WO2003090089A1

WO2003090089A1 - Cache device

Info

Publication number: WO2003090089A1
Application number: PCT/JP2002/003973
Authority: WO
Inventors: Masanori Goto; Yoshitake Shinkai
Original assignee: Fujitsu Limited
Priority date: 2002-04-22
Filing date: 2002-04-22
Publication date: 2003-10-30
Also published as: JPWO2003090089A1

Abstract

A cache system having two cache devices which holds data given from an access host or a secondary storage device, and stores the data in the secondary storage device. One cache device which receives data from the access host stores the data in a cache memory thereof, and transmits the data to the other cache device. The other cache device receives the data transmitted from the one cache device, and stores the data in a cache memory thereof. The one cache device or the other cache device outputs the data stored in the cache memory to the secondary storage device, and stores it. After storing the data in the secondary storage device, the cache device holding the data transmits a flash message to the other cache device. Since the data is stored in each of the two cache devices, any bottleneck of when the cache system has a single cache device can be eliminated, and the data is prevented from being lost even when one cache device fails.

Description

Description Cash device Technical field

The present invention relates to a cache device for temporarily storing data, a cache system, and a cache method. In particular, the present invention has two cache devices, and refers to data provided from an access host or a secondary storage device. In addition, each cache device, a cache system, and a cache method in a cache system for storing data from an access host in a secondary storage device are described. Background art

In the past, secondary storage devices such as hard disks were used only by one access host, but in recent years, multiple secondary storage devices shared by multiple access hosts via a storage network or the like have been used in recent years. Secondary storage devices consisting of are widely used. For these secondary storage devices, data is read and written from the access host via the storage network.

Due to the structure of such a secondary storage device, the throughput / response performance is often not sufficiently exhibited depending on the required access pattern. In addition, when using RAID (Redundant Array of Inexpensive Disks) 5, the overhead required for writing is large.

Therefore, performance is improved by using a cache device (cache memory) to temporarily store data that is used relatively frequently and data that is written to a secondary storage device in the cache device. Is commonly done.

A cache device plays a role of temporarily storing data read from or written to a secondary storage device using a high-speed primary storage device. It has a great effect on improving the performance of

In particular, data such as storage virtualization address translators and file system appliance devices are shared between the shared secondary storage device group and the access host. If it is necessary to go through a device that intercepts in the middle, using a cache device on that device is an effective method for improving performance.

However, a single cache device often becomes a bottleneck and degrades performance. That is, when there is a single cache device, accesses to the secondary storage device group pass through only one cache device. For this reason, even if the number of access hosts and secondary storage devices increases, a sufficient throughput for the cache device cannot be secured, and the overall performance is determined by the performance limit of the cache device. In addition, a single cache device loses its data due to a failure in its cache device, resulting in problems such as increased downtime.

In other words, the state where data exists before being written to the secondary storage device group in the cache device, and the data exists only in the cache device can easily occur at any time. If a failure occurs in the cache device in such a state, data that exists only in the cache device will be lost, making it unrecoverable.

Also, if the downtime of the system due to a cache device failure increases, the system will be damaged economically and in reliability.

To solve the bottleneck problem, there is a method to solve this problem by providing multiple cache devices. This aims to eliminate bottlenecks and secure scalability by assigning multiple cache devices to multiple access hosts.

However, if one access host writes data to one cache device and writes data with the same address but different contents to the other cache device to the other cache device, the caches of both caches will be lost. Even data with different contents between devices will be cached. If this state is left unchecked, cache data for a certain address will remain mismatched between cache devices, and consistency will not be maintained.

In addition, if one cache device stops due to a failure, another cache device will have a different cache, and the transparency will not be maintained.

In addition, as long as the cache devices operate independently, The problem of irreparable loss of a single crash due to a single failure remains unresolved.

On the other hand, there is a method to maintain the consistency of data written between multiple cache devices.

A common way to maintain this consistency is to use tokens. In this method, information called a token is communicated between multiple cache devices, and data consistency is guaranteed by exclusive control. However, token control requires a relatively large amount of time due to the high message cost (token communication cost) and the communication time of the token. Therefore, this method has the problem that it may be a bottleneck in devices provided for improving processing speed, such as cache devices.

There are also systems that use memory shared by multiple cache devices and perform exclusive control on this memory. A cache device that uses such a shared memory has the problem that the cache memory is lost if a failure occurs in the shared cache memory device and its management unit. If volatile memory is used as the cache memory, data may be lost due to a power failure. For this reason, it is necessary to use a non-volatile memory, but there is a problem that the cost increases accordingly.

Thus, when using a cache device, many existing methods are expensive, despite the need to maintain consistency and transparency over the written data. Etc. Disclosure of the invention

An object of the present invention is to eliminate bottlenecks in a single cache device.

Another object of the present invention is to prevent loss of data stored in a cache device even when a failure occurs in the cache device.

The cache device according to the present invention has two cache devices, stores data provided from an access host or a secondary storage device, and A cache device in a cache system for storing data from a host in the secondary storage device, wherein the data input unit inputs first data provided from the access host; A data receiving unit for receiving the second data input from the access host by the device and transmitted to its own cache device, and / or the first data and / or the second data. A cache storage unit that stores one of the cache units; a cache management unit that manages the cache storage unit; a data transmission unit that transmits the first data to the other cache device; And a data output unit for outputting the second data to the secondary storage device.

A cache system according to the present invention has two cache devices, stores data provided from an access host or a secondary storage device, and stores data from the access host in the secondary storage device. A cache input system for inputting first data provided from the access host; and a cache input device for inputting the other data from the access host. A data receiving unit that receives the second data transmitted to the host; a cache storage unit that stores the first data and / or the second data; and a cache that manages the cache storage unit. A cache management unit; a data transmission unit for transmitting the first data to the other cache device; A data output unit that outputs the first data or the second data to the secondary storage device.

A cache method according to the present invention has two cache devices, stores data provided from an access host or a secondary storage device, and stores data from the access host in the secondary storage device. A cache method in a cache system for remembering, wherein one cache device receiving data from the access host stores the data in its own cache memory and stores the data in the other cache device. The other cache device receives the data transmitted from the one cache device, stores the data in its own cache memory, and stores the one cache device or the other cache device. The cache device outputs the data to the secondary storage device. According to the present invention, two cache devices are provided in the cache system. Each cache device inputs the data to be stored in the secondary storage device from the access host and outputs (stores) the data to the secondary storage device. Therefore, it is possible to obtain twice the processing capacity of a single cache device and eliminate the bottleneck of a single cache device.

Further, according to the present invention, one of the cache devices that has received the data from the access host stores the data in its own cache memory, and transmits the data that has been input to the other cache device. Send. The other cache device stores the data sent from one cache device in its own cache memory. As a result, non-volatile memory in the cache system is achieved overnight. Therefore, even if a failure occurs in one of the cache devices in a state where the data is not stored in the secondary memory device, the data can be obtained from the other cache device, and the data due to the failure can be obtained. Loss is prevented.

Preferably, after the data input unit inputs the first data and before the data transmission unit completes the transmission of the first data, the data receiving unit transmits the second data. When the first data is received, the first data and the second data are used, and based on the address range on the secondary storage device and the contents of both data, the first data and the second data are used. A collision detection unit that determines the presence or absence of a collision between the first data and the second data; and when the collision detection unit detects a collision, transmits a collision detection message indicating that a collision has occurred to the other cache device. And a collision detection message receiving unit that receives the collision detection message from the other cache device.

The collision detection unit also detects a collision when the collision detection message receiving unit receives the collision detection message from the other cache device.

This makes it possible to detect a collision between these two data even if two caches with different contents in the same address range are received from the access host. .

When a collision is detected, the cache management unit gives priority to the first data and the second data based on a predetermined priority. Can be treated as valid and the other as invalid.

