JPH11345096A - Disk array system - Google Patents

Abstract

(57) [Summary] [PROBLEMS] In a disk array system connected to a disk device capable of command queuing, the disk occupation time in parity generation processing in the disk device is reduced, and the throughput of the entire disk array system is improved. . When a parity generation request in a disk device queued in a disk device is processed, the process is temporarily interrupted at the stage of reading an old parity, and subsequent generation and writing of a new parity are performed by another read / write request. The average time required for positioning the head of the new parity write in the parity generation processing in the disk device is reduced by performing the scheduling in the same manner as the above.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-performance disk device suitable for a disk array, and a disk array system comprising a high-performance disk device suitable for a disk array and a storage controller.

[0002]

2. Description of the Related Art Regarding disk arrays, D. Patterso
n, G.gibson, and RHKartz; A Case for Redundant Arr
ays of Inexpensive Disks (RAID), ACM SIGMOD Confer
ence, Chicago, IL, (June 1988), pp.109-116 is known.

A paper by Patterson discloses a technique relating to a data arrangement method in a disk array. In a disk array system, high performance is realized by physically appearing a plurality of disk devices as one logical disk device with respect to a data processing device. Further, when a failure occurs in a disk device storing data, redundant data enabling recovery of data stored in the failed disk device can be restored to a disk different from the disk device storing the data. High availability has been achieved by storing it in the device.

[0004] Usually, a read / write unit of data stored in a disk device is called a record. However, in the case of a disk array, a record which is a read / write unit viewed from the data processing device may have a different data length from a record actually recorded on the disk device. Hereinafter, the former is called a logical record and the latter is called a physical record.
According to Patterson's paper, some logical records are arranged in five levels of R1 to RAID5.
It has been introduced as an AID configuration, and results of studies on performance and reliability have been reported. Among them, the RAID 5 disk array stores each logical record as one physical record in the disk device, and performs exclusive-OR operation on m (m ≧ 1) logical records stored in different disk devices. Redundancy data called a certain parity is created and stored in another disk device as one physical record.

Hereinafter, a physical record storing a logical record to be directly read / written by the processing device is called a data record, and a physical record storing a parity is called a parity record. A group of records including m data records and parity records is called a parity group.

A RAID 5 disk array stores a physical record of old data and old parity, which are pre-update values of a data record and a parity record at the time of parity generation, for an updated data record. , A new parity which is an updated value of the parity record is generated by taking an exclusive OR with the new data which is an updated value of the data record, and the new parity needs to be written to the disk device. For this reason, the overhead of the storage control device accompanying the data update increases.

As a method of reducing the overhead of the storage control device due to the parity generation processing, information necessary for parity generation is transferred to a disk device storing a parity record, a new parity is generated on the disk device side, and the parity is generated on a disk medium. A writing technique is conventionally known. In this technique, new data and old data necessary for generating parity of certain update data are transferred to a disk device that stores a parity record. In the disk device, the head is positioned at the corresponding parity record, and the old parity is read out to a buffer provided inside the disk device.
The new data is created before the head rotates to the parity record again by rotating the disk medium, and the new parity is written to the disk medium when the head positioning is completed.

In the transaction processing, usually, the read data is updated, so this technique can be applied to most of the parity generation for the data update.
The load on the storage control device due to the parity generation processing can be greatly reduced.

On the other hand, as a measure for improving the performance of a disk array system, a technique for reducing the occupancy of a disk device which is a component of the disk array has been proposed. In a disk device, in order to position a head for reading and writing data on a disk medium to a target physical record, a seek operation of positioning the head on a track to which the target record belongs, and a seek operation until the head is positioned on the target record in the track. It is necessary to wait for rotation. In the case of random read / write that occurs in transaction processing, etc., the time required for these seeks / rotation waits occupies most of the disk occupation time, but involves physical operations, and is slower than a processor or bus. It tends to be a bottleneck in system performance.

As one method of reducing the disk occupation time, there is a method of simultaneously transmitting a plurality of read / write requests to a disk device and queuing the request in the disk device. In the processing of a certain read / write request, by receiving another read / write request while positioning the head, the use efficiency of the transfer bus and the execution efficiency of the read / write request can be improved. Since the other processes are performed in parallel, the apparent disk device occupation time can be reduced. In addition, since a plurality of queued read / write requests can be scheduled in the disk device, scheduling considering the time required for seek and rotation wait can be performed, and the average head positioning time can be reduced. It is.

