JPH09160729A - Disk array device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten the seek time of the head of each disk drive and speed up the data transfer process by rearranging plural input/output instructions which are received from outside so that the frequency of track movement of the head of each disk drive becomes smaller. SOLUTION: The I/O instructions received by an array controller 1 from a host computer are stored in a storage device 2. A scheduling device 3A takes I/O instructions properly out of the storage device 2 and changes the execution order of the I/O instructions (scheduling) so that the total seek time becomes shorter. The scheduled I/O instruction sequence is stored in the storage device 2 and the array controller 1 takes out the scheduled I/O instructions from the storage device 2 and executes them in order. Here, scheduling is performed even on each disk drive 10 in addition to load decentralization to the respective disk drives 10. Consequently, the seek time of each disk drive 10 is shortened and the performance is improved.

Description

Detailed Description of the Invention

(Technical Field of the Invention) Technical Field of the Invention Conventional Technology (FIGS. 15 to 17) Problems to be Solved by the Invention Means for Solving the Problems Embodiments of the Invention (a) Embodiments of the First Invention (FIG. 1) (b) Description of an embodiment of the second invention (FIG. 2) (c) Description of an embodiment of the third invention (FIG. 3) (d) Description of an embodiment of the fourth invention (FIG. 4) (e) Description of specific embodiment when performing static scheduling (FIGS. 5 to 11) (f) Description of specific embodiment when performing dynamic scheduling (FIGS. 12 to 12) 14) Effect of the invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a disk array device for accessing a plurality of disk devices in parallel to perform data input / output processing, and more particularly to RAID 5
I / O instruction scheduling technology for a disk type disk array device.

[0003]

2. Description of the Related Art In recent years, as the performance of CPUs has increased, the performance of memory subsystems has fallen relatively, and there is a demand for improved processing speed. The disk array device has a plurality of inexpensive and small-sized disk devices, stores data in a distributed manner among these disk devices, and accesses and drives them in parallel.

In the disk array device, by performing data transfer to a plurality of disk devices in parallel, it is possible to perform high-speed data transfer that is twice as many as the number of disk devices as compared with the case where only one disk device is used. In addition to data, by recording redundant information such as parity data, it is possible to detect and correct data errors caused by disk device failures, etc., and duplicate the contents of the disk device. It is possible to achieve the same high reliability as in the case of recording at a lower cost than in the case of duplication.

Such disk array devices are classified into five types depending on the difference in storage system. In any of these types, the disk array device is different from a single disk device in that it has a mechanism for extracting performance. (Array controller) is required. Where the 5
The two types are RAID (Redundant Arrays of Inexpe
nsiveDisks; cheap disk duplication array) 1 to 5 are abbreviated.

In the RAID 1 type disk array device,
It is provided with a mirror disk device that stores a copy (mirror image) of the data stored in the disk device. It is widely used because the disk device has low utilization efficiency but has redundancy and can be realized with simple control. . In the RAID 2 type disk array device, data is striped (divided) in units of bits or bytes, and reading and writing are performed in parallel in each disk device. The striped data is physically recorded in the same sector in all disk devices. A Hamming code generated from data is used as the error correction code. In addition to the data disk device, a disk device for recording the humming code is provided, and the failed disk device is specified from the humming code to restore the data. By providing the redundancy by the Hamming code in this way, correct data can be secured even if the disk device fails, but it has not been put into practical use because the utilization efficiency of the disk device is poor.

RAID 3 type disk array device 100
As shown in FIG. 15, for example, A is composed of an array controller 110 and four magnetic disk devices 111 to 114. The array controller 110 reads / writes to / from each of the disk devices 111 to 114 according to an input / output command (I / O command) from the host computer 120.
It controls write access. And RAID
In the 3-type disk array device 100A, for example, the data a, b, and c are data a1 to bit or sector units.
a3, b1 to b3, c1 to c3, the parity P1 is calculated from the data a1 to a3, and the data b1 to
Parity P2 is calculated from b3, parity P3 is calculated from data c1 to c3, and these data a1 to a3 and b are calculated.
1 to b3, c1 to c3, and parities P1 to P3 are written by accessing the disk devices 111 to 114 simultaneously and in parallel by the array controller 110, as shown in FIG.

In such a RAID3, data redundancy is maintained by parity. Further, the data writing time can be shortened by the parallel processing of the divided data. However, since a single seek or write access requires parallel seek operations of all the disk devices 111 to 114, it is effective when continuously handling a large amount of data, but a small amount of access is required. In the case of transaction processing in which data is randomly accessed, the high speed of data transfer cannot be utilized and efficiency decreases.

RAID 4 type disk array device 100
B has the same hardware configuration as that shown in FIG. 15, as shown in FIG.
In the ID4 type disk array device 100B, one data is divided into sector units and the same disk device 111 to
I am writing to 114. For example, the disk device 111
The data a is divided into sector data a1 to a4 and written therein. The parity is stored in the fixed disk device 114. Here, the data a1, b
1 and c1, the parity P1 is calculated, and the data a2
Parity P2 is calculated from b2 and c2, and data a
Parity P3 is calculated from 3, b3 and c3, and parity P4 is calculated from data a4, b4 and c4.

In such a RAID4 type disk array device 100B, data reading is performed by the disk device 1
11 to 113 are performed in parallel. For example, the reading of the data a is performed by accessing the sectors 0 to 3 of the disk device 111 and sequentially reading and combining the sector data a1 to a4. On the other hand, in the data writing, since the new parity is calculated and written after reading the data and the parity before the writing, a total of 4 is written in one writing.
You need to access it twice. For example, the disk device 11
When the sector data a1 of 1 is updated (rewritten), the original data (a1) _OLD of the update location and the original parity (P1) _{OL D} of the corresponding disk device 114 are read out and are consistent with the new data (a1) _NEW . The operation of writing for the obtained new parity (P1) _NEW is required besides the data writing for updating.

