JP2003084923A - Constituting method of cluster type disk array device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To optimize the performance and reliability of a disk array having a cluster constitution. SOLUTION: Parity groups of the disk array are dispersed and arranged among a disk column 211 (0, 02), a disk column 222 (1, 06), a disk column 233 (2, 10), and a disk column 244 (3, 14) of respective different collective disk devices 21, 22, 23, and 24 belonging to a plurality of different disk array devices. Thus, when an individual disk array device fails, data use is allowed by a data recovering procedure similar to that of a usual disk failure, and the performance of the disk array is optimized by changing the constitution of the parity groups in response to a type of load.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a configuration technology of a cluster type disk array device, and more particularly, to redundancy and performance improvement at the time of failure of the disk array device, and more particularly to combining low cost disk devices. The present invention relates to a technology for configuring a highly reliable disk array device.

[0002]

2. Description of the Related Art A disk array device is a device that uses RAID technology to improve data transfer performance and fault tolerance by accessing a plurality of disks in parallel. For this RAID technology, see “A Case for Red
undant Arrays of InexpensiveDisks (RAID) ”; In Pr
Details on oc. ACM SIGMOD, June 1988 (published by the University of California, Berkeley).

A disk array device using this RAID technology is currently widely used as a large-capacity storage device. Among them, RAID 1 described in this paper is particularly mentioned.
A method called (mirroring) and RAID5 is generally adopted. In RAID 5, several disks (generally 4 to 8 disks) are used. In the following description, assuming that there are 4 units, in RAID 5, 1 block of data called parity is provided for 3 blocks of data to improve fault tolerance, and 4 blocks are added to the above 4 disks due to this data and parity. The performance is improved by operating the three disks storing the data in parallel when writing the data and reading the data. In particular, in RAID 5, the disk for storing this parity data is not fixed to a specific one, but is distributed and stored in the above four disks in order to equalize the load. A plurality of disks thus set are called a parity group.

As another reference technology which is a premise of the present invention, a disk array device having a cluster configuration is known. As a method of constructing a high-performance and highly-reliable system at a lower cost than in the past, in the field of servers such as databases, a technology has been known from a comparatively early stage in which multiple independent servers are combined in a high-speed connection network to form a cluster configuration. Was there. As a method of constructing a large-scale system, a configuration in which a large number of systems with slightly inferior performance are combined in a high-speed connection network has a certain level of performance rather than building a single system with extremely high performance. This is because it has been proved that the system can improve the performance with respect to the cost. Up to now, the configuration using a single controller has been general in the disk array device, but recently, a similar configuration has begun to be considered even if it is placed in the disk array. For example, Japanese Patent Laid-Open No. 2001-27972
Japanese Unexamined Patent Publication (Kokai) Publication discloses a method of constructing such a disk device.

[0005]

The cluster type disk array device as described in the above publication has some technical problems. The following will be described in order.

First, in a cluster type disk array device, one disk controller receives a request for data from a server serving as a host, transfers it to a disk array device that actually has data, and transfers the actual data. By reading and writing, a single image disk array that is not just a set of disk array devices is realized.

However, due to cost and electrical restrictions, the data path for exchanging control information and data between the disk array devices has to be slower than the internal path of the disk array, resulting in a cluster. Type disk array devices have a difference in performance depending on the position of the interface connected to the host. Especially when access is concentrated on specific data (hotspots) from multiple hosts, both the disk array device that forms part of the cluster and the coupled network between the clusters become a bottleneck. The original high performance cannot be obtained.

[0008] Conventionally, in such a case, a software copy is used to cope with the load by placing a copy of data or the like, but in a usage such as an Internet server in which a hot spot changes rapidly, such a countermeasure is taken. Have difficulty.

As another technical problem, when a cluster is composed of a plurality of disk array devices, it must be taken into consideration that one of the disk array devices being configured is out of order or fails due to a failure. However, the cluster-type disk array device disclosed in the related art has not dealt with such a case.