This method of queuing read / write requests is implemented as a tagged queuing function in the SCSI standard. For more information about the tagged queuing function, see "SCSI-2 Detailed Explanation" (Seiichi Sugaya, CQ
Publishing, pp. 112-119) and so on. In the tagged queuing function, a plurality of different hosts simultaneously queue a plurality of read / write requests for a single disk device, and the disk device determines the execution order of the queued read / write requests. It is possible to perform processing by rearranging on the basis of a head positioning time or the like.

[0012]

When the in-disk parity generation technique is applied to a disk array system,
The load on the storage control device due to parity generation can be reduced. However, on the other hand, after reading the old parity, the disk medium rotates and waits until the head is positioned again at the same parity record, and then writes the new parity, so that the occupation time of the disk device increases. Will be.

Therefore, when the performance bottleneck of the disk array system is the storage controller, the system performance can be improved by applying the in-disk parity generation technique. In some cases, the system performance may be degraded.

Therefore, it is considered that the command queuing technique is applied simultaneously with the parity generation technique in the disk device. In this case, a normal read / write is
In addition to the write request, a parity generation request in the disk device is also queued.

However, as in the case of the read / write request, even if a schedule considering the head positioning time is applied to the in-disk parity generation request, the first old parity of the in-disk parity generation request is read. Can be reduced, but the rotation waiting time for one rotation of the disk from the old parity read to the new parity write still remains.

An object of the present invention is to increase the throughput of the entire disk array system by reducing the average head positioning time of a new parity write in the processing of a queued disk device parity generation request command.

[0017]

According to the present invention, there is provided a disk array system which realizes the following storage control method in order to achieve the object of increasing the throughput of the disk array system.

In response to an update to a certain data record from the data processing device, the storage control device transfers the parity generation request in the disk device together with the new data and the old data to the disk device storing the corresponding parity record. On the other hand, in the disk device, the queued read / write request and the parity generation request in the disk device are treated equally, and the scheduling for reducing the average head positioning time is performed. If the processing target is a parity generation request within the disk device due to scheduling, the disk device first positions the head at the corresponding parity record and reads the old parity.

When the reading is completed, a new parity is generated, the writing of the new parity and other read / write requests are treated in the same manner, and the one with the shortest head positioning time is scheduled. If there is a read / write request with a shorter head positioning time than the new parity write, the new parity write is queued together with other requests, and the read / write request with the shorter head positioning time is processed first.

As a result, the average head positioning time of the new parity write in the parity generation processing in the disk device can be reduced.

By applying the above contents, it is possible to reduce the average disk occupation time and increase the throughput of the entire disk array system.

[0022]

Embodiments of the present invention will be described below. Note that the present invention is not limited by this.

In the embodiment of the present invention, when processing the parity generation request in the disk device queued in the disk device, the reading of the old parity and the writing of the new parity are treated completely independently. The write is scheduled in the same manner as other read / write requests.

An embodiment will be described with reference to FIGS.

FIG. 1 is a block diagram of an information processing system to which the present invention is applied. The information processing system includes a data processing device 100, a storage control device 104, and one or more disk devices 105 connected to each other. ing.

The data processing device 100 includes a CPU 101
, A main memory 102, and a channel 103.

The storage control device 104 has one or more directors 106 and a cache memory 107.

The director 106 performs data transfer between the channel 103 and the disk device 105, data transfer between the channel 103 and the cache memory 107, and data transfer between the cache memory 107 and the disk device 105.

The cache memory 107 temporarily stores read data from the disk device 105 and write data to the disk device 105. The writing of the stored write data to the disk device 105 is executed by the director 106. As a method of managing the data held in the cache, a conventionally known method can be used.

The disk device 105 includes a disk 110 for recording data and a buffer 1 for temporarily storing data.
09, buffer management information 111, and tag queue management information 1
12 and a device for reading and writing recorded data. A device that reads and writes data manages queuing of a plurality of commands and schedules queued commands. The buffer 113 has a capacity capable of storing a plurality of data, and the buffer management information 111 is used for managing the plurality of data.

FIG. 2 shows the configuration of the buffer 109 in the disk device in this embodiment. The buffer 109 includes a plurality of buffer segments 200. One buffer segment 200 can store data for one data track. At the time of read processing, the value read from the disk 110 is temporarily stored, and at the time of write processing or parity generation processing within the disk device, the write value transferred from the storage controller 104 is stored.