Further, since access to the disk device 114 for parity always occurs at the time of writing, it is not possible to execute writing to a plurality of disk devices at the same time. For example, even if the writing of the data a1 of the disk device 111 and the writing of the data b2 of the disk device 112 are simultaneously performed, it is necessary to read the parities P1 and P2 from the parity from the same disk device 114 and write them after calculation. Therefore, writing cannot be executed at the same time. Although RAID4 is defined in this way, it has few merits, and therefore, there are currently few moves to put it into practical use.

RAID 5 type disk array device 100
As for C, as shown in FIG. 17, the hardware configuration itself is the same as that shown in FIGS. 15 and 16, but
In this RAID5 type disk array device 100C,
By not fixing the disk device for parity like RAID4, parallel read / write access to each disk device 111-113 is enabled. That is, FIG.
As shown in, the disk devices 111 to 114 for storing the parity are different for each sector. Again, Figure 1
6, the parity P1 is calculated from the data a1, b1 and c1, the parity P2 is calculated from the data a2, b2 and c2, the parity P3 is calculated from the data a3, b3 and c3, and the data a4, b4. c4
The parity P4 is calculated by.

In such a RAID5 type disk array device 100C, for parallel reading and writing, for example, the data a1 of the sector 0 of the disk device 111 and the data b2 of the sector 1 of the disk device 112 are parity P1.
Since P2 is placed in different disk devices 114 and 113, reading and writing can be performed simultaneously without overlapping. The overhead that requires a total of four accesses at the time of writing is the same as in the case of RAID4.

As described above, RAID 5 asynchronously accesses a plurality of disk devices 111 to 114 to read / write.
Since write can be executed, it is suitable for transaction processing in which a small amount of data is randomly accessed. here,
Generation of parity data in RAID 4 and RAID 5 data writing will be described as follows.

First, in the disk array device storing the redundancy information, the exclusive OR of the data at the same storage position in a plurality of disk devices is taken as parity by the following equation (1) and is held in the disk device for parity. doing. Data a (+) Data b (+) Data c (+) ... = Parity P (1) However, in the above equation, (+) represents an exclusive OR symbol.

Storage location of data and parity is RAID
In the case of No. 4, as shown in FIG.
It is fixed to. On the other hand, in RAID5, as shown in FIG.
As shown in FIG.
4 to eliminate the concentration of access to a specific disk device due to the parity read / write operation. When reading these RAID4 and RAID5 data,
Since the data in the disk devices 111 to 114 is not rewritten, the consistency of the parity is maintained, but the parity also needs to be changed according to the data at the time of writing. For example, one original data (a1) in the disk device 111
_{When OLD} is rewritten to new data (a1) _NEW , in order to obtain the consistency of parity P1, the following formula (2) is calculated, and the parity is updated to update the parity of the entire data of the disk device. The consistency can be maintained.

Original data (+) Original parity (+) New data = New parity (2)

[0018]

In the RAID 5 type disk array device 100C described above with reference to FIG. 17, data is cyclically stored in each of the disk devices 111 to 114 (striping). Therefore, if the array controller 110 determines that there is no contention for access to the same disk device, it simultaneously issues I / O commands (input / output commands) to different disk devices that are in an operable state.

However, the conventional disk array device 100
C considers only the speedup by parallel execution of I / O instructions,
The speedup of the I / O instruction access processing in each of the magnetic disk devices 111 to 114 is not taken into consideration, and there are the following problems (1) to (3). Each magnetic disk device 1 constituting the disk array device
In 11 to 114, the time (seek time) when moving the head to a desired track occupies most of the data transfer processing, which is a factor that hinders the high speed processing of the data transfer.

From the point of view of the CPU in the host computer 120, it can be considered that the write command to the disk array device 100C having the write-back type cache is completed when the data is written to the cache. However, the read command is issued by the host computer 120.
Since the processing ends when the data transfer to the main storage device is completed, the job on the CPU side is kept waiting during that time. The shorter the execution time in the CPU, the larger the waiting time becomes, and the job processing time is affected. That is, in particular, the response time at the time of a read request from the host computer 120 to the array controller 110 greatly affects the job processing time.

Data is stored in the magnetic disk devices 111 to 11
4 is written, the parity is updated. At that time, as described above, it is necessary to read the original data at the data write position and the original parity at the write position of the paity to be updated. However, each individual magnetic disk device 1
Since the access scheduling of the I / O instruction is performed for each 11 to 114, if the read processing of the original data and the original parity is synchronized, there is a problem that the scheduling effect is impaired.

The present invention was devised in view of the above problems, and realized the improvement of performance such as speeding up of data transfer processing by shortening / reducing the seek time of the head in each disk device. A first object is to provide a disk array device. Further, according to the present invention, the response time of the read command in each disk device can be shortened, the waiting time on the host side is improved (overhead time is shortened), and the processing performance on the host side is improved. A second object is to provide a disk array device having the above structure.

Further, the present invention provides a disk array device in which the read processing of the original data and the original parity at the time of updating the parity can be performed asynchronously so as to improve the throughput of each disk device. The purpose is 3.

[0024]

Therefore, in the disk array device of the first invention, a plurality of disk devices and a read / write operation for each disk device according to an input / output command from the outside are performed.
In an array controller for controlling write access, a plurality of I / O commands received from the outside are rearranged so that the number of head track movements in each disk device is reduced, and each I / O command is executed. Is provided with a scheduling function for instructing the array controller (claim 1).