Alternatively, when a collision is detected, the cache management unit may determine that the time at which the first data was input to the data input unit and the second data are the data of the other cache device. Among the times input to the input unit, the data with the earlier time

—Even if the evening is treated as valid, a night with a late time can be treated as invalid.

Further, when a collision is detected, the cache device transmits, after a lapse of a random time, the first data or a retransmission message indicating the retransmission of the first data. The second data or the second data transmitted from the data / message retransmitting unit of the other cache device.

A cache / message receiving unit for receiving an evening retransmission message, the cache management unit including: a retransmission time by the data / message retransmission unit; and a reception by the data / message receiving unit. Of the times, the data corresponding to the earlier time can be treated as valid, and the data corresponding to the later time can be treated as invalid.

Either of these can resolve the collision condition and ensure data consistency and transparency between the two cache devices.

Preferably, the data transmission unit includes, together with the first data, a first flash indicating which cache device outputs the first data to the secondary storage device. The data receiving unit transmits authority information, and the data receiving unit indicates, with the second data, which cache device outputs the second data to the secondary storage device. Receiving the flash right information, and outputting the first data to the secondary storage device when the first flash right information indicates the self-cache device; If the second flush right information indicates the own cache device, the second flush right is output to the secondary storage device. This allows load balancing between the two cache devices.

Preferably, the cache device monitors the occurrence of a fault in the other cache device, and upon detecting the occurrence of the fault, the first data stored in the cache storage unit. Of the above, the data output unit or the data output unit of the other cache device has not completed the output to the secondary storage device, and the data transmission unit does not transmit the data to the other cache device. Of the completed data and the second data stored in the cache storage unit, wherein the data output unit or the data output unit of the other cache device has completed output to the secondary storage device. A fault monitoring unit that controls the data output unit to output the missing data to the secondary storage device. As a result, even if a failure occurs in one of the cache devices, data that exists in the cache device but does not exist in the secondary storage device can be reliably flushed (stored) in the secondary storage device. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a block diagram showing the overall configuration of a secondary storage device access system using a cache system according to an embodiment of the present invention.

FIG. 2 shows a configuration example of a cache control table held by the cache management unit. Figure 3 shows an example of the data structure of a control message communicated between message communication units.

Figure 4 is a sequence diagram showing the flow of the process of writing storage data sent from the access host to the secondary storage device group.

Figures 5A to 5C are sequence diagrams showing the flow of the collision detection process when a collision occurs between both cache devices.

Figure 6 is a sequence diagram showing the flow of storage data write processing when load distribution is performed.

FIGS. 7A to 7E show a case where the address ranges of the two stored data overlap. BEST MODE FOR CARRYING OUT THE INVENTION ''

FIG. 1 is a block diagram showing the overall configuration of a secondary storage device access system using a cache system according to an embodiment of the present invention. This secondary storage device access system (hereinafter simply referred to as “access system”) consists of a cache system 3, an access host group 4, a secondary storage device group 5, an access network 6, And a storage network 7.

The cache system 3 has two cache devices 1 and 2. Cache devices 1 and 2 have the same configuration. The cache device 1 includes a cache management unit 11, a cache memory 12, input / output units 13 and 14, a message communication unit 15, and a failure monitoring unit 16. The cache device 2 has a cache management unit 21, a cache memory 22, input / output units 23 and 24, a message communication unit 25, and a failure monitoring unit 26.

Access Host group 4, n (n is an integer of 2 or more) having an access host 4 I～4 _n of. Each access host writes data (hereinafter referred to as “storage data”) to the secondary storage device group 5 via the cache system 3, and writes the data to the secondary storage device group 5 (or cache storage device 5). The stored data stored in the system 3) is read out via the cache system 3. Each access host is composed of, for example, a convenience store.

Secondary storage device group 5 is a secondary storage device that will be shared by the access host access host group 4, m pieces (m is an integer of 2 or more) have a secondary storage device 5 5 _m of. Each secondary storage device is assigned a device number (for example, a serial number) to uniquely identify it, and the access host group 4 and the cache system 3 can specify this device number by specifying this device number. , One secondary storage device in the secondary storage device group 5 can be specified, and by specifying the address (or block number), the storage data on the specified secondary storage device can be specified. . Each secondary storage device is composed of, for example, a hard disk, a magneto-optical disk (MO), and an optical disk (for example, DV D-RAM).

The access network 6 includes, for example, an SCS I network, a fiber channel, a LAN (Ethernet), and the like. The storage network 7 is composed of, for example, Fiber Channel.

The access host i An, the input / output unit 13 of the cache device 1, and the input / output unit 23 of the cache device 2 are connected to the access network 6. As a result, the stored data is transmitted to the cache device 1 or 2 via the access host 4! -4 ^^: and the access network 6, and transmitted from the cache device 1 or 2. You can receive the stored data overnight.

In addition, the IS device 5 i to 5 _m , the input / output unit 14 of the cache device 1 and the input / output unit 24 of the cache device 2 are connected to the storage network 7. As a result, the cache devices 1 and 2 transmit (write) the storage data to the secondary storage device group 5 via the storage network 7, and receive the storage data from the secondary storage device group 5. Yes (read).

Cache devices 1 and 2 are configured independently of each other, so that one fault does not affect the other. As described later, the cache devices 1 and 2 independently receive I / O from the access host group 4 and maintain twice the performance of one cache device while maintaining the transparency of the storage device. Can be obtained.

The input / output unit 13 (23) (codes in parentheses indicate corresponding components in the cache device 2. The same applies to the following.) Is for communication such as protocol processing for stored data transmitted and received via the access network 6. Execute the process. In addition, the input / output unit 14 (24) executes communication processing for stored data transmitted and received via the storage network 7.

The cache memory 12 (22) is composed of a storage device (RAM, etc.) that can access (read and write) faster than the secondary storage devices of the secondary storage device group 5.

The cache management unit 11 (21) holds a cache control table (described later), and manages the storage data stored in the cache memory 12 (22) based on the cache control table. The cache management unit 11 (21) controls the input / output units 13 (23) and 14 (24), the failure monitoring unit 16 (26), and the message communication unit 15 (25). 13 (23) or 14 (24) or writing of storage data input from the message communication unit 15 (25) to the cache memory 12 (22), transmission from the input / output unit, cache memory 12 (22) Transmission of the storage data read from the I / O unit via the input / output unit 13 (23) or 14 (24), transmission / reception of the control message with the other cache device and storage data via the message communication unit 15 (25), etc. I do. Fig. 2 shows an example of the configuration of the cache control table held by the cache management unit 11 (21). The cache control table is provided for the cache memory 12 and for the cache memory 22. The cache control table is maintained by the cache management unit 11 for the former, and is retained by the cache management unit 21 for the latter. You.

The cache control table has a cache control list of each storage device currently stored in the cache memory 12 (22). Each cache control list has, as data items, an element number, a device number, a device start address, a data length, a cache start address, a status, and a flag.

The “element number” is the element number of the storage data currently stored in the cache memory 1 2 (2 2). Here, one element corresponds to a set of stored data written from or read from a certain access host in the access host group 4. One element may correspond to one byte of storage data, or may correspond to multiple bytes of storage data.

For example, from a certain access host 4 i (i is an integer from l to n), a storage data with a block of 512 bytes is transmitted, and the cache memory 1 2 (2 2) stores this 1 block. When stored data is stored, one element corresponds to one block (512 bytes) of stored data. When two blocks of stored data are transmitted from another access host 4 (j is any of l to n) and the stored data of these two blocks is stored in the cache memory 12 (22). , One element corresponds to two blocks of memory.

The “device number” is the device number (for example, a serial number) of the secondary storage device of the secondary storage device group 5 in which the storage data is to be stored. “Device start address” is the storage start address (start address) of the storage device in the secondary storage device of the corresponding device number. “Data length” is the length (number of bytes) of the corresponding storage data.