Specifically, in the cluster type disk array device disclosed in the prior art, all the data on the disks connected to the disk array device forming the cluster becomes inaccessible, so the disk as a system is considered. As an array device, it is inevitable that the availability will drop.

The present invention solves the above problem by arranging data in consideration of the cluster configuration.

That is, it is an object of the present invention to realize optimization of performance and reliability in a cluster type disk array device composed of a plurality of disk array devices having a cluster configuration.

Another object of the present invention is to prevent performance degradation in a cluster type disk array device comprising a plurality of disk array devices of a cluster configuration due to concentration of access to a specific disk array device.

Another object of the present invention is to prevent a decrease in availability of a disk device due to a failure of a specific disk array device or the like in a cluster type disk array device consisting of a plurality of disk array devices of a cluster configuration. .

[0015]

According to the present invention, there is provided the following configuration means in a cluster type disk device having a configuration in which a plurality of disk array devices are connected by information transmission means.

(1) A disk device forming a parity group is selected from disk devices connected to a plurality of disk array devices. That is, the disk devices forming the parity group are distributed and arranged in a plurality of disk array devices.

When all the disk devices constituting the parity group are placed on different disk array devices, failure or stop of one disk array device is the same as failure of one disk device for the parity group. The data can be read as usual although the data generation penalty occurs. Further, assuming that the access is concentrated on a specific parity group from each of the multiple disk array devices, the disk devices of the parity group are similarly specified by being distributed among the multiple disk array devices. The load is not concentrated on the disk array device, and the load can be effectively distributed.

(2) Further, it is possible to configure a disk array device capable of dynamically changing the configuration of the distributed arrangement of the disk devices across a plurality of such disk array devices. In a cluster-type disk array device, there is a trade-off between balancing the load between individual disk array devices and reducing the load on the cluster coupling device. Therefore, the optimum placement of the disk devices that make up a parity group according to access from the host. Will change. A cluster type disk array device with optimum performance can be realized by providing a function of dynamically changing this in advance or by statistical information or the like.

(3) As a further advanced configuration, when a disk interface in each disk array device is provided with an alternate interface for the disk controller, the alternate interface is used to control the disk of another disk array device. By connecting a disk device to a plurality of disk array devices by connecting to the device, it is possible to adopt a system configuration of a cluster type disk array device with further improved fault tolerance.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

(Embodiment 1) FIG. 1 is a conceptual diagram for explaining an example of the operation of a method of constructing a cluster type disk array device according to a first embodiment of the present invention, and FIG. It is a conceptual diagram which shows an example.

As illustrated in FIG. 2, in the present embodiment, four disk array control devices (hereinafter referred to as boxes for clarifying the distinction by terms) 11, 12, 1
3 and 14 are connected by the cluster coupling device 30 to equivalently form one disk array device (cluster type disk array device). Disk array controller 11,
Each of 12, 13, 14 is a collective disk device 21, 22, 23, 24, each of which comprises a plurality of physical disks.
Are connected to the disk array control device 11 via the interface 21a, the collective disk device 22 to the disk array control device 12 via the interface 22a, and the collective disk device 23 via the interface 23a. The disk array control device 13 has an interface 24a for the collective disk device 24.
Are sequentially connected to the disk array control device 14 via.

The collective disk devices 21, 22, 23,
The interior of each of the 24 is further composed of four rows of disks. The aggregate disk device 21 internally has disk columns 211 (disk numbers 00 to 03) and 212 (disk numbers 04 to 0).
7), 213 (disk numbers 08 to 11) and 214 (disk numbers 12 to 15). Collective disk device 2
The same applies to 2, 23 and 24.

That is, the aggregate disk device 21 and the disk array controller 11, the aggregate disk device 22 and the disk array controller 12, and the aggregate disk device 23.
And each of the disk array control device 13, the aggregate disk device 24, and the disk array control device 14 constitute a plurality of independent disk array devices (box numbers 0 to 3). A cluster-type disk array device is configured by connecting them to each other by the cluster connecting device 30.