The disk array device of the second invention comprises a plurality of disk devices and an array controller for controlling read / write access to each disk device in response to an external input / output command. In, the decoding function that analyzes the instruction type of the input / output instruction received from the outside, and the execution of each input / output instruction by rearranging the plurality of input / output instructions received from the outside based on the instruction type analyzed by this decoding function Is provided with a scheduling function for instructing the array controller (claim 2).

Further, in the disk array device of the third invention, a plurality of disk devices and a disk device for parity according to an input / output command from the outside are read in parallel to each disk device without being fixed. In a RAID5 type disk array device having an array controller for controlling execution of write / write access, tag numbers are added to original data required for parity update and input / output instructions relating to the original parity. When executing an input / output instruction assigned a tag number and a tag number, the tag number is registered and the read state of both the original data and the original parity is managed to read both the original data and the original parity. The parity control function that notifies the array controller that the parity update process has become possible It is characterized in that (claim 3).

The disk array device of the fourth invention comprises a plurality of disk devices and an array controller for controlling read / write access to each disk device in accordance with external input / output commands. In the above, the plurality of I / O commands received from the outside are rearranged so that the number of head track movements in each disk device is reduced, and the array controller is instructed to execute each I / O command. When another I / O instruction to be executed with priority over the instruction is accepted, a dynamic scheduling function for instructing the array controller to execute the other I / O instruction prior to the scheduled I / O instruction It is characterized in that it is provided (Claim 4).

[0028]

Embodiments of the present invention will be described below with reference to the drawings. (A) Description of Embodiment of First Invention FIG. 1 is a block diagram showing a disk array device as an embodiment of the first invention. As shown in FIG. Each magnetic disk device 1 according to an input / output command (hereinafter referred to as an I / O command) from a disk device 10 and a host computer (not shown) side.
Array controller 1 for controlling read / write access to 0 and storage device 2 for storing I / O instructions
And a scheduling device 3A having information necessary for determining the execution order of I / O instructions and an arithmetic function for executing scheduling according to a predetermined scheduling algorithm.

Here, the storage device 2 and the scheduling device 3A rearrange a plurality of I / O commands received from the outside (host computer) so that the number of head track movements in each magnetic disk device 10 is reduced. A scheduling function for instructing the array controller 1 to execute each I / O instruction is realized. Next, the operation of the disk array device configured as shown in FIG. 1 will be described.

The I / O command received by the array controller 1 from the host computer is stored in the storage device 2. The scheduling device 3A appropriately fetches I / O instructions from the storage device 2 and changes (schedules) the instruction execution order among the I / O instructions so that the overall seek time is shortened. The scheduled I / O instruction sequence is stored in the storage device 2. Array controller 1
Retrieves scheduled I / O instructions from the storage device 2 and sequentially executes them. When the scheduled I / O instructions stored in the storage device 2 run out, the scheduling device 3A performs the scheduling process again.

The scheduling algorithm executed by the scheduling device 3A will be described below.
The scheduling device 3A distributes a plurality of I / O commands around a track on which the head is located. At this time, a group of I / O instructions having a larger track number than the head-located track number is sorted in ascending order, while a group of I / O instructions having a smaller track number than the head-located track number. Sorts in descending order. For convenience, the terms are defined as follows.

“Row [Ascend]”: I / O instruction sequence sorted in ascending order “row [Descend]”: I / O instruction sequence sorted in descending order “leng [Ascend]”: I / O instruction sorted in ascending order The distance between the head position and the track number accessed by the I / O instruction located farthest from the head in the column "leng [Descend]": The farthest from the head in the I / O instruction sequence sorted in descending order Distance between the track number accessed by the located I / O instruction and the head position "num [Ascend]": Number of instructions contained in the I / O instruction sequence sorted in ascending order "num [Descend]": Sorted in descending order The number of instructions included in the I / O instruction sequence In addition, an operation of concatenating another I / O instruction sequence at the end of a certain I / O instruction sequence is represented by “+”.

In the scheduling algorithm executed by the scheduling device 3A, "leng [Ascend]"
And "leng [Descend]" are compared with each other, and the execution order of the I / O instructions is determined as follows in accordance with the magnitude relation. 1. When leng [Ascend]> leng [Descend]: row [Descend] + row [Ascend] 2. When leng [Ascend] <leng [Descend]: row [Ascend] + row [Descend] 3. If leng [Ascend] = leng [Descend]: (a) If num [Ascend]> num [Descend], then row [Descend] + row [Ascend] (b) If num [Ascend] <num [Descend] , Row [Ascend] + row [Descend] (c) If num [Ascend] = num [Descend], there is no particular order specification, and even if starting from row [Ascend], row [Des]
cend] 4. When the access requests are biased in the ascending or descending direction (when the track number to be accessed exists on only one side of the head position): row [Ascend] or row [Descend].

In this way, each magnetic disk drive 10
In addition to load distribution to each disk device 10, scheduling is also performed in each magnetic disk device 10.
It is possible to shorten or reduce the seek time of the head in, and to improve the service time of the magnetic disk device 10.
Performance such as speeding up of data transfer processing will be greatly improved.

(B) Description of Embodiment of Second Invention FIG. 2 is a block diagram showing a disk array device as an embodiment of the second invention. As shown in FIG. In addition to having a plurality of magnetic disk devices 10, an array controller 1 and a storage device 2 similar to those shown in FIG. 1, an instruction decoding device for analyzing the instruction type of an I / O instruction received from the outside (host computer) 4 and a scheduling device 3B having information necessary for determining the execution order of I / O instructions and an arithmetic function for executing scheduling according to a predetermined scheduling algorithm.