FIG. 3 shows the structure of the buffer management information 111. The buffer management information 111 includes buffer segment management information 300, valid buffer segment management information queue pointer 301, and free buffer segment management information queue pointer 302. The buffer segment management information 300 exists as many as the number of the buffer segments 200 and corresponds to the buffer segments 200 on a one-to-one basis.

FIG. 4 shows buffer segment management information 30.
0 indicates information related to the present invention. The buffer segment pointer 400 points to the buffer segment 200 corresponding to the buffer segment management information on a one-to-one basis. The related tag number 401 stores the tag number 500 attached to the request command transferred from the storage control device 104 as a processing request for the data track. The valid flag 402 indicates that valid data is stored in the buffer segment 200 pointed to by the buffer segment pointer 400.

The queue pointer 403 is used for connection to a valid buffer segment management information queue or an empty buffer segment management information queue, and connects the buffer segment management information 300 with each other. The group buffer segment pointer 404 connects the plurality of buffer segment management information 300 allocated to store the new data and old data of the target data record of the parity generation request command in the disk device, and the old parity and the new parity. Used for

The valid buffer segment management information queue pointer 301 and free buffer segment management information queue pointer 302 in FIG. 3 are the root pointers of the respective queues. The valid buffer segment management information queue connects buffer segment management information 111 holding valid data to the corresponding buffer segment 200. FIG. 10 shows the state of the valid buffer segment management information queue.

Buffer segment management information 3 in queue
00 are connected by a queue pointer 403. A plurality of buffer segment management information 300 assigned to one disk device parity generation request command constitutes one group, and the representative buffer segment management information 300 in the group uses the queue pointer 403 to generate an effective buffer. Connected to the segment management information queue. The empty buffer segment management information queue connects the buffer segment management information 111 corresponding to the buffer segment 200 in which the internal data is not valid.

FIG. 5 shows information related to the present invention in the tag queue management information 112. The tag number 500 stores the tag number assigned to the request command transferred from the storage control device 104. The tag number is determined in each director 106 so as not to be duplicated in all request commands transmitted from the director 106,
It is possible to uniquely determine which data record the processing request is based on the tag number. Note that a conventionally known method can be used to determine the tag number. Command information 50
1 stores the request command transmitted from the storage control device 104. The target buffer segment management information pointer 502 points to the segment management information 300 allocated to the request command. In the case of the parity generation request command in the disk device, it points to one of the segment groups storing the data to be processed and the parity.

Next, the disk drive 10 according to the embodiment of the present invention will be described.
Operation 5 will be described. The operation of the storage control device 104, such as request command transmission to the disk device 105 and cache control, can be performed by a conventionally known method, and a description thereof will be omitted.

FIGS. 6, 7, and 11 are diagrams showing the operation of the disk device 105 in this embodiment in detail.
FIG. 11 shows an operation when the processing request command transmitted from the storage control device 104 is received and queued. FIG. 6 shows the operation of the parity generation processing in the disk device at the time of the old parity read, and FIG. 7 shows the operation of the parity generation processing in the disk at the time of the new parity write.

First, in the command receiving process 610, the disk device 105 queues the processing request command transmitted from the storage control device 104. If the processing request command is a write or a parity generation request in the disk device, a segment is allocated and the target data is stored. Next, in a command execution process 620, a plurality of accumulated processing request commands are scheduled and executed so as to minimize the average seek / rotation wait time. When the command to be processed is a parity generation request in the disk device, when the old parity is read into the buffer, the processing of the command is interrupted and the processing type is changed to “new parity write”. Thereafter, the command is scheduled in the same manner as other request commands, and a new parity is generated and written.

In this embodiment, the new parity generation process in the parity generation process in the disk device is performed continuously with the new parity write process. However, the new parity generation process may be performed continuously after the old parity read process. .

The command receiving process 610 and the command executing process 620 will be described below in order.

First, the command receiving process 610 will be described.

From the storage control device 104 to the disk device 10
5 is composed of a processing request command and data to be processed. The processing request command has the following contents.

Request processing type: Request processing transmitted from the storage controller 104 includes "read", "write", "parity generation in disk device", and the like. Further, when the old parity read for generating the parity in the disk device is completed, the request processing type is changed to “new parity write” in the disk device.

Processing target range: request processing target range in the disk device.

