, first-out basis. CPLT24 blocked By the time the last container 80 in the network is loaded, the CPU 32 has already loaded the first Empty the container. Therefore, after loading into the last container, CPU2 4 returns to the first container. The control message block 74 is thus Operates as a fa.

The control message block 76 transmits control messages from the CPU 32 to the CPU 24. The same method is used for transfer. The CPU 32 stores messages in a circular buffer. and the CPU 24 retrieves it on a first-in, first-out basis.

Next to the control message blocks 74, 76 are some data storage blocks 82. Ru. This data storage block is connected to the control message blocks 74 and 76. between CPU 24 and CPU 32 in the same manner used to transmit the used to transmit data. More specifically, the CPU 24 the data of the block is loaded into the first storage block 84. new block data Each time it arrives, the CPU 24 loads this data into the next storage block. .

When CPU 24 loads a data block into storage, CPU 32 loads it first. Take out on a first-in, first-out basis. CPU 24 loads last memory block 86 By this time, CPU 32 has already emptied the first storage block 84. that Therefore, when the last storage block 86 is loaded, the CPLT 24 loads the first storage block 86. Return to 84. Storage block 72 therefore operates as a circular buffer.

Operation of circular buffers in dual-ported storage devices, see 15(a) and 5(b) Now let's take a look at how circular buffers work by using data block transfers as an example. will be explained in detail.

Control register block 72 registers the next available data block in storage 31. contains a next entry pointer 88 pointing to the next entry pointer 88; It is also the next to be emptied. a removal pointer 90 pointing to the data block currently stored in the storage device 31; It includes a count register 92 that specifies the number of data blocks present.

When CPU 24 desires to load a block of data into storage device 31, First, read count register 92 to determine if the circular buffer is full. (Circulation (Step) 210). The CPU 32 generally blocks quickly enough Because of the removal, the circular data buffer can never be full. however , when the buffer is full, the CPU 24 removes the block from the CPU 32. Count register 92 is read repeatedly until the count is decremented (step 212).

If the buffer is not full, CPtJ24 reads entry pointer 88. to determine the location of the next available data block (step 214). CPU2 4 is the address of the first location in the next selected block and is stored in register 40°4. 2 and sets the write increment bit WT in register 40 (step 21 6). The CPU 24 then writes data to the storage device 31 by successively writing to the storage device 31. the data into the selected storage block (step 218). A block of data is loaded When the CPTJ 24 updates the entry pointer 88 to the next available block. Point to the lock (step 220).

Finally, CPU 24 increments the counter to indicate that a new block has been added. must be divided. However, CPU 32 also performs data block removal. At the same time change the count to reflect, i.e. try to decrement the count be able to. For example, CPU24 and CPU32 read the current value of a count of 5. Assume that you will take it. CPU 24 that has just loaded a new data block attempts to increment the count by six. The CPU 32 that has just removed the block counts attempts to decrement by 4. The mechanism for synchronizing the CPU 24 with the CPU 32 is standard. If not, CPU 24 can write 6 to count register 92. Then, the CPU 32 tries to write 4 on top of this value. However, five blocks Since the lock remains in storage 31, the count should remain at 5. Ru.

To avoid this type of error, the counting semaphore word 96 is (Figure 4). Before reading count register 92, CPU 24 first reads Read counting semaphore 96 (step 222). If this semaphore is 0 ( Step 224. ), the CPU 24 has control over the counting registers. Assume that

When the CPU 24 reads O from the semaphore 96, the storage device 31 sets the semaphore to 1. automatically, thereby placing the count register 92 under exclusive control of the CPU 24. It indicates to the CPU 32 that the The CPU 24 reads the count and increments it. (steps 226, 228), writes 0 to semaphore 92, thereby Free count register 92 for use by U 24 (step 230).

The count read at step 226 is 0, which causes the CPU 24 to If it shows that the circular buffer was empty before appending the last block For example, CPU 24 interrupts CPU 32 and tells it that the buffer currently contains the data. (steps 232 and 234) (when CPU 24 assigns CPU 32 The loading mechanism is explained in detail below. ). If the count is greater than 0, it is still removed. An interrupt is already pending for a previously loaded data block that has not been It is. Therefore, CPU 24 returns to other operations (step 236).

Referring to Figures 6(a) and 6(b), the CPU handles interrupts as explained below. It responds to the interrupt by first clearing it (step 310). That's first Read the removal pointer 90 to determine the location of the next block in the buffer begins removing blocks of data from storage device 31 by (step 311) . CPU 32 then specifies the address of the first location in the block to be emptied. Initialize registers 56, 58 and read increment bit R to register 56. 2 (step 312). The CPU 32 then continuously reads from the storage device 31. Empties the block by performing a search and writing the data to cache 33. (step 314). When the entire block is emptied, CPU 32 points to the removal pointer. Update 90 to point to the next block in the buffer (step 316). .