Here, the instruction decoding device 4 performs the decoding function, and the storage device 2 and the scheduling device 3B have the same scheduling function as that of the above-described scheduling device 3A (the number of head track movements in each magnetic disk device 10 is changed). A scheduling function of rearranging a plurality of I / O instructions so as to reduce the number of I / O instructions and instructing the array controller 1 to execute each I / O instruction), and a plurality of I / Os based on the instruction type analyzed by the instruction decoding device 4. It performs a scheduling function of rearranging O instructions and instructing the array controller 1 to execute each I / O instruction.

Next, the operation of the disk array device configured as shown in FIG. 2 will be described. Upon fetching the I / O instruction from the storage device 2, the instruction decoding device 4 analyzes the fetched I / O instruction. Next, I analyzed
The / O command is classified according to its command type, that is, whether the I / O command is a read command or a write command for the disk array device (magnetic disk device 10).

The scheduling device 3B performs scheduling for each of the read instruction sequence and the write instruction sequence based on the above-mentioned algorithm. However, the track accessed by the tail I / O instruction of the read instruction sequence after scheduling is used as the reference track of the head during the write instruction scheduling process.

After scheduling, the write command sequence is connected to the end of the read command sequence and stored in the storage device 2. The array controller 1 fetches scheduled I / O instructions from the storage device 2 and sequentially executes them. When the scheduled I / O instructions stored in the storage device 2 run out, the scheduling device 3B performs the scheduling process again.

By such processing, the read command can be executed with priority over the write command, and the response time of the read command can be reduced as compared with the case where the read / write command is mixedly executed. Therefore, the waiting time on the side of the host computer making the I / O request is also improved, which contributes to the improvement of the processing performance on the side of the host computer.

(C) Description of Embodiment of Third Invention FIG. 3 is a block diagram showing a disk array device as an embodiment of the third invention. As shown in FIG. In addition to having a plurality of magnetic disk devices 10 similar to those shown in FIG. 1, an array controller 1A, a storage device 2 and a scheduling device 3C,
The parity control device 5 is provided.

Here, the array controller 1A reads / reads in parallel to each magnetic disk device 10 without fixing the magnetic disk device 10 for parity according to an I / O command from the outside (host computer). By controlling the write access, the RAID5 type disk array device is realized. Further, the array controller 1A has a tag number assigning function for assigning a tag number to the input / output command relating to the original data and the original parity which is required at the time of updating the parity.

As with the scheduling device 3A shown in FIG. 1, the scheduling device 3C schedules the I / O instruction fetched from the storage device 2 and, at the same time, detects the tagged I / O instruction, and then the parity control device 5 It has a function to register the tag number. The parity control device 5 is for supporting the asynchronous execution of the read instruction at the time of updating the parity, and when executing the I / O instruction to which the tag number is added, the tag number is registered by the scheduling device 3C. A parity control function that manages the read states of both the original data and the original parity and notifies the array controller 1 that the parity update processing is possible when both the original data and the original parity are read out. Is playing.

Next, the operation of the disk array device configured as shown in FIG. 3 will be described. I / O related to original data and original parity required for parity update
The O instruction is converted by the array controller 1 into an I / O instruction with a tag number.

Then, the scheduling device 3C schedules the I / O instruction fetched from the storage device 2 and, at the same time, detects a tagged I / O instruction, registers the tag number in the parity control device 5, and Guarantees the association between the data and the original parity. The array controller 1 fetches I / O commands from the storage device 2 and sequentially executes them. When the processing in the magnetic disk device 10 is completed, the array controller 1 transfers to the parity control device 5.
This is notified.

The parity control device 5 determines whether or not the I / O processing end notification from the array controller 1 is a read command at the time of parity update. As a result, when both the original data and the original parity data are prepared, the parity control device 5 reports to the array controller 1 that the parity calculation process can be executed. If only one of the data is included, that fact is registered.

By such processing, there is no need to synchronize the original data / the original parity at the time of updating the parity, that is, the original data and the original parity at the time of updating the parity can be asynchronously read. Therefore, the effect of the scheduling performed for each magnetic disk device 10 is not impaired, and each magnetic disk device 1
The throughput of 0 can be improved.

(D) Description of Embodiment of Fourth Invention FIG. 3 is a block diagram showing a disk array device as an embodiment of the third invention. As shown in FIG. In addition to a plurality of magnetic disk devices 10, an array controller 1 and a storage device 2 similar to those shown in FIG. 1, a dynamic scheduling device 6 that dynamically determines the execution order of I / O instructions, and this dynamic scheduling. I determined by the device 6 to be executed preferentially
Priority register 7 for storing / O instruction
It is configured with.

Here, the dynamic scheduling device 6
Multiple I / s received from outside (host computer)
The O commands are rearranged so that the number of head track movements in each magnetic disk device 10 is reduced, and each I / O command is reordered.
When the array controller 1 is instructed to execute an O instruction and another I / O instruction to be executed with priority over the scheduled I / O instruction is accepted, the other I / O instruction is given priority. Store in register 7,
It performs a dynamic scheduling function of instructing the array controller 1 to execute another I / O instruction prior to the scheduled I / O instruction.

Next, the operation of the disk array device configured as shown in FIG. 4 will be described. The I / O command received by the array controller 1 from the host computer is stored in the storage device 2. The dynamic scheduling device 6 fetches an I / O instruction from the storage device 2 as appropriate,
The instruction execution order is changed (scheduled) between I / O instructions so that the overall seek time is shortened.