When the storage data is read from each secondary storage device in blocks of a plurality of bytes (for example, 512 bytes) and written to each secondary storage device, the device start address is used. Let λ be the starting block number where the first block is written, and let the length be the end block number where the last block is written. Can also. For example, when two blocks of storage data are stored in the storage area of the fifth block and the storage area of the sixth block of the secondary storage device, the start block number may be 5 and the end block number may be 6.

The “cache start address” is the storage start address (start address) of the storage data in the cache memory 12 (22).

The “state” indicates the state of the storage device, which includes “received” (hereinafter referred to as “received”), “duty” (hereinafter referred to as “dirty”), and “non-volatile” ( Hereinafter, "non-volatile", "flushing" (hereinafter, "flushing"), "flash message" (hereinafter, "flushed msg"), "clean" (hereinafter, "clean") , And "invalidated" (hereinafter "invalidated").

The received state is a state in which the stored data is received from the access host, the state until the transmission of the stored copy message (described later) is completed, or a first acknowledgment message (described later) for the copy message. ) The state until is received.

The dirty state refers to a state in which the first acknowledgment message regarding the stored data has been received, and the stored data has not been written to the secondary storage device. The non-volatile state is a state in which storage data included in a copy message transmitted from the other cache device is stored in the cache memory, and this storage data has not yet been written to the secondary storage device group. Say.

The flushing state differs from the dirty state in that the storage data is being written to the secondary storage device group.

The flushed msg state changes from the non-volatile state to the clean state when the same storage data stored in the other cache device has been completely written to the secondary storage device (flash). This is the state in which a flash message (described later) is notified to the other cache device so as to change to the other cache device.

The clean state refers to the state in which writing to the secondary storage device in the storage device has been completed and notification to that effect to the other cache device has been completed. In this state, the data stored in the cache memory 12 (22) can be overwritten or deleted at any time. Although it can be discarded, it is stored in the cache memory 1 2 (2 2) in preparation for reading out this memory from the access host group 4 overnight.

The invalidated state is a state in which the stored data has been invalidated after the clean state. In the storage device in this state, discard processing such as overwriting and erasing is performed immediately or after a predetermined time has elapsed.

The “flag” indicates whether or not the flash is authorized, whether or not writing is in a collision state, whether or not a host acknowledgment message (ACK) has already been returned to the access host. The presence / absence of the flash authority, the write collision status, and the acknowledgment message will be described later.

When the cache management unit 11 (21) receives the data from the access host group 4 or the secondary storage device group 5 and processes it, the cache management unit 11 (21) refers to the cache control table held by itself, and as a result of the processing, Update the contents of the cache control table as needed.

The message communication units 15 and 25 are connected to each other by a communication line L, and send and receive storage messages and control messages to and from each other. As the communication line L connecting the two, for example, PCI, Peripheral Component Interconnect Bus, Gigabit Ethernet, or the like is used, and its communication speed is preferably higher than that of the access network 6.

Fig. 3 shows an example of the structure of a control message transmitted between the message communication units 15 and 25. The control message has a header section in which control information is stored and a data section in which stored data is stored. The header contains data items of type, device number, device start address, data length, sequence number, status, and flash authority.

“Type” is the type of the message. This type includes "COPY" indicating a copy message, "C-ACK" indicating an acknowledgment message (first acknowledgment message) for a copy message, "FLUSHED" indicating a flash message, and a flash message. "F-ACK" indicating the acknowledgment message (second acknowledgment message), "COLLISION" indicating the collision detection message, and "COL-" indicating the acknowledgment message (third acknowledgment message) for the collision detection message ACK "and" MONITOR "indicating a failure monitoring message. This Each of these control messages will be described later.

The “device number” is the storage data added to the data section in the case of a copy message.

—The device number of the secondary storage device where the evening is stored, in the case of a flash message, the device number of the secondary storage device in which the flashed storage device is stored, and in the case of a collision detection message, In the case of an acknowledgment message, the device number is the same as the device number of the corresponding copy message, flash message, collision detection message, etc.

The “device start address” is the start address of the secondary storage device that stores the storage data added to the data section in the case of a copy message, and the flashed storage in the case of a flash message. ^ —The start address of the secondary storage device where the evening is stored, and in the case of a collision detection message, the start address of the secondary storage device where the storage location where the collision is detected is stored, and the acknowledgment message In this case, the start address is the same as the start address of the corresponding copy message, flash message, collision detection message, and the like.

The “delivered length” is the storage data added to the data section in the case of a copy message—the length of the evening (number of bytes, blocks, etc.), and the flashed memory in the case of a flash message. In the case of a collision detection message, this is the length of the stored data at which the collision was detected. The value of the data length of the acknowledgment message is set to 0.

“Status” is the same data as the status in the cache control table described above. “Sequence number” is the serial number assigned to the transmitted control message. Each of the cache management units 11 and 21 sequentially assigns a serial number starting from 1, for example, to the control message transmitted by itself in accordance with the transmission order. This makes it possible to clarify the time jl order of the control message to be transmitted. Therefore, the cache management unit 11 (21) on the receiving side manages the sequence numbers of the control messages to be received, so that the transmission order of the control messages is different from the transmission order even if the reception order is different from the transmission order. You can know the order exactly.

The “flash authority” is defined as a cache device that performs a flash (a process of storing the stored data stored in the cache memory 12 (22) in a secondary storage device). Shows whether the device is 1 or 2. If the flush authority is not specified, this area is set to a value other than the value indicating cache devices 1 and 2 (for example, Null 1).

The failure monitoring unit 16 (26) monitors the failure of the other cache device, and if a failure is detected, performs the necessary processing to recover from the failure. For example, the failure monitoring units 16 (26) mutually transmit and receive failure detection messages via the message communication unit 15 (25) at regular time intervals. If one fault monitoring unit does not receive a fault detection message from the other fault monitoring unit within a predetermined time, it determines that a fault has occurred in the other cache device.

Further, the failure monitoring unit 16 (26) detects the occurrence of a failure in the other cache device, and is held by the cache management unit 11 (21) of the cache device 1 (2) to which it belongs. Rewrites the status of the cache control table as needed, and executes failure recovery processing. Details of the failure recovery processing will be described later.

In the access system having such a configuration, the access host group 4 writes the storage data to the secondary storage device group 5 via the cache system 3, and the secondary storage device via the cache system 3. Read the stored data from group 5. The following describes the details of the write processing and the read processing, and the failure recovery processing when one of the cache devices in the cache system 3 fails.

The processing when one of the access hosts 4 i of the access host group 4 writes the stored data to the secondary storage device group 5 via the cache device 1 or 2 will be described below.

FIG. 4 is a sequence diagram showing a flow of a write process to the secondary storage device group 5 in the storage device transmitted from the access host 4i.

First, the access host 4 sends the stored data to the cache device 1 or 2 via the access network 6. Whether the storage device is to be transmitted to the cache device 1 or 2 'is preset in the access host 4i. It may be set to transmit to only one of the cache devices 1 and 2, or The transmission may be set to be alternately transmitted to the hash devices 1 and 2. The access host 4 sends a write request signal to the cache devices 1 and 2, and the cache device 1 or 2 that is in the idle state returns a data receivable message to the access host _{4 i} . Thus, the transmission of the stored data may be started (when both the cache devices 1 and 2 transmit the data receivable message, the access host 4i selects one). The following describes an example in which stored data is transmitted from the access host 4i to the cache device 1.

The access host 4 i transmits the storage data and the control data to the input / output unit 13 of the cache device 1 via the access network 6. The control data is added, for example, as a header for storing data. This control data includes the device number, device start address, and data length of the secondary storage device to be stored in the storage data (hereinafter, the device number, device start address, and data length are referred to as “address range”). ) Is included.

The input / output unit 13 gives the control data and the storage data transmitted from the access host 4 i to the cache management unit 11.

The cache management unit 11 stores in the cache memory 12 the storage data (storage data b) having an address range that overlaps the address range of the storage data (storage data a) transmitted from the access host 4i. Whether it is stored or not is determined based on the address range of the cache control table of the cache memory 12.

