JPH0330012A - Disk device - Google Patents

Abstract

PURPOSE:To guarantee write data by providing a write buffer memory as a non-volatile memory on a head disk assembly HDA unit. CONSTITUTION:When read-out becomes possible in a magnetic disk 1 and a head 2, by the switch 3 of the head 2 and a read-out/write R/W circuit 4, the head 2 and the circuit 4 are selected, and in the case a reconnection to a CPU is possible, a through operation path 8 is selected by a buffer memory switch 5. However, in the case the reconnection is impossible, a read-out buffer memory 7 is selected, data is stored from the disk 1 through the head 2 and the circuit 4 and transferred to the CPU as soon as the reconnection becomes possible. At the time of write, a write buffer memory 6 is selected by a buffer memory and path switch 9 and the data is stored and the end of a write operation is reported to a controller. In such a manner, write data against a fault can be guaranteed, an overhead is reduced, and the data control can be simplified.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to a disk device having a data transfer device equipped with a buffer memory. [Conventional technology]

A conventional example of a data transfer device in which a buffer memory was attached to a head disk assembly (HDA) was published in 1983.
3455. In general, disk subsystems use read/write (R
/W) When a request is issued by the controller and accepted by the storage device, it disconnects from the upper layer and seeks.
It waits for rotation and tries to connect with the host again when R/W becomes possible. At this time. If the path connecting to the upper level is being used for other R/W processing, reconnection is not possible, and reconnection is attempted after waiting one more rotation. In the conventional example described above, when reconnection is not possible, the data is temporarily stored in a buffer memory within the data transfer device at the time of reading, and as soon as reconnection becomes possible, the data is transferred to the upper level. Also, when writing, data is first stored in the buffer memory from the upper level, then sought, waits for rotation, and as soon as it becomes possible to write to the magnetic disk, the data is transferred from the buffer memory. By doing so, if the bus with the host is used by another process during reading/writing and reconnection cannot be performed, the writing process can be performed quickly without waiting for one revolution.[Problem to be Solved by the Invention] Sometimes, the process does not end until the data is written to the magnetic disk, and you have to wait for seek and rotation wait times. The time required for such mechanical operation is R/W.
Very large in processing. Therefore, when the number of R/W requests increases significantly, the throughput decreases due to this waiting time. Therefore, a method has been considered in which the write data is stored in a buffer memory and the writing ends at that point. However, if a failure such as a power outage occurs before writing from the buffer memory to the magnetic disk, there is a risk of data loss.

[Means to solve the problem]

The configuration of the present invention is that when a write request is accepted during writing, the data is stored in a buffer memory, and at that point, the upper side is notified that the processing has been completed. In this way, write data is stored in the puffer memory at QL, and as soon as it becomes possible to write, it is written to the magnetic disk. At this time, read data is stored in a volatile memory, and write data that requires data preservation is stored separately in a nonvolatile memory.

[Effect]

In the present invention, read data and write data are stored in separate memories, the write data ends when it is written to the buffer memory, and is written to the magnetic disk as soon as it becomes possible to write later. From a higher level perspective, the waiting time required for mechanical operations is eliminated, and data management becomes easier. At this time, by using a non-volatile memory to store the write data, the data is guaranteed even if a failure such as a power outage occurs.

【Example】

Examples of the present invention will be explained below using figures. (Embodiment 1) The configuration of this embodiment includes a magnetic disk 1, a head 2, an R/W circuit 4, a read buffer 7, a write buffer. Bunkyu Memory 6 is attached. Generally, an R/W request issued from a controller is accepted by the drive, the drive disconnects from the CPU, waits for seek rotation, and when R/W becomes possible, attempts to reconnect with the CPU. Here, if connection is possible, data transfer is performed. However, if all paths are being used by other processes, reconnection is not possible. Therefore, in such a case, the data is temporarily stored in the buffer memory and transferred as soon as reconnection becomes possible. Such data management methods are publicly known. In this embodiment, the operation shown in FIG. 2 is performed. That is, during reading, the drive accepts a read request issued by the controller, and when the magnetic disk 1 and head 2 become ready for reading, the head and R/W circuit switch 3
selects head 2 and R/W circuit 4, and if reconnection with the CPU is possible, selects path 8 for through operation (transferring without storing in buffer memory) by R/W circuit and buffer memory switch 5. select. However, if reconnection is not possible, the read buffer memory 7 is selected, data is stored from the magnetic disk 1 through the head 2 and the R/W circuit 4, and as soon as reconnection is possible, the data is transferred to the CPU. Furthermore, during writing, the buffer memory and bus switch 9 selects the write buffer memory 6 to store data, and at this stage the end of the write operation is reported to the controller. FIG. 3 shows the configuration of the data transfer device. At the time of reading, the data management unit 10 checks the path with the controller, and if reconnection is possible, selects the path 8 for through operation to the switch 5, supports the switch 9 to select an empty path, and transfers the data. . However, if reconnection is impossible, the data management unit 10 instructs the switch 5 to select a path to the read buffer memory 7 and transfers the data to the read buffer memory 7. Next, the data management unit 10 constantly monitors the path to the CPU, selects an empty path as soon as reconnection becomes possible, and performs data management, recognizing that the write request is issued from the CPU. The data is recognized by the unit 10, and the switch 9 immediately selects the path connected to the write buffer memory 6 to transfer the data. Seek and wait for rotation on the magnetic disk 1 and head 2, and as soon as writing becomes possible, R/W is performed with the switch 5.
Select a circuit and write data from the write buffer memory 6. Note that during writing, when data is stored in the write buffer memory 6 in this manner, the CPU is notified that the writing has been completed. Therefore, if an accident such as a power outage occurs before writing to the magnetic disk 1, the data in the buffer memory will be lost. Therefore, the write buffer memory 6 is
By using a non-volatile memory such as RAM, written data is protected in the event of a failure. When the number of read requests increases significantly compared to the number of write requests, the data management unit 10
It is also possible to store read data in the memory, making it possible to use memory with a very high degree of freedom. (Embodiment 2) An embodiment will be described in which a disk device with a data transfer device as shown in the second construction drawing is provided with a function of pre-reading data at the time of reading. As shown in FIG. 4, the read buffer memory 4 is divided into a normal online buffer memory 11 and a pre-read buffer memory 12. For sequential files, read requests are issued one after another to the same cylinder. Therefore, as shown in FIG. 5, even if R/W request A from the CPU is processed and data transfer is completed, the next R/W request B
The data is continuously stored in the prefetch buffer memory 12 under the control of the data management unit 10 until the prefetching buffer memory 12 arrives. In addition,
Regarding the stored data, a management table of the stored data is created in the data management section 10. Data for the next read request is transferred from the buffer memory 12 sofa memory 12. In such a case, when a write request is issued from the higher level and write data is written to the write buffer memory 6, the table in the data management section is referred to, and only if the same data is in the read-ahead buffer memory 12, access the read-ahead buffer memory 12;
The data in it will also be rewritten at the same time to unify the data. In the above case, in order to simplify data management, it is also possible to cancel the data in the pre-read buffer memory 2. In this embodiment as well, it is of course possible to use the write buffer memory 6 as a nonvolatile memory and to use the write buffer memory for reading. A disk device equipped with a transfer device as in the above embodiments 1 and 2 includes a magnetic disk device, an optical disk device,
Data is stored on a rotating storage medium, such as a floppy disk device. It is clear that any form of reading and writing using the head is possible.