At this time, the dynamic scheduling device 6
Starts the scheduling process when there is no scheduled I / O command in the storage device 2 or when the I / O command process starts in the magnetic disk device 10. In the latter case (when the processing start of the I / O instruction in the magnetic disk device 10 is triggered), the I / O instruction fetched from the storage device 2 and the scheduled I / O stored in the storage device 2 Scheduling processing is performed with the instruction.

The scheduled I / O instruction group is
It is stored in the storage device 2. Further, the I / O that needs to be executed prior to the I / O instruction stored in the storage device 2
The instruction is stored in the priority register 7. The array controller 1 receives the scheduled I / O from the storage device 2 or the priority register 7.
The O instruction is fetched and sequentially executed.

By such processing, it becomes possible to dynamically rearrange the execution order of the I / O instruction sequence for each magnetic disk device 10, further improve the service time of the magnetic disk device 10, and perform data transfer. Performance such as speeding up of processing can be further improved. (E) Description of Specific Embodiment When Performing Static Scheduling FIG. 5 is a block diagram showing a specific embodiment of a disk array device that performs static scheduling. The disk array device shown in FIG. 5 is a combination of the disk array devices described above with reference to FIGS. In FIG. 5, the same reference numerals as those used above indicate almost the same parts, and the detailed description thereof will be omitted.

The disk array device shown in FIG. 5 is RAID
This type of disk array device uses the array controller 1A having the functions (tag number assigning function, etc.) described above with reference to FIG. In the disk array device shown in FIG. 5, the storage device 2 described above is configured by the FIFO buffer 21, the instruction decoding device 4 described above is configured by the fetch unit 41 and the decode unit 42, and the parity control device 5 described above is used. Is a reservation register 51 and a tag register 52
It consists of.

Further, the scheduling device 3 (see FIGS.
The scheduling device 3A to 3C shown in FIG. 3) has a general register 31, a head information register 32, an ordering register 33, and a counter 3.
4, a sequencer 35, an arithmetic unit 36 and an I / O register 37. Although the ordering register 33 of the scheduling device 3 is included in the storage device 2 in the embodiments shown in FIGS. 1 to 4, it is provided here in the scheduling device 3. Therefore, in the embodiment shown in FIGS. 1 to 4, the I / O instruction to be executed by the array controller 1A was read from the storage device 2, but in the embodiment shown in FIG. The / O instruction is adapted to be read from the ordering register 33 of the scheduling device 3 via the I / O register 37.

In FIG. 5, reference numeral 38 denotes an external interface (I / F) which operates as an interface between the array controller 1, the FIFO buffer 21 (storage device 2) and the scheduling device 3. . The FIFO buffer 21 is an external I / F 3
It stores a plurality of I / O commands sent from the host computer via 8 and accepted by the array controller 1.

In the instruction decoding device 4, fetch /
The unit 41 fetches the I / O instruction from the FIFO buffer 21, and the decode unit 42 analyzes the I / O instruction fetched from the fetch unit 41. In the parity control device 5, the reservation register 51 has a configuration described later with reference to FIG. 8 and holds a queue of asynchronous read instructions at the time of updating the parity, and the tag register 52 is shown in FIG. The tag number is registered by the scheduling device 3 in order to determine whether or not the I / O command executed in each magnetic disk device 10 is related to parity update, and has a configuration described later.

In the scheduling device 3, the general-purpose register 31 is used as a working buffer when executing scheduling, and the head information register 3 is used.
2 has a configuration which will be described later with reference to FIG. 6, and data (HDD0 to HD) indicating the position of the head on each magnetic disk device
D n-1) is held. Also, ordering
The register 33 has a configuration described later with reference to FIG. 7, and stores a scheduled I / O instruction sequence,
The counter 34 is the I / O in the ordering register 33.
The number of O commands is shown as a count value.

The sequencer 35 gives a predetermined timing to the arithmetic unit 36 and the fetch unit 41 in order to instruct the arithmetic unit 36 to perform a scheduling process or to instruct the fetch unit 41 to fetch an I / O instruction. To send the instruction. The arithmetic unit 36 performs an actual scheduling process according to an instruction from the sequencer 35 based on the above-described predetermined algorithm and the analysis result by the decoding unit 42. The I / O register 37 is provided in each magnetic disk device. This is for temporarily storing the I / O command executed in 10 when the scheduling device 3 outputs it to the target magnetic disk device 10 via the array controller 1.

The FIFO buffer 21, the head information register 32, the ordering register 33, the counter 34, and the tag register 52 are provided for the magnetic disk device 10, respectively. FIG. 6 shows the configuration of the head information register 32. In FIG. 6, reference numeral 321 represents a field in the head information register 32 corresponding to each magnetic disk device, and stores the track number where the head is located. In the figure, HD
D0 to HDD n-1 are device numbers set corresponding to the n magnetic disk devices 10. FIG. 7 shows the configuration of the ordering register 33. In FIG. 7, reference numeral 331 denotes a field corresponding to each magnetic disk device in the ordering register 33, and m number of I / Is for every n magnetic disk devices 10. It has a structure capable of storing an O instruction.

FIG. 8 shows the structure of the reservation register 51. In FIG. 8, 511 is a field for tag number. By registering the tag number in this field 511, the original data at the time of parity update and It is possible to judge the dependency of the original parity.
512 is a status bit, and this status bit 512
Shows the execution state of the read request at the time of updating the parity.

For example, the status bit 512 includes 2-bit information "00 (S0)", "01 (S1)", "1".
0 (S2) ”and“ 11 (S3) ”are set, where 00 (S0): indicates“ this field is invalid ”. 01 (S1): indicates“ NO DEFINITION ” 10 (S2): Indicates that this field is used for parity update.