If there is no storage data b having an address range that overlaps with the address range of the storage data a, the cache management unit 11 generates a cache control list for the storage data a and stores it in the cache control table. Add to it. Then, the cache management unit 11 writes the stored data a to the memory cell of the cache memory 12 starting from the cache start address of the generated cache control list (S1).

On the other hand, when the address ranges overlap, as shown in Figs. 7A to 7E, (1) When the address range of the storage device a is exactly the same as the address range of the storage device b (Fig. A), (2) When the address range of storage data a includes the address range of storage data b (Fig. 7B), (3) The address range of storage data a is the storage data. (B) When the address range of the storage device (a) and the address range of the storage device (b) partially overlap with each other, (Figs. 7D and 7E).

In the case of (1), the cache management unit 11 updates the device number, device start address, and the like of the cache control list of the storage data b to those of the storage device a. Alternatively, the cache management unit 11 generates a cache control list for the storage device a, adds it to the cache control table, and sets the status of the cache control list for the storage device b to "invalidated". , Or the cache control list in the storage device b may be deleted. Then, the cache management unit 11 writes the stored data a to the area of the cache memory 12 starting from the cache start address in the cache control list (S1).

The area on the cache memory 12 where the storage data a is written may be the same as the area on the cache memory 12 where the storage data b was stored, or may be another free area. Is also good. In the former case, the stored data b is overwritten by the stored data a. In the latter case, the stored data b remains in the cache memory 12 but the cache $ IJ list is set to "invalidated" or is deleted from the cache control table (overwriting is disabled). It is not treated as a valid memory. Therefore, the stored data b will be overwritten by another stored data thereafter. The same applies to the following cases (2) to (4).

In the case of (2), the same processing is performed as in the case of (1) (S1). Therefore, the cache control list of storage data b is deleted from the cache control table.

In the case of (3), the cache management unit 11 generates a cache control list of the storage data a and adds it to the cache control table for the part where the address ranges overlap, and stores the storage data a in the cache control table. Write to the cache memory 1 2 (S 1). In addition, the cache management unit 11 generates (or updates) a cache control list for each of the two parts of the storage data b except for the part that overlaps the storage data a, and generates a cache control table. To be added.

In the case of (4), the cache management unit 11 A list is generated and added to the cache control table, and the stored data a is written to the cache memory 12 (S1). In addition, the cache management unit 11 generates or updates a cache control list for a portion (one) of the storage data b except for a portion that overlaps with the memory storage a, and generates a cache control table. It comes with calories.

"Received" is written in the cache control list state of the storage data a, indicating that the stored data a is in the received state.

Subsequently, the cache management unit 11 transmits a copy message (COPY) to the message communication unit 25 of the cache device 2 via the message communication unit 15 and the communication line L (S2). The device number, device start address, and data length of the cache control list of the stored data are written in the device number, device start address, and data length in the header of the copy message. . In addition, in the data message section of the copy message, there is a memory message a.

The message communication unit 25 gives the copy message transmitted from the cache device 1 to the cache management unit 21. When the cache management unit 21 receives the copy message, the cache management unit 21 executes the same processing as that of the above-described step S1 of the cache management unit 11 on the memory a in the data portion of the copy message. I do.

That is, the cache management unit 21 updates or generates the cache control list based on the cache control table of the cache memory 22 and writes the stored data a to the cache memory 22 (S7). When all of the stored data a is written to the cache memory 22, the stored data a is not lost even if one of the cache devices 1 and 2 stops due to a failure. That is, the nonvolatile storage of the stored data a is completed. Therefore, to indicate this non-volatility, "non-volatile" is written in the state of the cache control list of the storage device a held by the cache management unit 21.

After writing the stored data a to the cache memory 22, the cache management unit 21 sends a first acknowledgment message (C-ACK) indicating that the writing has been completed normally to the message communication unit 25 and the communication line L. The message is transmitted to the message communication unit 15 via the server (S8).

The message communication unit 15 sends the first acknowledgment message to the cache management unit 11 Give to. When the cache management unit 11 receives the first acknowledgment message from the message communication unit 15, the cache management unit 11 sends an acknowledgment message for the access host (host acknowledgment message) to the input / output unit 13 and the access network 6. It is transmitted to the access host 4i via the server (S3). By transmitting the host acknowledgment message, the access host 4 knows that the storage data a has been stored in the cache system 3 (cache devices 1 and 2) and has been nonvolatilely stored.

When transmitting the host acknowledgment message, the cache management unit 11 updates the state of the cache control list of the server a from "received" to "dirty".

Subsequently, at an appropriate timing thereafter, the cache management unit 11 updates the state of the cache control list corresponding to the storage data stored in the cache memory 1.2 to "flushing", and updates this storage data. the Isseki a, and transmits the secondary storage device of the output unit 1 4 and through your storage network 7 secondary storage device group 5 (shall be the _{5 k.) (S 4)} . The secondary storage device 5 _k is a secondary storage device corresponding to the Kiyadzushi Interview Control List Bok device number of the storage de Isseki a. Transmitted stored de Isseki a is Ru written in an area starting from the device start address in the secondary storage device 5 _k (S 4).

When the transmission (writing) of the storage data “a” to the secondary storage device 5 _k is completed, the cache management unit 11 updates the state of the cache control table to “flushed msg” and updates the flash message. The message (FLUSHED) is transmitted to the message communication unit 25 via the message communication unit 15 and the communication line L (S5). This flash message is given from the message communication unit 25 to the cache management unit 21. As a result, the cache management unit 21 knows that the flushing (that is, the non-volatile storage of the storage device a in the secondary storage device) has been completed, and safely stores the storage data a stored in the cache memory 22. Recognize that it can be erased.

Subsequently, the cache management unit 21 updates the state of the cache control list of the storage data stored in the cache memory 22 to “clean” and notifies the cache device 1 that the state has been shifted to the clean state. To notify, a second acknowledgment message (F-ACK) is transmitted to the message communication unit 15 via the message communication unit 25 and the communication line L (S9). Upon receiving the second acknowledgment message from the message communication unit 15, the cache management unit 11 updates the state of the cache control list to “clean” in order to treat the stored data a as a clean state. .

The cache management units 11 and 21 can always update the status of the cache control list in the clean state to "invalidated" (S6, S10). For example, when the storage data a in the clean state becomes unnecessary or when the cache memory 12 or 22 runs out of free space and the storage data a needs to be erased, the storage data a is deleted. The status can be invalidated. Also, by using the LRU (Least Recently Used) algorithm, the storage data a that has not been read most can be made invalidated.

The cache control list in the invalidated state is then deleted from the cache control table when a new storage data is written to the cache memory, or overwritten by the cache control list of another newly stored data. Will be done. In addition, the area of the cache memory in the invalidated storage area is also overwritten by another new memory area.

Thus, the storage data write processing of the cache system 3 ends. The cache devices 1 and 2 of the cache system 3 can independently receive the storage data from the access host group and execute the write processing independently. Therefore, the cache system 3 has almost twice the processing capacity as compared to the case where only one cache device exists. This eliminates the bottleneck when there is one cache device. Further, since the storage data is held in at least one of the cache devices or the secondary storage device group 5, even if a failure occurs in one of the cache devices, the stored data is not lost.

When the cache device 2 receives the stored data from the access host 4i, only the cache device 1 and the cache device 2 are switched, and the same processing as described above is executed.

In step S7, the cache device 2 accesses another storage data (storage data c) having the same address range as the storage data a transmitted from the cache device 1 via the communication line L. Received from host 4 j and stored data c May be placed in the received state in the cache device 2. That ₃ have the same address range, a different storage de Isseki a and c of the content, Kiyadzushi Interview apparatus 1 and 2, almost simultaneously, there is a case where it it received from different access host. The processing in this case will be described later in the processing at the time of collision of write data.