Finally, CPU 32 updates count register 92 to remove the block from the buffer. reflect the past. For this purpose, CPU 32 first uses counting semaphore 9. 6 (step 318). The semaphore is set to 1, and the CPU 24 performs counting. If so, the CPU 32 returns the semaphore to O. (step 320). When the semaphore is cleared, C The PU32 reads the count from the register 92, decrements the register 92, and blocks the block. (steps 322 and 324). After decrementing the count, C PU32 releases the counting register by clearing semaphore 96 (stage Floor 326).

CPU 32 then examines the updated count to determine if the buffer is empty. (step 328). If the buffer contains another data block, then C PU 32 continues reading blocks until the buffer is emptied (step 330). ). When the buffer is empty, the CPU 32 receives a new interrupt as it appears on the bus 62. return to other tasks until it indicates that new data has been loaded into the data buffer. (step 332).

Control message blocks 74 and 76 operate in essentially the same manner to send control messages. transfer the file. More specifically, control register block 72 registers the next available register block 72. An entry pointer 98 indicating the location of the storage container and the next entry pointer to be emptied. It includes a removal pointer 100 indicating the location of the message container (FIG. 4).

It further includes a counter indicating the number of control messages stored in message block 74. It includes a number register 102. Semaphore block 70 goes to count register 102 associated semaphore 1 for use in arbitrating competing requests for access to Contains 04.

Similarly, control register block 72 provides information through control message block 76. Removal pointer 105, entry for use in controlling the passage of messages. It includes a pointer 106 and a count register 107. Semaphore block 70 is a separator for arbitrating between competing requests for access to counting register 107. Contains Mafo 108.

interrupt As explained above, CPU 24 writes the first data or message block. When loading into storage device 31, i.e. storage device 31 was previously empty), this Notify CPU 32 that the block is available by interrupting CPU 32 do. The mechanism for generating interrupts will now be described in more detail.

2 and 3, interface 30 generates an interrupt to CPU 32. It includes a control status register 66 that is used by CPU 24 to control the CPU 24. Furthermore Specifically, CPU 24 writes to register 66 and interrupts CPU 32. Sets a certain bit in a register that requests interface 30 to do so.

In order to write to register 66, CPU 24 reads the address given to register 66. Assert the dress. Decoders 44 and 46 decode this address and decode it. control register access signal R4. Register signal R4 Upon reception, register 66 transfers the data from data bus DB to its register cell. Herod.

CPU 24 sets interrupt bit IGI in control status register 66 and Loading the identification code IDI of 2 into bits 0 to 5 (MIDO to MID5) requests the interface 30 to interrupt the CPU 32 by. identification mark The number rD1 is applied to the comparator 110, which assigns rDl to the CPU 32. Compare with the identified identification code rcODE'J. Comparator output and interrupt generation bits IGI is applied to a three-input AND gate 112, thereby requesting an interrupt. seek

Interface 30 includes a second control status register 6 which is substantially the same as register 66. 8, which sets or clears interrupt enable bit rE2. enable or disable interrupt requests from the CPU 24 by It is used by the CPU 32 for this purpose. Interrupt enable bit IE2 is an AND game. will be added to the third manpower of 112. If IE2 is set, the other two inputs Activation of the power causes the output of AND gate 112 to become asserted, causing the interrupt alarm to become asserted. trigger the switch 114 to set it. When set, interrupt latch 114 Assert interrupt signal rlnterrupt2J on line 62. CPU32 is Sets the interrupt pending bit IP2 in its status register 68, which causes the Clears the interrupt by being cleared immediately.

CPU 32 interrupts CPU 24 in the same manner. More specifically, the CPU 32 Set interrupt bit rG2 in register 68 and set identification code ID2 to bit MID. OM [Load into D5. In addition to the single power of AND gate 113, Bit IGI It will be done. Identification code bit ID2 is added to the power of comparator 111.

Comparator 111 compares ID2 with l”Code2J. If so, the comparator provides a "match" signal to AND gate 113. most Later, the interrupt enable bit IEI from status register 66 is applied to AND gate 113. will be added to the third person force. All three inputs to AND gate 113 are asserted. If so, AND gate 113 sets interrupt latch 115.

When set, interrupt latch 115 outputs interrupt signal rln through cable 19. Assert interruptLJ. Interface 28 routes interrupts to bus 22 This interrupts the CPU 24. CPU 24 writes its status register 66 to Clear the interrupt by setting the interrupt pending bit IPI.