11 (S3): indicates that "either the original data or the original parity has been read". FIG. 9 shows the configuration of the tag register 52. In FIG. 9, 521 is a V (Valid) bit, and this V bit 521 is provided for each magnetic disk device 10. If "1" is set,
It indicates that the I / O access for updating the parity is being executed to the magnetic disk device 10. Also,
522 is a tag field, and this tag field 5
The tag number is registered and stored in 22 as in the field 511 described above.

Next, the operation of the disk array device for static scheduling according to the present embodiment, which is configured as described above, will be described. When an I / O request is issued to the disk array device shown in FIG. 5 from a host computer (not shown) which is a higher-level device, the I / O command is stored in the FIFO buffer 21 from the array controller 1 through the external I / F 38. To be done. At this time, when the read processing of the original data or the original parity accompanying the parity update occurs, the array controller 1 adds the tag number to the I / O command.

The fetch unit 41 is the sequencer 3
According to the instruction from 5, a plurality of I / O instructions are fetched from the FIFO buffer 21. The upper limit of the fetchable I / O instruction is the I that can be stored in the ordering register 33.
The number of / O instructions is m. After fetching the I / O instruction, the decode unit 42 analyzes the I / O instruction and classifies the I / O instruction according to the read / write instruction. I /
The O instruction sequence is divided into a read instruction / write instruction and stored in the general-purpose register 31. Sequencer 3 after storing I / O instructions
Reference numeral 5 instructs the arithmetic unit 36 to start the scheduling process.

The arithmetic unit 36 determines the execution order of the I / O instructions in the general-purpose register 31 based on the head position information 321 in the head information register 32 according to the above-mentioned algorithm. The I / O instruction sequence after scheduling is stored in the ordering register 33 corresponding to the magnetic disk device. The number of I / O instructions stored in the ordering register 33 is written in the counter 34.

When the read instruction generated by the parity update is detected during the scheduling in the arithmetic unit 36, the tag number 511 registered in the reservation register 51 is searched. At this time, (a) When there is a match with this tag number in the reservation register 51: No particular processing is performed.

(B) When there is no match with this tag number in the reservation register 51: "10 (S2)" is set in the status bit 512 of the reservation register 51. Upon receiving an instruction to transfer an I / O command from the array controller 1 through the external I / F 38, the sequencer 35 requests the arithmetic unit 36 to start processing.

The arithmetic unit 36 fetches an I / O instruction from the ordering buffer 33 and extracts it from the I / O register 3
Store in 7. At this time, the count value of the counter 34 is decremented every time the I / O instruction is fetched. Also,
If the I / O instruction is a tagged I / O instruction, the V bit 521 of the tag register 52 corresponding to the magnetic disk device 10 is turned on (“1”), and the tag number field 52 is displayed.
Fill in and register the tag number in 2.

External I / F instructed by the sequencer 35
38 transmits the I / O command stored in the I / O register 37 to the array controller 1. The array controller 1 inputs the received I / O command to the magnetic disk device 10. When the I / O command processing in the magnetic disk device 10 is completed, the sequencer 35 informs the external I / F 3
8 to receive from the array controller 1. The sequencer 35 issues a processing request to the arithmetic unit 36.

At this time, the arithmetic unit 36 executes the following processing. 1. The magnetic disk device 10 in the tag register 52
V bit 521 corresponding to (A) When V bit 521 is off: No processing is performed. (B) When the V bit 521 is on: The tag number 522 is read and the following processing is performed.

2. The tag number in the reservation register 51 is searched. The one that matches the tag number 522 is sought, and the status bit 512 is set as follows. (A) When the status bit 512 is S2, the status bit 512
Is set to S3. (B) When the status bit 512 is S3, the status bit 512
After setting S to S0, the sequencer 35 notifies the array controller 1 that the parity update process can be started.

As described above, according to the disk array device of the embodiment shown in FIGS. 5 to 11, the seek time is reduced by scheduling the access request in each magnetic disk device 10 of the disk array device. It becomes possible to do. Therefore, the performance of the disk array device (performance such as speeding up of data transfer processing) is significantly improved.

When the I / O instructions are not distinguished by read / write, the response times of the read requests are T0, T1 and T2, respectively, as shown in FIG.
And the average response time is (T0 + T1 + T
It was 2) / 3. On the other hand, in the present embodiment, as shown in FIG. 10B, by giving a read command to the magnetic disk device 10 preferentially as compared with a write command, the response time of each read request is different. t0 (≦ T0), t1 (≦ T1) and t2 (≦ T
2), and the average response time is (t0 + t1 +
t2) / 3 [≦ (T0 + T1 + T2) / 3] is improved. That is, the waiting time in the CPU (host computer) is significantly improved, and the processing performance on the host side is greatly improved. In FIGS. 10A and 10B, the CPU and the main storage device are provided on the host computer side.

Further, when the original data / original parity read at the time of updating the parity is synchronized, the waiting time of the I / O instruction k becomes T0, as shown in FIG. 11 (a). On the other hand, as in the present embodiment, by performing the original read / original parity read asynchronously, the waiting time of the I / O instruction k is t.
It is shortened to 0 (≦ T0). That is, by controlling the read states of both the original data and the original parity by the tag number, the original parity / original data can be read asynchronously, and the throughput of each magnetic disk device 10 is greatly improved. .

(F) Description of Specific Embodiment when Dynamic Scheduling is Performed FIG. 12 is a block diagram showing a specific embodiment of a disk array device that performs dynamic scheduling. The disk array device shown in FIG. 12 is similar to that shown in FIGS.
It is a combination of the disk array devices described above. In FIG. 12, the same reference numerals as those used above indicate almost the same parts, and the detailed description thereof will be omitted.