In step S1, the stored data b overwritten and erased by the stored data a is in the received state or the flushed msg state, and the first acknowledgment message or the second acknowledgment message for the stored data b is In some cases, data is being transmitted from the cache device 2 to the cache device 1. In this case, the cache device 1 (cache management unit 11) ignores these confirmation response messages sent from the cache device 2. That is, even if the cache device 1 receives these acknowledgment messages from the cache device 2, it only discards them and does not execute the processing accompanying the reception of the acknowledgment messages.

Whether to ignore the acknowledgment message is determined based on the sequence number contained in the message. For example, if the sequence number is incremented by one in order from 1 and if the cache device 1 receives two first acknowledgment messages, the sequence numbers of the two acknowledgment messages will be changed. Among them, the one with the young (small) value is the response message corresponding to the stored data b. Therefore, in this case, the first acknowledgment message with the lower sequence number is ignored.

In addition, a storage device a having a different content in the same address range as the storage data (storage device data d) which is flushed by the cache device 2 and is in the flushed msg state is transmitted from the access host to the cache device 1. , Sometimes received by cache device 1. In this case, the cache device 1 ignores the flash message from the cache device 2, and the cache device 2 converts the stored flashed data d into the copy message transmitted from the cache device 1. By replacing the stored data with the stored data a, the inconsistency of the stored data in the same address range can be resolved.

By the time the storage device 1 of the cache device 1 enters the flushing state, the storage device 1 When the cache device 1 receives new storage data (storage data e) in the same address range as the data a from the access host, the cache device 1 (cache management unit 11) stores the secondary storage data. stops writing storage de Isseki a to device 4 _k (flushing), only the stored data e by writing to the secondary storage device 4 _k, it is possible to avoid the write Mino overlap.

If the storage data a transmitted from the access host 4i to the cache device 1 is a block data consisting of a plurality of bytes, the copy message divides the block data into a plurality of parts, May be transmitted. In this case, a plurality of first acknowledgment messages are also transmitted corresponding to each part. In addition, the state of the cache control list for this block is updated after reception and writing to the cache memory 22 are completed for all blocks. In addition, a host acknowledgment message to the access host 4i is also transmitted after receiving the second acknowledgment message for all of the block data.

The flash message transmitted in step S5 can be transmitted separately from other messages, or can be transmitted by piggyback in addition to other control messages. That is, immediately after step S4, a flash message can be transmitted as an individual message separated from other messages, and then added to a copy message for another storage message and piggybacked. Can also be sent as

When the storage data b stored in the cache memory 12 and the storage data a received from the access host 4 i have the same contents, the cache device 1 sends the access data to the access host 4 丄. An acknowledgment message can be sent immediately to avoid writing (updating) to the secondary storage (and cache memories 12 and 22). As a result, the cost required for writing can be reduced.

“Overwrite collision” means that the access host group sends the cache system 3 In this case, multiple storage data with different contents are written in the same address range, and these multiple storage data are in the received state in the cache system 3. The collision includes a collision in only one cache device and a collision in both cache devices 1 and 2. The following describes the processing when a collision occurs in these two cases.

(1) Processing when a collision occurs with one cache device

After the cache device 1 receives a certain storage data A1 from the access host 4 ，, the access host 4 stores the contents of the memory A2 with different contents in the same address range as the storage data A1. When the data is received from the cache device 1, a collision between the stored data A1 and A2 occurs in the cache device 1. The access hosts 4i and 4j may be the same or different.

In this case, the cache management unit 11 of the cache device 1 overwrites the cache control list of the stored data A1 with the cache control list of the storage data A1 or the storage control A1. The cache control list of the storage device A2 is deleted or invalidated, and a cache control list of the storage device A2 is newly generated and added to the cache control table. In addition, the cache management unit 11 writes the storage data A2 to the same area as the storage data A1 or a different area in the cache memory 12.

When the cache management unit 11 receives the first acknowledgment message for the stored data A1 from the cache device 2, the cache management unit 11 ignores the first acknowledgment message. Note that the two first acknowledgment messages can be distinguished, for example, in the same manner as described above, by the large and small sequence numbers (old and young), and the first acknowledgment messages with small sequence numbers (young). The response message is ignored.

Similarly, the cache management unit 21 of the cache device 2 stores the cache control list generated based on the copy message (the first copy message) of the stored data A1 in the storage device A2. Either overwrite based on the copy message (the second copy message) or erase or invalidate the cache control list of storage data A1 and cache control of storage data A2. A new list is created and added to the cache control table. Also, the cache management unit 21 In the flash memory 22, the memory A 2 is written to the same area as the memory A 1 or a different area.

When the communication line L is, for example, a gigabit Ethernet, the second copy message transmitted later is received by the cache device 2 before the first copy message transmitted earlier. There is. Also in such a case, the cache management unit 21 ignores (discards) the first copy message based on the sequence numbers of the first copy message and the second copy message, and deletes the first copy message. 2 ■ The storage of the copy message can be stored in the cache memory 22.

Although the collision of the storage device 1 in the cache device 1 has been described, the same processing is executed when a similar collision occurs in the cache device 2.

(2) Collision detection processing when a collision occurs between both cache devices

5A to 5C are sequence diagrams showing the flow of the collision detection process when a collision occurs between the two cache devices. In these figures, storage data A1 and storage data A2 are storage data having the same address range and different contents.

In each of FIGS. 5A to 5C, the storage data A1 transmitted from the access host 4 is received by the cache device 1 and is in the receive state, and the storage data A1 transmitted from the access host 4j is also received. A2 is received by the cache device 2 and is in the receive state. Therefore, the stored data A1 and the stored data A2 are in a collision state.

Note that the reception time of the stored data A1 of the cache device 1 and the reception time of the stored data A2 of the cache device 2 may be simultaneous, or one may be earlier or later than the other. .

In this collision state, when both memory devices A 1 and A 2 shift to the dirty state or non-volatile state in both cache devices 1 and 2, the consistency and transparency of the data held by both cache devices are restored. Cannot be maintained. Therefore:-To maintain persistence and transparency, both cache units must first detect a collision condition. Therefore, the following processing is executed.

Figure 5A shows that one cache device (cache device 2) is stored in memory A2. This figure shows the flow of the collision detection process when a copy message of the storage device A1 is received from the other cache device (cache device 1) before sending the copy message.

The cache device 1 attempts to transmit the copy message by receiving the stored data A1 from the access host 4i and the cache device 2 receiving the stored data A2 from the access host 4j (Sll). , S13). However, if the copy message of the storage data A1 is received from the cache device 1 before the cache device 2 transmits the copy message, the cache device 2 (cache management unit 21) stores the data in the storage device. Evening A2 detects that a collision has occurred (S14).

That is, the cache device 2 (cache management unit 21) calculates (a) the address range (that is, the device number, the device start address, and the data length) included in the header of the copy message of the storage data A1; By comparing the address range of A3 with the address range of A2, it is detected that the address range is the same, and (b) the address of A1 is stored in the data area of the copy message. By comparing the contents with the contents of the stored data A2, it is detected that the contents of the stored data A2 are different, and (c) the stored data A2 is determined by the cache control list of the stored data A2. It detects that the storage device A1 is in the receive state based on the receive state and the status of the header of the copy message. From these (a) to (c), the cache device 2 detects the occurrence of a collision between the stored data A1 and the memory A2.

Upon detecting the collision, the cache device 2 stops transmitting the copy message of the stored data A2, and transmits a collision detection message (COLLISION) to the cache device 1 instead of the first acknowledgment message ( S14).

The cache device 1 (cache management unit 11) detects that a collision has occurred in the storage device A1 by receiving the collision detection message from the cache device 2 (S12). That is, based on the fact that the message is a collision detection message and the address range included in the header of the collision detection message, the cache device 1 (cache management unit 11) detects a collision in the storage device A1. Detect that it has occurred. When the cache devices 1 and 2 are switched, that is, when the cache device 2 transmits the copy message of the storage device A2 to the cache device 1 before the cache device 1, the cache device 1 The collision detection message is transmitted to the cache device 2.