Tag number: An identification number uniquely determined so that there is no duplication in each request command transmitted from the storage control device 104.

The processing request command transmitted by the storage controller 104 is received by the disk device 105, the processing request command is queued, and the target data is stored in the buffer 10.
9.

FIG. 8 is a flowchart of the command receiving process.

In step 801, the storage controller 104
Queues the processing request command sent from.
Specifically, the tag information 500 and the command information 501 of the request command are stored in the tag queue management information 112. The target buffer segment management information pointer 502 is initialized to an initial value.

In step 802, the processing type of the received request command is determined. If the write command or the parity generation in the disk device is performed, the process proceeds to step 803, and if the command is read, the process ends.

In step 803, the buffer segment 200 for storing the target data of the request command
Assign. Specifically, if the request command is a write command, one buffer segment 200 is allocated for new and old data in the case of generating parity in the disk device. In the latter case, the buffer segments to be allocated are connected by the group buffer segment pointer 404. The buffer segment 200 to be allocated is selected from those in which the corresponding buffer segment management information 300 is connected to the empty buffer segment management information queue, and reconnected to the valid buffer segment management information queue. At this time, the buffer segment management information 30
The tag number of the request command is set in the related tag number 401 in 0, and the valid flag 402 is made valid. Further, the head address of the buffer segment management information 300 is set in the target buffer segment management information pointer 502 in the tag queue management information 112 of the command.

At step 804, the storage controller 104
, Receives the target data of the request command, and stores it in the buffer segment 200 allocated in step 803.

Second, the command execution processing 620 will be described.

The command request processing 610 schedules the processing request commands queued in the disk device 105, and executes the request processing of the selected command.

FIG. 9 is a processing flowchart of the command execution processing.

In step 901, a processing target is selected from the queued commands. In the present embodiment, in order to reduce the average seek / rotation wait time, the processing request command that minimizes the seek / rotation wait time from the current head position to the processing target range is given priority. Alternatively, the selection may be performed in consideration of the priority order of the higher designation. A conventionally known method can be used for the schedule, and a description thereof will be omitted.

In step 902, the request processing type of the command determined to be processed in step 901 is determined.
If the request process is a read, the process proceeds to a step 903. If the request process is a write, the process jumps to a step 906.

In step 903, a buffer segment 200 for temporarily storing target data read out from the disk 110 is allocated.

In step 904, data in the processing range is read out from the disk 110 and stored in the buffer segment 200 allocated in step 903. At this time, the valid flag 402 in the buffer segment management information 300 is made valid, indicating that the held data is valid.

At step 905, the target data stored in the buffer segment 200 is transferred to the storage controller 104.

In step 906, the target data of the write request command is written in the processing target range of the disk 100. The target data of the command is specified using the target buffer segment management information pointer 502 in the tag queue management information 112.

In step 907, a buffer segment 200 for storing the read old parity is allocated. The buffer segment management information 300 of the allocated buffer segment 200 is connected to the buffer segment management information 300 already allocated to the command using the group buffer segment pointer 404 in order to store new and old data.

In step 908, the old parity specified in the processing target range is read out from the disk 110, and the buffer segment 200 newly allocated in step 907 is read out.
To be stored. At this time, the valid flag 402 in the buffer segment management information 300 is made valid.

In step 909, the command information 50 of the tag queue management information 112 corresponding to the request command
The request processing type in 1 is “Parity generation in disk device”
To “new parity write”.

In step 910, a buffer segment 200 for storing the created new parity is allocated.

In step 911, a new parity is calculated from the old and new data and the old parity held in the buffer 109, and stored in the buffer segment 200 allocated in step 910. At the same time, the valid flag of the buffer segment management information 300 is made valid.

At step 912, the new parity created at step 911 is written to the disk 110.

At step 913, the buffer segment 200 allocated to the request processing command is released. Specifically, each value of the buffer segment management information 300 of the buffer segment 200 is returned to the initial value,
Transition from the valid buffer segment management information queue to the empty buffer segment management information queue. Also, the information in the tag queue management information 112 of the request processing command is cleared.

At step 914, the completion of the request processing is reported to the storage controller 104.

In the present embodiment, the new parity is generated when the new parity write process of the parity generation process in the disk device is scheduled. However, the new parity may be generated following the old parity read. In this method, it is possible to release the buffer segments 200 other than those storing the new parity and wait for the new parity write processing.