The disk array device shown in FIG. 12 is also RAI
The D5 type disk array device also uses the array controller 1A having the functions (tag number assigning function, etc.) described above with reference to FIG. Further, the disk array device shown in FIG. 12 does not have the function described above with reference to FIG. 2, so the decode unit 42 shown in FIG. 5 is omitted. Further, the function as the head information register 32 shown in FIG. 5 is incorporated in the scheduling information register 61 as described later with reference to FIG.

In the disk array device for performing this dynamic scheduling, as shown in FIG. 12, the dynamic scheduling device 6 includes a general-purpose register 31, an ordering register 33, a counter 34, and a register similar to those described above with reference to FIG. Sequencer 35, arithmetic unit 36 and I / O
In addition to the register 37, the scheduling information register 6
1 is configured.

This scheduling information register 61
13 has a configuration described later with reference to FIG. 13, and stores head position information and I / O command information stored in the ordering register 33. Further, the priority register 7 described above in FIG. 4 is also provided,
In the priority register 7, as described above,
I / O instructions to be preferentially executed are stored with respect to the I / O instructions stored in the ordering register 33.

FIG. 13 shows the scheduling information register 6
In FIG. 13, reference numeral 611 is an entry field corresponding to each magnetic disk device 10, and this entry field 611 stores the track number of each instruction. Further, 612 is a U / D _I bit, and this U / D _I bit 612 is set with information indicating the positional relationship of the track of the command target with respect to the head. U indicates "UP" (the track number is larger than the head position), and D indicates "D".
OWN "(the track number is smaller than the head position), and the subscript" I "indicates that the bit 612 corresponds to an instruction.

Reference numeral 613 is a field for storing the track number of the head (head position on the magnetic disk device 10), and 614 is the U / D _H bit, which is the U / D
_{The H} bit 614 is set to indicate the moving direction of the head. Note that U indicates “UP” (moving from the current head position in the direction of the larger track number),
D indicates "DOWN" (moving from the current head position in the direction of the smaller track number), and the subscript "H"
Indicates that the bit 614 corresponds to the head.

Next, the operation of the disk array system for dynamic scheduling according to the present embodiment configured as described above will be described. This embodiment performs the same processing as that of the disk array device for performing static scheduling described above with reference to FIGS. 5 to 11, except for dynamic scheduling of I / O instructions. In the present embodiment, the fetch unit 41 that requests fetching of I / O instructions is
When the I / O instruction is fetched from the IFO buffer 21, the instruction is stored in the general-purpose register 31. After storing the I / O instruction, the sequencer 35 instructs the arithmetic unit 36 to start the scheduling process.

The arithmetic unit 36 compares the track number accessed by the I / O instruction with the head number 613 in the scheduling information register 61. As a result of the comparison, (a) [track number of head in field 613]>
In the case of [track number of I / O instruction]: U / D _I bit 612 = D (b) of the I / O instruction [track number of head of field 613] <
In the case of [track number of I / O instruction]: U / D _I bit 612 = U of the I / O instruction.

Next, the U / D of the I / O instruction that was previously obtained
_I bit 612 and scheduling information register 61
U / D of the above I / O instruction [i, j] (0≤i≤m-2)
Compare with _I bit 612. At this time, 1. If it matches the U / D _I bit 612: Compare with the track number [i, j] 611.

(A) [Track number [i, j] 611]
> [Track number of the I / O command] • If U / D _I = U, before I / O command [i, j],
The I / O instructions are sorted and the ordering register 3
3 is stored. At this time, if it is the I / O instruction [0, j], the I / O instruction is the priority register 7
To store.

If U / D _I = D, I / O instruction [i +
1, j], the U / D _I bit 612 is compared again. (B) [track number [i, j] 611] <[corresponding I /
When the track number of the O command] U / D _I = U, the U / D _I bit 612 is compared again for the I / O command [i + 1, j].

If U / D _I = D, I / O instruction [i,
j], the I / O instructions are sorted and stored in the ordering register 33. At this time, I / O instruction [0,
j], the I / O instruction is stored in the priority register 7. 2. If it matches U / D _I bit 612: See U / D _H bit 614.

If it matches the U / D _H bit 614, the I / O instruction is sorted before the I / O instruction [i, j] and stored in the ordering register 33. If the U / D _H bit 614 does not match, the U / D _H bit 614 is compared again for the I / O instruction [i + 1, j]. If it is the last I / O instruction stored in the scheduling information register 61, this instruction is registered at the end of the ordering register 33.

When the sequencer 35 receives an instruction to transfer an I / O command from the array controller 1 through the external I / F 38, the sequencer 35 requests the arithmetic unit 36 to start processing. The arithmetic unit 36 has a priority
If there is an I / O instruction in the register 7, the I / O instruction is fetched from the priority register 7 and, if not, from the ordering register 33 and stored in the I / O register 37.

External I / F instructed by the sequencer 35
38 transmits the I / O command stored in the I / O register 37 to the array controller 1. Thus, FIG.
According to the disk array device of the embodiment shown in FIGS. 14A-14D, the seek time can be reduced by scheduling access requests in each magnetic disk device 10 as in the case of the disk array device shown in FIGS. The array device performance (performance of speeding up data transfer processing, etc.) is significantly improved, and original parity / original data can be read asynchronously.
The 0 throughput is greatly improved.

Further, as shown in FIG. 14A, the head track movement distance during static scheduling is set to "Di
stance _S ”, Distance _s = leng (i) + leng (j) + leng (k), but as shown in FIG. 14B, the track movement distance of the head when dynamic scheduling is performed is Di
If stance _D ”, Distance _D = leng (i) + leng (j).