The interrupt mechanism also allows the backup file server to determine when the primary file server fails. equipment is being prepared. More specifically, in normal operation, the CPU 24 is Interrupting CPU 32 regularly with new data to be loaded onto disk 34 . If the CPU 32 does not receive an interrupt within a specified time, it is assumed that the main unit has failed. and begins to take on the responsibilities of the Lord.

A properly functioning primary file server may be running for an extended period of time due to lack of activity on the network. It is possible that no data will be sent to the backup file server during that time. Therefore, the Lord It also monitors the amount of time since it last interrupted CPU 32. finger If the specified time is about to expire, CPU 24 sends rAl to control block 74. ive message and interrupts the CPU 32, which has a buffer. Report that a message is being contained. CPU 32 responds to the interrupt and A1 1ve message and returns to its normal operation. In this way, the CPU 24 tells the CPU 32 during the time when data does not need to be copied to the backup disk 34. Even reports that it is operational. Similarly, the CPU 32 Regularly send live messages to the PU 24, and the CPU 32 is ready for operation. Report that the vehicle remains in

Figures 7(a) to 7(d) indicate that the CPU 24 is running for a specified period of time. If the interrupt to U32 fails, the backup file server assumes the main responsibility. Ru. For backup, the bar 16 sends a "stop" message to block 76 and stops it. informs the master that he/she is taking over (step 410). Backup file server 1 Step 6 then attaches the backup file system as a local file system (step 4). 12) and activates its network interface 116 (FIG. 1) (step 416).

Next is rAddress Re5solution ProtocolJ package The packet (rARPJ packet here) is sent to the network interface via link 14. 116, and all traffic that was previously directed to the will be processed. More specifically, each client connection point of the master is used by each attachment point recognized by the requester's operating system. It maintains a Node ID table containing the network interface addresses for each node.

The rA and RPJ packets are the main interface of the network interface 2o. Replacing the address with the address of the network interface 116 in the backup Instruct each client connection point to change its table by (stage N4 I 8 ).

After sending the ARP packet, the backup file server will delete the main old data block. All disk data blocks that are corrected so that they can be replaced at a future point. Begin keeping track of records. More specifically, the backup file server is Block CPU storage is allocated for storage of the map (step 420).

Each bit in this map corresponds to a single disk data block.

If a given block is written or otherwise modified, the journal The corresponding bit in the null bitmap is set.

A backup file server is used to update the newly generated journal bitmap. The server describes all changes to the file system within the last 15 seconds. The null file is first read (step 422). backup file server is journa The journal is examined for each data block modified in the last 15 seconds. The corresponding bit in the null bitmap is set (step 424).

The backup file server continues to monitor the journal file to update the file system. Update the bitmap with each change.

The primary file server will eventually be restarted after the failure condition is removed.

Once restarted, the master sends the [Al1veJ message to control message block 74 send to The backup file server receives “A11veJ Metsu” from the control message block. message and begin returning control to the master (step 426). for this purpose In order to do so, it first stops its network 116 from receiving more packets. C stage 428). That is all pending network packets to be processed next. Wait 15 seconds (step 430). More specifically, a backup file server is a file server. Are packets already being received from networks that are not yet integrated into the system? Maybe. Furthermore, it has already started preparing the packet for transmission by the network. It may be. Therefore, during the 15 second waiting period, the backup file server is Update the file system to reflect the desired changes and clear all pending changes. send a packet of

After waiting 15 seconds, the backup file server updates the disk data block allocation bitmap. 432). block allocation Each bit of the bitmap corresponds to a disk data block. set to 1 A set bit indicates that the corresponding disk block is in use, and a 0 indicates that the corresponding disk block is in use. Indicates that the disk block is free. The backup file server is then Examine the journal bitmap to see if this is the most recent change to the file system. (Step 1111434). If it is not reflected , the backup file server will journal BitMatsu to reflect the most recent changes. (steps 436, 438).

When the journal bitmap is updated, the file server Scans the disk for each disk data block modified in the 15 seconds before the main failure. (step 440). It then uses the data block transfer procedure described above. Sending each modified data block primarily via dual boat storage 31 (step 442) ). After completing this transfer, it is transferred to the CP containing the journal bitmap. Deallocate the U storage block (step 444) and set the ``Journal Complete'' message. Messages are sent primarily via control message block 76 (stage wI446).