In this way, when one cache device receives a copy message having a storage content of a different content in the same address range with respect to the storage data in the receive state, the other cache device transfers the copy data to the other cache device. By transmitting a collision detection message notifying the occurrence of a collision instead of a message, both cache devices can detect the occurrence of a collision.

Figure 5B shows the flow of the collision detection process when both cache devices send and receive a copy message.

Cache device 1 sends a copy message of stored data A 1 (S 21), cache device 2 sends a copy message of stored data A 2 (S 24), and both cache devices send When receiving the other party's copy message, each of the cache devices detects a collision by the other party's copy message (S22, S25) o

In this case as well, as in FIG. 5A, the cache device that has detected a collision transmits a collision detection message to the other party instead of the first acknowledgment message (S22, S25). In other words, in Fig. 5B, both cache devices transmit a collision detection message to the other cache device. As a result, both cache devices will again detect a collision that has already been detected.

In this way, when both cache devices transmit and receive a copy message, both cache devices can detect a collision based on the copy message. Furthermore, in this case, both cache devices can send a collision detection message so that both cache devices can mutually notify that they have detected a collision.

In FIG. 5C, as in FIG. 5B, both cache devices transmit a copy message, but the copy message of the stored data A1 is received by the cache device 2 after the collision detection message of the cache device 1. The flow of the collision detection process Is shown.

In this case, the cache device 2 detects the collision based on the collision detection message transmitted from the cache device 1 (S35), and then detects the detected collision by the received copy message of the storage device A1. Is detected again (S36). As a result, a collision is detected in both cache devices (S32, S35, S36)

Note that the cache device 2 may transmit a third acknowledgment message (COL-ACK) to the cache device 1 after detecting the detected collision, as indicated by the broken line in FIG. 5C. When the third acknowledgment message is transmitted, the cache device 1 again detects the detected collision by this message (S33). A situation opposite to that of FIG. 5C may occur. That is, the copy message of the storage device A2 arrives at the cache device 1 with a delay. In this case, the same processing is performed only by switching between cache devices 1 and 2.

(3) Collision recovery processing between both cache devices

When a collision is detected between the two cache devices described above, the following three methods are used to resolve (recover) the situation.

(a) First method

In the first method, when a collision is detected, it is determined in advance which of the cache devices 1 and 2 is to be prioritized, and the stored data received by the prioritized cache device 1 is made valid. This is a collision resolution method. In this case, only the data stored in the priority cache device is treated as valid, and the data stored in the non-priority cache device is treated as invalid.

Here, “the stored data is treated as valid” means that the cache control list of the stored data exists in the cache control table in a state other than the invalidated state, and the stored data is stored in the cache memory. That is being done.

“The stored data is treated as invalid” means that the cache control list of the stored data is deleted from the cache control table (including the case where it is overwritten by another cache control list). , Or that the status of the cache control list is invalidated. In the memory day, In some cases, it may exist in memory, and in other cases it may have been erased from the cache memory (including cases where it is overwritten by other storage devices).

For example, in FIG. 5A, when the cache device 1 is prioritized, the cache management unit 11 of the cache device 1 detects the collision (S12), but the cache control list of the storage data A1 There is no need to update the cache memory 12. That is, in the cache device 1, the stored data A1 is treated as valid, and the stored data A2 is treated as invalid.

On the other hand, after detecting a collision, the cache management unit 21 of the cache device 2 adds the cache control list of the storage data A1 to the cache control table based on the information in the header of the copy message of the storage data A1, and adds the cache memory. Write the memory data A 1 to 2 2. The cache control list of the stored data A2 is deleted or placed in the invalidated state. As a result, even in the cache device 2, the stored data A1 is treated as valid, and the stored data A2 is treated as invalid.

In FIG. 5A, when the cache device 2 is prioritized, the cache management unit 21 of the cache device 2 needs to update the cache control list and the cache memory 22 of the storage device A2 after detecting the collision. Absent. Then, the cache management unit 21 transmits a copy message of the stored data A2 to the cache device 1. This copy message may be sent separately from the collision detection message, or may be sent as a background with the collision detection message.

The cache management unit 11 of the cache device 1 adds the cache control list of the storage data A2 to the cache control table based on the information in the header of the copy message, and writes the storage data A2 to the cache memory 12. The cache control list of the stored data A1 is deleted or placed in the invalidated state. As a result, even in the cache device 1, the storage data A2 is treated as valid, and the storage data A1 is treated as invalid.

Although the cache device 1 may return the first acknowledgment message in response to the copy message from the cache device 2, it is preferable that the cache device 1 does not return the first acknowledgment message in order to reduce the communication cost of the message. If the first acknowledgment message is returned Then, the cache device 2 may ignore this first acknowledgment message. In FIGS. 5B and 5C, both cache devices 1 and 2 hold both storage data A1 and storage data A2. The stored data received by the device is treated as valid, and the stored data received by the non-priority cache device is treated as invalid.

In this way, the collision state is resolved, and in the cache devices 1 and 2, consistency and transparency of the stored data are ensured.

(b) Second method

The second method compares the reception time of the storage data of the cache device 1 with the reception time of the storage data of the cache device 2, and gives priority to the storage data of the earlier reception time. This is a conflict resolution method.

In the second method, after detecting a collision, the cache devices 1 and 2 communicate the reception time via the communication line L, or transmit the reception time by a copy message or a collision detection message. As a result, the reception time is mutually notified. Then, the storage data received at the earlier reception time is prioritized, the storage data is treated as valid, and the storage data received at the later reception time is treated as invalid. Become. When the reception time is notified by a copy message or a collision message, an area for storing the time is provided in the header or data section of these messages.

In the second method, it is assumed that the times of both cache devices 1 and 2 are synchronized. Similarly to the first method, in FIG. 5A, when the cache device 1 is prioritized, a copy message of A2 is transmitted from the cache device 2 to the cache device 1. It will be. In Fig. 5B and Fig. 5C, no matter which of cache devices 1 and 2 is prioritized, both of them hold the stored data A1 and A2, so there is no need to newly send a copy message. .

This second method also eliminates the collision state, and ensures consistency and transparency of the stored data in the cache devices 1 and 2. (c) Third method

The third method is a conflict resolution method in which both cache devices transmit a copy message (retransmission copy message) or a retransmission instruction message again after a random time has elapsed after collision detection.

The cache device that transmitted the retransmission message or the retransmission instruction message earlier has priority, and the stored data received by this cache device from the access host is treated as valid.

The random time is, for example, a time obtained based on pseudo-random numbers generated by the cache management units 11 and 21 respectively.

The message transmitted after the elapse of the random time may be a retransmitted copy message containing the stored data A1 or A2, but the stored data A.1 and A2 are transmitted to the cache device on the other side. If it already has one, it is preferable to use a retransmission instruction message to indicate retransmission without including data storage in order to reduce communication costs. In order to distinguish the resent copy message from the normal copy message, the type of the header part is a resend copy message. ":". The contents of the other header parts are the normal copy message. The retransmission instruction message has only a header part and no data part.The type of the header part of the retransmission instruction message is ": RETX", which indicates the retransmission instruction message, and the address of the header. The range is set to the same address range as the previously transmitted storage data, so that the cache device on the receiving side can identify the retransmission message for the previously received storage data.

For example, in FIG. 5A, the cache device 1 (cache management unit 11) transmits a retransmission instruction message to the cache device 2 via the communication line L after a random time has elapsed. After a random time elapses, the cache device 2 (cache management unit 21) transmits a retransmission copy message including the stored data A2 to the cache device 1 via the communication line L.