Further, after the old parity read in the parity generation processing in the disk device, the idle time of the disk device 105 is used until the new parity write process is scheduled, for example, waiting for the head position of another processing request. Alternatively, a new parity generation process may be executed.

[0073]

According to the storage control method and the disk array system of the present invention, the old parity read and the new parity write of the parity generation processing in the disk device can be executed independently, and the average head positioning time of the new parity write is reduced. By doing so, the disk device occupancy can be reduced, and the throughput of the entire disk array system can be increased.

[Brief description of the drawings]

FIG. 1 is a block diagram of an information processing system targeted by the present invention.

FIG. 2 is a diagram illustrating a configuration example of a buffer memory;

FIG. 3 is a diagram illustrating a configuration example of buffer management information;

FIG. 4 is a diagram illustrating a configuration example of buffer segment management information;

FIG. 5 is a diagram illustrating a configuration example of tag queue management information.

FIG. 6 is a block diagram showing an operation of the disk device when a processing request command transmitted from the storage control device 104 is received.

FIG. 7 is a block diagram showing the operation of the disk device when executing the old parity read in the parity generation processing in the disk device.

FIG. 8 is a process flowchart of a command receiving process.

FIG. 9 is a process flowchart of a command execution process.

FIG. 10 is a diagram illustrating a configuration example of a valid buffer segment management information queue.

FIG. 11 is a block diagram illustrating an operation of the disk device when a new parity is generated and a write is performed in the parity generation processing in the disk device.

[Explanation of symbols]

100: data processing device, 101: CPU,
102: Main memory, 103: Channel, 104: Storage controller, 105: Disk device, 106: Director, 107: Cache memory, 109: Buffer, 110: Disk, 111: Buffer segment management information, 112: Tag queue management information, 200
... buffer segment.

Continued on the front page (72) Inventor Takao Sato 2880 Kozu, Odawara-shi, Kanagawa Prefecture Storage Systems Division, Hitachi, Ltd.

Claims (5)

[Claims] (1) The number of each parity group is m (m ≧
1) one or more pieces of data belonging to at least one or more of the parity groups, or a storage medium that stores one or more pieces of the redundant data, the data being composed of the data of 1) and redundant data for recovering the data. A disk comprising one or more disk devices, a data processing device, and a storage control device interposed between the data processing device and the storage device and controlling data transfer between the data processing device and the disk device An array system, wherein the disk device is a processing target based on processing request information received from the storage control device and a unit for storing a plurality of the data targeted by the processing request and the processing request information stored therein. Means for selecting and executing the data, and wherein the storage control device updates the data updated from the data processing device. For the first disk device storing the corresponding redundant data, an updated value of the redundant data is created in the first disk device together with an updated value of the data and a pre-update value, and the storage medium Means for transmitting first processing request information requesting that the first data is written to the storage medium from the storage medium with the pre-update value of the redundant data for the first processing request information. In a disk array system having means for reading and creating an update value of the redundant data and writing the updated value to the storage medium, the processing request information selected as a processing target by the disk device was the first processing request information In this case, the pre-update value of the redundant data, which is the object of the first processing request information, is read from the storage medium, and the first processing request information is stored in the redundant data. Means for changing the value to second processing request information requesting that the value be created and written to the storage medium and holding the second processing request information; and, if the selected processing request information is the second processing request information, Means for generating an update value of the redundant data, which is a target of the processing request information of No. 2, from the update value of the data, the pre-update value, and the pre-update value of the redundant data, and writing the updated value to the storage medium. Disk array system. 2. The disk array system according to claim 1, wherein the processing request information selected as a processing target by the disk device is the first processing request information.
The pre-update value of the redundant data that is the object of the first processing request information is read from the storage medium, and the update value of the redundant data is calculated from the update value of the data, the pre-update value, and the pre-update value of the redundant data. Means for generating and changing the first processing request information to the second processing request information and holding the processing information; and when the processing request information selected as a processing target is the second processing request information, A disk array system comprising means for writing an updated value of the redundant data, which is a target of the second processing request information, to the storage medium. 3. The storage device according to claim 1, wherein the disk device selects the storage target to be processed from the plurality of processing request information held by the disk device as a criterion. A disk array system using a time required to position a head to an area of a medium. 4. The disk array system according to claim 3, further comprising a time when said processing request information is held in said disk device. 5. The disk array system according to claim 4, wherein a priority specified by said storage controller is used for said processing request information in addition to the selection criterion of claim 4.