Referring to Figures 8(a) and 8(b), the main file when returning to operation from a failure The operation of server 15 will be explained. The main thing is to first control the double boat storage device 31. Sends a “live” message to the backup file server through the message block 74 ( Step 510). Backup file server allocates disk data blocks When the master responds by sending a data block allocation bitmap, the master The bitmap is replaced with the newly arrived allocation bitmap (step 512). it is Then read each incoming modified data block from the dual port interface. begins writing the block to its disk 26 (step 514). reserve file The server sends a journal complete message to confirm that all blocks have been sent. (step 516), the master activates its network interface (step N 516), and attaching the file system (stage #520). It then uses the ARP send a message to all requester connection points, correct their connection point ID tables, and confirm that the host 522) to indicate that the service has returned to service.

It finally sends an "online" message to the backup via control message block 76. command the reserve to return to its role as a reserve (step 5). 24).

When the backup file server is first powered up, the contents of control register block 72 The contents become invalid (ie, in an unknown state). Therefore, the semaphore block 7° is A mechanism for notifying the CPU 24 and CPU 32 that the control entry is not valid. Includes power abuse muff R109 to provide structure. CPU24 and C The PU32 is configured to perform semaphore when each of these file servers is first powered. Read the bus line. Semaphore 109 is backed up by backup file server. indicates that control register block 72 has not been initialized. is set to O. CPU 24 which should read 0 in semaphore 109 and The first of CPUs 32 assumes responsibility for initializing these locations. Which When read by a new CPU, the semaphore is automatically set to 1 and Whichever CPU then reads the semaphore, the first CPU is responsible for initializing it. Let us know that you have accepted the offer.

Referring again to FIG. 1, the backup file server 16 is a backup file server and a second backup file server. It can act as both a primary file server. For this purpose, The file server 16 includes two network interfaces 116 and 118.

As explained above, interface 116 indicates that file server 16 has failed. Used to access communication link 14 when assuming primary responsibility.

Interface 118 is connected by link 14 in its capacity as a second primary file server. Used by file server 16 to communicate.

Additions, reductions, deletions, and other changes to the adopted specific embodiments of this invention are technical. and are within the scope of the following claims.

FIG. 6(b) Special table 15-508506 Cl0) FIG. 7(d) Fault-tolerant network file system Summary book A fault-tolerant network file server system consists of multiple Contains connection points. The main file server attachment point is a connection point that receives files from multiple attachment points. The backup file server connection point stores files from the main file server. memorize a copy of In the improved file server system, the main and Reports that the storage device contains data. In response to this interrupt, The processing unit in the file server reads data from the dual-ported storage device. Write this to the storage device in the backup file server. In a similar way, dual ports A storage device is used to send control messages between the primary and backup file servers. It will be done. Dual-ported storage devices provide a backup solution for accessing the same location. and semaphore storage locations to arbitrate competing requests by the primary file server. We are prepared.

international search report 1m - 1yu. N*　ρCT/uS　92103001

Claims (17)