In addition, all the boxes 11, 12, 13,
There is a management console 16 coupled to 14, and settings and the like are performed using this management console 16. The internal structures of the boxes 11, 12, 13, 14 are the same. For example, inside the box 11 side, a host controller 111, a control storage device 112, a disk controller 113, which are mutually connected via a bus, It has a cache storage device 114.

Further, in the present embodiment, description will be given in the context of a disk array device which adopts RAID5 control using 3D + 1P as a parity group, that is, parity 1 for the data area 3. In this embodiment, the placement of the parity group on the disk array is performed from the management console 16.

In such a disk array device, for ease of management, it is common to show to the host side that an access from the host is made to a virtual disk called a logical volume. Correspondence from the logical volume to the actual physical disk is box 11-1
4 has a management table inside, and a host controller 111 and a disk controller 113 inside the disk array.
Is being done in. This also has the effect of hiding the actual physical configuration from the host, and this embodiment utilizes this effect as described below.

FIG. 3 shows an example of the configuration of the management table 300 used in this embodiment. The management table 300 is used, for example, by storing the same ones in some of the control storage devices 112 to 142 of the boxes 11 to 14 and the like, and setting status of a parity group across the entire cluster in any box. It is possible to grasp.

The management table 300 of this embodiment has a logical volume number 301 and a logical volume number 3
A parity group number 302 corresponding to 01, three data (data 1 to data 3) physical disks configuring the parity group of each parity group number 302, and a physical disk entry designating one physical disk for parity. 303 to 305 and a physical disk entry 306.

Each physical disk entry can specify a parity group consisting of a plurality of physical disks arranged across boxes in the format of (box number, disk number). The setting example of the parity group 0 in FIG. 3 corresponds to the arrangement example of the 3D + 1P physical disks in FIG. The setting example of the parity group 7 corresponds to the arrangement example of the 3D + 1P physical disks in FIG. 5 of the reference technology.

As described above, in the case of the present embodiment, by setting the management table 300, the disk groups forming the parity group are distributed over a plurality of boxes, or the disk groups of a specific disk array device are distributed. It is possible to freely set parity groups in cluster type disk array devices, such as allocating parity groups centrally to the aggregate disk device or mixing and distributing distributed and centralized parity groups. is there.

In the cluster type disk device of the reference technology, the parity group is set in the collective disk device connected to the same disk array device (box), specifically, for example, in the collective disk device 21 under the control of the disk array controller 11. One or a plurality of disks are selected from the four rows of the disk rows 211, 212, 213 and 214, and they are configured as a parity group. This arrangement is shown in FIG. This configuration has two advantages.

(1) It is possible to manage the parity groups of the boxes forming the cluster in the same manner as when the cluster is not formed. The disks that make up the parity group are managed only by that box, so it is not necessary to be aware of the fact that it is a cluster, neither in terms of hardware nor software.

(2) Also, since paridi data and other data need only be sent to the disks connected to the same cluster, generation and processing of parity data is closed in the box,
To be easy.

Further, in the configuration of the reference technique using the arrangement as shown in FIG. 5, when access concentrates on specific data, both the specific disk array device forming the cluster and the connection network become a bottleneck, which is high in the original cluster. There is a technical problem that the performance cannot be exhibited, and if the specific disk array device fails, the data belonging to the disk array device becomes completely unusable.

On the other hand, one of the preferable arrangements in the case of adopting the configuration of arranging the data in a distributed manner among the boxes in this embodiment according to the present invention is shown in FIG. In this arrangement, boxes 11 to 14 are used, one disk of each aggregate disk device connected to each box is selected, and three data and parity are distributed and arranged therein.
Hereinafter, for convenience of explanation, the disk row 211 (0, 02),
Disk row 222 (1,06), disk row 233
(2,10) and the disk row 244 (3,14). At this time, the processing from the host is as follows.