If the retransmission instruction message transmitted by the cache device 1 is transmitted before the retransmission copy message transmitted by the cache device 2, the cache device 1 takes precedence, and therefore, the storage device A1 is regarded as valid. Will be handled. On the other hand, when the retransmission copy message transmitted by the cache device 2 is transmitted before the retransmission instruction message transmitted by the cache device 1, the cache device 2 has priority, and therefore, the storage device A2 is valid. Is treated as something In this case, since the cache device 1 does not have the storage device A2, the cache device 2 sends the memory A2 to the cache device 1 by a copy message or the like.

Since the random time by the cache device 1 and the random time by the cache device 2 are the same, the retransmission instruction message transmitted by the cache device 1 and the retransmission copy message transmitted by the cache device 2 are simultaneously transmitted and received. In this case, a random time is counted again and the same process is repeated.

Similar processing is performed in the case of FIGS. 5B and 5C.

Until the random time elapses, another storage device having the same address range may be transmitted from the access host and received by the cache device 1 or 2. In this case, the cache device 1 or 2 that has received another storage device transmits the other storage device to the other cache device again by a copy message.

This third method also eliminates the collision state, and ensures consistency and transparency of the stored data in cache devices 1 and 2.

If at least one of the memory data received by the cache device 1 from the access host and the storage data received by the cache device 2 from the access host has a plurality of bytes, the storage data of a plurality of bytes is stored. A collision may occur in a part of the area. In this case, the collision detection processing and the recovery processing are executed for a part where the collision has occurred.

A description will be given of the process of reading the stored data when the access host group 4 requests the cache system 3 to read the stored data. Here, the case where the access host 4i issues a read request for storage data to the cache device 1 will be described.

The access host 4i needs a read request including the address range of the storage data to be read. The request is transmitted to the input / output unit 13 of the cache device 1 via the access network 6. This read request is given from the input / output unit 13 to the cache management unit 11.

The cache management unit 11 determines whether stored data in the address range included in the read request exists in the cache memory 12 based on the cache control table. Except for the stored data in the received state and the invalidated state, the stored data in other states is determined to exist in the cache memory 12. If all or part of the storage data corresponding to the address range is not stored in the cache memory 12, the cache management unit 11 deletes the part not stored in the cache memory 12. The data is read from the corresponding secondary storage device of the secondary storage device group 5 via the input / output unit 14 and the storage network 7 and stored in the cache memory 12. Along with this, the cache management unit 11 generates a cache control list related to the storage data stored in the cache memory 12 from the secondary storage device and adds it to the cache control table. Note that "clean" is written in the status of this cache control list.

Subsequently, the cache management unit 11 reads out the storage data corresponding to the read request from the cache memory 12 and transmits it to the access host 4 via the input / output unit 13 and the access network 6.

At this time, in this embodiment, even if the storage data read from the secondary storage device by the cache device 1 already exists in the cache memory 22 of the cache device 2, the data is read from the secondary storage device. It is guaranteed that the stored data and the stored data in the cache memory 22 have the same contents. Therefore, there is no need to communicate a confirmation message or the like between the cache devices 1 and 2 to confirm the consistency of the stored data. As a result, the communication overhead and communication cost associated with this communication can be reduced. .

If the storage data requested to be read from the access host 4 does not exist in the cache memory 12 but exists in the cache memory 22, the cache management unit 11 operates in the cache device 2 (the cache memory 2). You may receive this memory from 2). When the time order of the read and received word 31t data at the access host 4i is important, the time sequence can be changed by using a general control synchronization mechanism of the access host 4 丄. Can be guaranteed.

Load distribution processing>

In the above description, the cache device that has received the storage data from the access host writes (flashes) the storage data to the secondary storage device. However, this flash can be distributed between the two cache devices.

For example, when the access frequency of one cache device to the secondary storage device is higher than that of the other cache device, the write process can be shared by the other cache device. Also, when there is a difference between the processing capacity and performance of the two cache devices, more write processing can be performed by a cache device with higher performance and performance. As a result, load distribution between cache devices 1 and 2 is achieved.

In order to perform such load distribution, the cache management units 11 and 21 both measure the load and periodically use control messages (other control messages different from the copy message and acknowledgment message described above). Mutually inform the load. The measured load includes, for example, the number of times of writing to the secondary storage device group 5 and the amount of data written (the number of bytes and blocks). Then, the cache device with a low load executes the flush.

FIG. 6 is a sequence diagram showing a flow of a write process in a storage device when load distribution is performed.

When the cache device 1 receives the stored data from the access host 4 i, the cache management unit 11 generates a cache control list and stores the cache control list in the cache memory 12, as in step S 1 in FIG. The memory of the memory is remembered (S40).

Subsequently, the cache management unit 11 sends the copy message to the cache device 2 (S41). Here, when the load of the cache device 1 is higher than the load of the cache device 2, the cache management unit 11 specifies the cache device 2 as the flush authority of the header of the copy message.

The cache management unit 21 receives the command for which the cache device 2 is specified for the flash right. When a copy message is received, a cache control list is generated, and the state of the generated cache list is set to a dirty state instead of the non-volatile state normally set, and the storage data included in the copy message is stored. The evening is stored in cache memory 22 (S42).

Subsequently, the cache management unit 21 returns a first acknowledgment message to the cache device 1 (S8).

When the cache management unit 11 receives the first acknowledgment message, it sets the cache control list to a non-volatile state instead of the dirty state, and sends a host acknowledgment message to the access host 4i. Yes (S43).

Thereafter, the cache management unit 21 changes the dirty storage state to the flushing state at an appropriate timing, and writes the flushed state to the secondary storage unit (S44). In the subsequent steps S45 to S48, the cache management unit 11 changes the state of the cache control list from non-volatile to clean, and the cache management unit 21 changes the state of the cache control list from flushing to flushed msg. This is the same as the corresponding processing of steps S5 to S10 shown in FIG.

In this way, the load can be distributed by transferring the flash authority for storing data in the appropriate address range to the other cache device.

In the above description, the flush authority is given to the other cache device. However, the fact that the own cache device has the flash authority can be notified to the other cache device by a control message such as a copy message. .

Also, the load value can be the ratio of the load to the performance of each cache device. For example, the cache device 1 divides its own load value by its own performance value (that is, (the load value of the cache device 1) ÷ (the value of gender g of the cache device 1)), and calculates the result of this division. The data is transmitted to the cache device 2, and the cache device 2 also transmits the division result of the own device to the cache device 1. The flushing authority may be given to a cache device having a lower value of the two ratios.

As described above, the failure monitoring units 16 and 26 operate the other cache unit. Monitor the situation and determine if a failure has occurred. For example, the failure monitoring units 16 and 26 send failure detection messages via the communication line L to the other cache device at fixed time intervals. If the failure monitoring units 16 and 26 do not receive the failure detection message transmitted from the other cache device even after the lapse of a predetermined time, a failure occurs in the other cache device. Judge that it is.

The following describes recovery processing when a failure is detected in the cache device 1 and the failure monitoring unit 26 detects this failure.

When the failure of the cache device 1 is detected, the cache device 2 replaces the cache device 1 with the dirty state storage data that is to be written from the cache memory 12 to the secondary storage device group 5. Must be written to secondary storage.

For this reason, the failure monitoring unit 26, based on the cache control table held by the cache management unit 21 after the failure is detected (that is, the cache control table related to the stored data stored in the cache memory 22), Rewrite the non-volatile state of storage data to dirty state. As a result, the storage data changed to the dirty state is written (flushed) to the secondary storage device group 5 by the cache management unit 21.

Even when the cache device 1 is writing data to the secondary storage device group 5 in the middle of writing to the secondary storage device group 5, it is impossible to determine which portion of the data has been written to the secondary storage device group 5. The storage device in the flushing state in the cache device 1 is also stored in the cache memory 22 in the cache device 2 as the storage device in the non-volatile state. Therefore, the storage device in the flushing state is also changed from the non-volatile state to the dirty state in the cache device 2 in the same manner as described above, so that the cache device 2 can reliably store the data in the secondary storage device. It will be memorized.