[Claims] 1. network communication link Multiple attachment points connected to a network communication link, before connected to a network communication link a primary file server for storing files from multiple attachment points; and A backup file for storing copies of files from the primary file server mentioned above. server, In a fault-tolerant network file server system that includes The server is main computer processing unit, main memory disk, a first network interface connected to said network communication link; and a first network interface connected to said network communication link; It is connected to the preparatory file server and executes the above operations in response to commands from the main processing unit. a first independent interface capable of communicating information to a backup file server; It contains The backup file server mentioned above is back-up computer processing equipment; backup memory disk, a backup network interface connected to said network communication link, and said main network interface; connected to said first independent interface to receive information from the file server; and stored in the storage device in response to a command from the backup processing device. a second independent interface capable of reading information stored in the main interface; Equipped with a dual-ported storage device for storing information received from a file server said second independent interface, An improvement consisting of. 2. Said second independent interface further interrupts said back-up computer processing device. Then, the dual port storage device is connected to the main file storage device in the backup computer processing unit. includes an interrupt device to notify that the server is accommodating information received from the server. 2. The file server system according to claim 1. 3. The dual port storage device described above is a group of data for storing data to be replicated to the backup storage disk; memory block, to remember the entry pointer to identify the next available data block. entry pointer register for Exclude to remember the purge pointer that identifies the next data block to be emptied. left pointer register, and a counting register for storing the number of data storage blocks containing data; 2. The file server system of claim 1, comprising: 4. said dual-ported storage device further comprises said primary file server to said subsequent a first control message block for storing control messages to the backup file server; and a control link from the backup file server to the primary file server. a second control message block for storing messages. 3 file server system. 5. The second independent interface further comprises: the contents of a common register accessible to a back-up computer processing unit in response to a signal from A device that can be modified to indicate the entry of data into a storage block. and the second independent interface further includes a back-up computer processing device. The common register is activated to remove data from the storage block in response to a signal from The above dual-ported storage device can be modified as shown in the above common register. 4. The method of claim 3, further comprising a semaphore register for arbitrating access to the star. file server system. 6. Said semaphore register has first and second operating states, whereby Therefore, when read by the processing unit in the first state mentioned above, the semaphore register is The register automatically enters the second state with the register available code and When read by a processing unit while in state 2, the semaphore register described above becomes a register. remain in the second state with a register unavailable code; When written by a processing unit at some time, said semaphore register is 6. The file server system according to claim 5, wherein the file server system returns to the state. 7. The common register is the counting register, and the main file server is The server increments the count register to register the data in the data storage block. and the back-up processing device has a device for indicating the bird, and the back-up processing device to indicate the removal of data from the data storage block by decrementing the count register of 7. The file server system of claim 6, further comprising a device for. 8. at least one of said dual-ported storage devices having first and second operational states; Contains one semaphore bit, which allows the second independent interface to said at least one semaphore bit is in said first state when said base is powered on; the at least one semaphore when read by the processing device. 2. The file server system of claim 1, wherein the file server system automatically enters said second state. 9. The main file server further includes: said main storage disk of a file received by said first network interface; A disk operator for controlling replication to the disk and dual-ported storage device described above. This disk operating system contains The above operation is performed in response to a reliable disk write command and an uncertain disk write command. Instructs the rating system to write the specified file to the main storage disk mentioned above. includes a device for instructing said reliable command to write said uncertain Requests writing to the main storage disk more quickly than the instruction, and causes the operator to In response to certain and uncertain commands, the to the dual-ported storage device described above, and using the unreliable write procedure described above. The file server of claim 1, wherein a specified file is written to the main disk storage device. server system 10. If the disk operating system mentioned above responds to synchronous disk write instructions. responsively writing said specified file to said dual-ported storage device; and Also, write the specified file to the main storage disk using the delayed write procedure. 9. The file server of claim 8, which is a modified Unix operating system. server system. 11. A backup file system that reflects the primary file system of the primary network attachment point. A method for supporting a system at a back-up network connection point, the method comprising: , using a secure write procedure at the aforementioned main connection point in response to a secure disk write command. directing other specified files to be written to main persistent storage; At the main connection point, in response to an uncertain disk write command, the main connection point writes the specified file to main persistent storage using unreliable write procedures. At that time, the above-mentioned secure write procedure writes the above-mentioned specified file to the above-mentioned uncertain write. writing to said main persistent storage faster than said procedure; In response to either a certain command or an indeterminate command, the specified file is forwarded. to the shared storage in the backup connection point, and using the unreliable write procedure described above. writing the specified file to the persistent storage device; The aforementioned method comprising: 12. Writing a specified file to the shared storage device From the entry pointer register in the device, the next available reading an entry pointer identifying the data block portion; writing said specified file to said next available data block; and the method further comprises: from the removal pointer register in said shared storage to be emptied; reading a removal pointer identifying a next data block portion of the storage device; Beauty The next data block portion is used to store the backup file system. a step of replicating to the storage device of 12. The method of claim 11, comprising: 13. Writing the specified file to the shared storage device further writes the specified file to the shared storage device. From the counting register in the storage device, the shared storage containing the specified file a count identifying the number of data blocks within said data block portion of the device; a step of reading, and Increment the count register mentioned above and move the specified file to the next available data block. and the method further comprises: Decrement the count register above to obtain the data from the next data block portion to be emptied. 12. Indicating said reading of a specified file. the method of. 14. The method further comprises: The counting register is available from the semaphore register in the shared storage. reading an available code indicating whether the counting register is After said increment or decrement, said semaphore register is reset and said counting register is 14. The method of claim 13, further comprising the step of: indicating that the star is available. Method. 15. From the power-up semaphore register in said shared storage, said shared Selected registers in said shared storage device are first accessed after the storage device is powered. reading an initialization code indicating whether the selected period has been set; If the initialization sign above indicates that the registered register was not initialized, then For example, the method further comprises the step of initializing the selected register. 12 ways. 16. The main connection point operates by interrupting the backup processing device in the backup connection point. Claim further comprising the step of notifying the backup connection point that the connection point is possible. 11 methods. 17. Interrupting the back-up processing unit sends the live control message to the shared writing to a control message block in storage; and The shared storage device interrupts the back-up processing device and sends the control message to this processing device. Claim 16, further comprising the step of notifying the user that the product contains message. the method of.