【Effect of the invention】

As explained above, in the present invention, the following effects can be obtained by separating the buffer memories used for reading and writing, and installing the nonvolatile memory for writing in each HDA. (1) When writing write data, since the controller manages many drives, the amount of data handled increases, and the overhead of processing in the buffer memory becomes large. However, since the buffer memory according to the present invention is handled in units of HDA, the amount of data handled is relatively small, so the overhead is small. (2) If the buffer memory in the controller is used, the bus between the controller and the drive must be occupied during that time, and if an R/W request is issued from the CPU,
There is a risk that the throughput at that time will decrease. but,
By providing a buffer memory in the ODA as in the present invention, the bus with the controller is not occupied.
Writing becomes possible, and throughput is not affected by R/W requests to other drives. (3) In the buffer memory in the controller, reading and writing are stored in the same volatile memory, and only write data is simultaneously written in the nonvolatile memory. However, in the present invention, the memory is separated for reading and writing, and by using non-volatile memory for the written data, data management becomes easier, and the written data can be guaranteed against failures such as power outages. be. (4) When writing data in the controller, data is written to both the normally used buffer memory and the nonvolatile memory in case of a failure. However, in the present invention, data control is simple because it is only necessary to access the read buffer memory and rewrite the data when a write request is received for the same data as the pre-read data.

[Brief explanation of drawings]

Fig. 1 is a model diagram showing the configuration of a disk device according to an embodiment of the present invention, Fig. 2 is an explanatory diagram of the overall operation in the configuration of Fig. 1, and Fig. 3 is a data transfer in the configuration of Fig. 1. FIG. 4 is a block diagram showing the internal structure of the device, FIG. 4 is a block diagram showing the internal structure of the read buffer memory, and FIG. 5 is an explanatory diagram of the pre-read operation. Explanation of symbols 1...Magnetic disk, 2...Magnetic head, 3...
Head and R/W circuit switch, 4... Read/write circuit, 5... R/W circuit and buffer memory switch, 6... Buffer memory for writing, 7... Buffer memory for reading, 8.・Through operation path, 9...
- Buffer memory and path switch, 10... Data management section, 11... Online buffer memory, 12
...Buffer memory for read ahead! Ishibako l-input de"4" Fig. 1 7 C data 71 method hJ high speed knob Fig. 2 Fig. 3 R/W circuit controller/θ111?-9g controller ■

Claims (1)

[Claims] 1. In response to an input/output request from a host (CPU), in a disk device that stores or will store the data, the disk device responds to a read request from the host at the required timing. When transferring the input/output data, if connection with the control device is not possible, the data is temporarily stored in the buffer memory in the data transfer device attached to each disk device, and then transferred to the buffer memory at a predetermined timing. In response to a write request, a data transfer device that has the function of storing the write data in a buffer memory and writing it out when a predetermined timing arrives, performs the following operations at the time of read and write.
A disk device characterized by using separate buffer memories. 2. The disk device according to claim 1, wherein in the data transfer device, at least a buffer memory for writing is made non-volatile. 3. The disk device according to claim 1, wherein in the data transfer device, when the number of read requests increases significantly compared to the number of write requests, the buffer memory for writing can be used for reading. . 4. In the data transfer device, by separating the read and write buffer memories, a means is provided for accessing the read buffer memory only when there is data in the read buffer memory when writing. The disk device according to claim 1.