If the cache device 1 recovers from the failure, or if the cache device 1 is switched to another backup cache device, a situation may occur in which only one of the cache devices has a dirty state storage device. obtain. Therefore, In order to prevent such a situation from occurring, the return timing of the cache device 1 is controlled so that the cache device 1 is not returned unless all cache data stored in the cache memory 22 is in the clean state. Is done. Alternatively, the storage device in a dirty state before the cache device 1 returns may be transmitted to the cache device 1 after the recovery.

When a failure occurs in the cache device 2, the failure monitoring unit 16 of the cache device 1 executes the same processing as the failure monitoring unit 26 described above. Industrial applicability

INDUSTRIAL APPLICABILITY The present invention can be used for a cache system arranged between an access host (group) such as a computer and a secondary storage device (group).

According to the present invention, since the two cache devices independently receive and process the storage data from the access host, almost double the processing capacity can be obtained for a single cache device. Therefore, the bottleneck of a single cache device can be eliminated.

Also, since the same storage data is stored in both cache devices, the storage data in the cache system can be made non-volatile. As a result, even if a failure occurs in one of the cache devices, loss of stored data is prevented.

Furthermore, since the conflict resolution processing is executed between the two cache devices, consistency and transparency of the stored data between the two cache devices are ensured. Therefore, each cache device does not need to check the consistency of the stored data each time the data is read in the evening, eliminating the overhead for checking.

As a result, it is possible to achieve consistency and transparency of stored data while improving performance.

Claims

Claims: A cache system having two cache devices, storing data provided from an access host or a secondary storage device, and recalling data from the access host in the secondary storage device. Each cache device in

A data input unit for inputting a first data provided from the access host; a data receiving unit for receiving the second data input by the other cache device from the access host and transmitted to the own cache device;

A cache storage unit that stores the first data and / or the second data,

A cache management unit that manages the cache storage unit;

A data transmission unit that transmits the first data to the other cache device; and a data output unit that outputs the first data or the second data to the secondary storage device. When,

A cache device having: In claim 1,

By the time the data transmitting unit completes the transmission of the first data, the data input unit sets the same address range as the address range of the first data on the secondary storage device. When the third host having a different content from the first host is input from the access host, the cache management unit determines that the third host is valid, and Stored in the cache storage unit, and treats the first data as invalid.

Cache device. In claims 1 or 2,

After the data input unit inputs the first data, and before the data transmission unit completes the transmission of the first data, the data reception unit receives the second data. No one In the case of receiving the evening, based on the address range on the secondary storage device of both the first data and the second data and the content of both data, A collision detection unit that determines whether there is a collision between the first data and the second data; and, when the collision detection unit detects a collision, a collision detection message indicating that a collision has occurred is transmitted to the other cache device. A collision detection message transmitting unit that transmits the collision detection message from the other cache device; and a collision detection message receiving unit that receives the collision detection message from the other cache device.

A cache device further comprising: 4. In claim 3,

Cache device. 5. In Claims 3 or 4,

The cache management unit, when the collision detection unit detects a collision, among the first data and the second data, a data which is prioritized based on a predetermined priority. Treats as valid, treats the other as invalid,

Cache device.

6. In Claims 3 or 4,

The cache management unit, when the collision detection unit detects a collision, stores the time at which the first data was input to the data input unit and the second data at the data input unit of the other cache device. Of the times entered in, the one with the earlier time is treated as valid and the one with the later time is treated as invalid.

Cache device.

7. In Claims 3 or 4,

When the collision detection unit detects a collision, after a lapse of a random time, the data transmission unit transmits the first data or a retransmission message indicating retransmission of the first data. ,

A data message receiving unit that receives the second data or the retransmitted message of the second data transmitted from the data message retransmitting unit of the other cache device;

Further has

The cache management unit treats the data corresponding to the earlier time out of the retransmission time by the data Z message retransmitter and the reception time by the data message receiver as valid. Treats a night that corresponds to a later time as invalid,

Cache device. 8. In claim 7,

If the retransmission time of the data / message retransmitting unit is the same as the reception time of the data / message receiving unit, the data retransmission unit retransmits the data again after a random time elapses. Repeat,

Cache device.

9. In any one of claims 1 to 8,

After the data output unit outputs the first data or the second data to the secondary storage device, a flash message indicating the completion of the output is stored in the other cache. A flash message transmitting unit for transmitting to the device; a flash message receiving unit for receiving a flash message transmitted from the other cache device;

A cache device further comprising:

10. In claim 9, The cache management unit, after transmitting the flash message by the flash message transmitting unit, or after receiving the flash message by the flash message receiving unit, the first data or the flash message corresponding to the flash message. Treat the second day as invalid,

Cache device.

1 1. In any one of claims 1 to 10,

The data transmission unit, together with the first data, outputs the first data to the secondary storage device, and outputs first flash authority information indicating whether or not the shift cache device performs the output. Send,

The data receiving unit receives, together with the second data, second flash authority information indicating which cache device outputs the second data to the secondary storage device,

The data output unit outputs the first data to the secondary storage device when the first flash right information indicates the own cache device, and outputs the second flash right. Outputting the second data to the secondary storage device if the information indicates the own cache device;

Cache device. 1 2. In claim 11,

A load information transmitting unit for measuring the load of the own cache device and transmitting the measured load value to the other cache device;

A load information receiving unit that receives a load value of the other cache device transmitted from the other cache device;

The load value of the own cache device is compared with the load value of the other cache device, and the cache device having the smaller load value is stored in the first storage device or the second storage device. A cache device further comprising: a flash right setting unit for setting a flash right.

1 3. In claim 12,

The cache device, wherein the value of the load is a ratio of the load to the performance of each cache device. 14. In any one of claims 1 to 13,

Monitoring the occurrence of a failure in the other cache device, and detecting the occurrence of the failure, among the first data stored in the cache storage unit, the data in the data output unit or the data in the other cache device. The evening output unit has not completed the output to the secondary storage device, and the data transmission unit has completed the transmission to the other cache device, and has been stored in the cache storage unit. Of the second data, the data output unit or the data output unit of the other cache device that has not completed the output to the secondary storage device, A fault monitoring unit that controls output to the secondary storage device;

Cache device.

15. In any one of claims 1 to 14,

A read request input unit for receiving a data read request including the address range of the secondary storage device from the access host;

When the data corresponding to the address range included in the read request is stored as a valid data in the cache storage unit, the data is read from the cache storage unit and the cache is read. If it does not exist in the storage unit, or if it is not stored as valid data, a data read unit that reads from the secondary storage device;

A read data transmitting unit for transmitting the data read by the data reading unit to the access host;

A cache device further comprising:

16. In claim 15,

The data management unit reads the data from the secondary storage device by the data reading unit. A cache device that stores the extracted data as a valid data in the cache storage unit. 7. It has two cache devices, remembers the data given from the access host or the secondary storage device, and remembers the data from the access host.

A cache system for storing data in a secondary storage device,

Each of the two cache devices is

A data input unit for inputting a first data provided from the access host; and a data receiving unit for receiving the second data transmitted from the access host to the other cache device and transmitted to the own cache device. One night receiver,

A cache storage unit for storing one or both of the first data and the second data;

A cache management unit that manages the cache storage unit;

A data transmitting unit that transmits the first data to the other cache device; and a data output unit that outputs the first data or the second data to the secondary storage device. Department and

A cache system having 8. It has two cache devices, stores data given from an access host or a secondary storage device, and stores data from the access host in the

A cache method in a cache system for storing data in a secondary storage device,

One cache device that has received the data from the access host stores the data in its own cache memory and transmits the data to the other cache device,

The other cache device receives the data transmitted from the one cache device and stores the data in its own cache memory.

The one cache device or the other cache device outputs the data to the secondary storage device; Cache method.