That is, by performing the dynamic scheduling, the track movement distance for the I / O command k is
“Distance _s −Distance _D (= leng (k))”, seek time is also reduced by O (leng (k)). Therefore, the dynamic I / O request scheduling reduces the seek time more than the static scheduling, further improves the service time of the magnetic disk device 10, and further improves the performance such as speeding up of data transfer processing. It can be further improved.

[0093]

As described above in detail, according to the disk array device of the first invention, in addition to the load distribution to each disk device, the scheduling is performed for each disk device. There is an effect that the seek time of the head is shortened / reduced, the service time of each disk device is improved, and the performance such as speeding up of data transfer processing can be greatly improved.

Further, according to the disk array device of the second aspect of the present invention, the read command can be executed with priority over the write command, and the response time of the read command is greatly increased as compared with the read / write command mixed execution. Therefore, the waiting time on the higher-level device (host) side that makes an input / output request is improved, which also contributes to the improvement of the processing performance on the higher-level device side.

Furthermore, according to the disk array device of the third invention, it is not necessary to synchronize the read processing of the original data / the original parity at the time of updating the parity, that is, the original data and the original parity at the time of updating the parity. Since the read processing can be performed asynchronously, when the scheduling is performed for each disk device as described above, the effect of the scheduling is not impaired, and the throughput of each disk device can be improved.

According to the disk array device of the fourth aspect of the present invention, it is possible to dynamically rearrange the execution order of the input / output instruction sequence for each disk device, and further improve the service time of the disk device. The performance such as speeding up of data transfer processing can be further improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a disk array device as an embodiment of a first invention.

FIG. 2 is a block diagram showing a disk array device as an embodiment of a second invention.

FIG. 3 is a block diagram showing a disk array device as an embodiment of a third invention.

FIG. 4 is a block diagram showing a disk array device as an embodiment of a fourth invention.

FIG. 5 is a block diagram showing a specific embodiment of a disk array device that performs static scheduling.

FIG. 6 is a diagram showing a configuration of a head information register in the present embodiment.

FIG. 7 is a diagram showing a configuration of an ordering register in the present embodiment.

FIG. 8 is a diagram showing a configuration of a reservation register in the present embodiment.

FIG. 9 is a diagram showing a configuration of a tag register in the present embodiment.

10 (a) and 10 (b) are diagrams showing a response time shortening effect for a read instruction in the present embodiment, as compared with the related art.

11A and 11B are diagrams showing an effect of an asynchronous operation at the time of parity update processing according to the present embodiment, as compared with the related art.

FIG. 12 is a block diagram showing a specific embodiment of a disk array device that performs dynamic scheduling.

FIG. 13 is a diagram showing a configuration of a scheduling information register of this embodiment.

FIGS. 14A and 14B are diagrams showing the effects of performing the dynamic scheduling according to the present embodiment of FIGS. 14A and 14B in comparison with the case where the dynamic scheduling is not performed.

FIG. 15 is a block diagram for explaining a RAID 3 type disk array device.

FIG. 16 is a block diagram for explaining a RAID 4 type disk array device.

FIG. 17 is a block diagram for explaining a RAID 5 type disk array device.

[Explanation of Codes] 1 Array Controller 1A Array Controller (Tag Numbering Function) 2 Storage Device (Scheduling Function) 3,3A, 3B, 3C Scheduling Device (Scheduling Function) 4 Instruction Decoding Device 5 Parity Control Device (Parity Control Function) 6 Dynamic Scheduling Device (Dynamic Scheduling Function) 7 Priority Register 10 Magnetic Disk Device 21 FIFO Buffer 31 General-purpose Register 32 Head Information Register 33 Ordering Register 34 Counter 35 Sequencer 36 Arithmetic Unit 37 I / O Register 38 External Interface 41 Fetch Unit 42 Decode unit 51 Reservation register 52 Tag register 61 Scheduling information register

Claims (4)

[Claims] 1. A disk array device comprising a plurality of disk devices and an array controller for controlling read / write access to each of the disk devices in response to external input / output commands A plurality of input / output instructions are rearranged so that the number of head track movements in each disk device is reduced, and a scheduling function for instructing the array controller to execute each input / output instruction is provided. , Disk array device. 2. A disk array device comprising a plurality of disk devices and an array controller for controlling read / write access to each of said disk devices in response to an external input / output command. A decode function for analyzing the instruction type of the input / output instruction, and a plurality of input / output instructions received from the outside are rearranged based on the instruction type analyzed by the decode function, and the array controller is instructed to execute each input / output instruction. A disk array device, which is provided with a scheduling function to perform. 3. A plurality of disk devices and a read / write access to each of the disk devices are executed in parallel without fixing the disk device for parity according to an input / output command from the outside. In a RAID 5 type disk array device including an array controller for controlling the above, a tag number assigning function for assigning a tag number to an input / output instruction related to the original data and the original parity required at the time of updating the parity, and a tag number When executing an input / output instruction assigned with, the tag number is registered, the read state of both original data and original parity is managed, and parity is updated when both original data and original parity are read. A parity control function for notifying the array controller that processing has become possible is provided. Kuarei apparatus. 4. A disk array device comprising a plurality of disk devices and an array controller for controlling read / write access to each of said disk devices in response to an external input / output command. The plurality of input / output instructions are rearranged so that the number of head track movements in each disk device is reduced, the execution of each input / output instruction is instructed to the array controller, and the input / output instructions have priority over the scheduled input / output instructions. When another I / O instruction to be executed is accepted, a dynamic scheduling function is provided to instruct the array controller to execute the other I / O instruction prior to the scheduled I / O instruction. A disk array device characterized by the above.