(1) The read process from the host is simply reading from the target data disk, and the host controller (for example, the host controller 111) is read to a predetermined host.
Transfer more data. When a large data area is taken on the disk array, continuous data may be arranged in a plurality of boxes without taking the configuration of the present invention. It is a natural extension of the normal processing of.

(2) The writing from the host is the same as the method of the reference technique as the operation of the data, but the disk control device of another box (for example, the disk control device 143) is started to create the parity data at the time of the writing. Then, a process for obtaining parity data is added.

In the present embodiment, since data and parity are distributed and arranged over a plurality of boxes, the availability of data is not impaired even if one of the boxes 11-14 is stopped or closed. . For example, assume that the box 12 is stopped. At this time, the data of the disk row 222 that constitutes the parity group cannot be accessed, but the disk rows 211, 233, and
Since 244 is accessible, it is possible to obtain the required data according to the normal RAID5 data recovery procedure.

Further, in the present embodiment, since the data and the parity are arranged in a distributed manner over a plurality of boxes, the load can be distributed in a relatively small unit of the data forming the parity group. However, when adopting such a cluster type configuration, it is necessary to consider the data arrangement. This is because, for example, when focusing on the box 11, access to a certain short data from the host connected thereto is assigned to any box in the reference technology, and in the arrangement of the present embodiment, a plurality of boxes 11 are allocated.
14 are distributed to the aggregated disk devices 21 to 24 under control. This has the advantage that a substantially constant data transfer performance can be realized regardless of the data arrangement, the amount of data transfer between boxes increases, and the cluster coupling device 30 between boxes is easily saturated, and the data arrangement is optimized. It has two drawbacks: higher performance cannot be achieved. Therefore, the distributed arrangement according to the present invention is particularly effective when (1) the availability of data that can withstand a box failure is required.

(2) When it is required that the performance does not fluctuate significantly due to the data arrangement in a situation where an average load is applied to all boxes and lower than the peak value.

(3) In the case where a load is applied such that access is concentrated in a short period of time on some and unspecified data from all hosts according to the access pattern. Web server data for the Internet is considered to have this behavior. On the other hand, as a typical case, it is more suitable to arrange the data by concentrating on the aggregate disk device under the control of a specific box as in the reference technology of FIG. ) When peak throughput is required. However,
In this case, it is necessary to know the pattern of data to be accessed with the host to some extent or to assume that it does not fluctuate rapidly in the short term, and place the data in the optimum box according to the knowledge of this access pattern to maximize the cluster performance. be able to.

(2) When the number of writes is very large, frequent exchange of data to be written and parity data through the cluster combiner 30 for parity generation occurs, and the cluster combiner 30 is likely to be saturated. Depending on the system implementation, in the case of writing mainly, the arrangement of the reference technology of FIG. 5 may be able to improve the average system performance. There are two possibilities. Therefore, maintenance personnel and system administrators (SE, etc.) from the management console 16 rewrite the management table 300 as appropriate based on system requirements to optimally arrange the system data, thereby achieving optimum performance and reliability as a whole. The owning disk array device can be configured.

Although the present embodiment has been described on the assumption that maintenance personnel and SEs judge these two data arrangements and select and instruct them from the management console 16, the statistical data etc. are acquired and the disk array device alone is acquired. It is also possible to change these two data arrangements dynamically without any manual intervention in cooperation with software. In that case, in the management table 300 illustrated in FIG. 3, the mapping from the logical volume to the physical disk described above is changed inside the disk array device according to the maintenance worker's instruction, so that the host is not aware of the change. It can be carried out.

Further, in the present embodiment, the case of adopting the RAID5 configuration is described for convenience of description, but RAID1 and R are used.
Other parity configurations can be adopted even with AID5, and the above embodiment can be applied as it is.

In this embodiment, the number of disk areas and the number of boxes forming the parity group are the same. However, the method of arranging the parity group of the present invention is such that the number of boxes is larger than the number of disk areas. It is easy to expand to a box that has only the number of disks that make up the parity group in consideration of load distribution.

In terms of improving fault tolerance among the effects of the present invention, the number of boxes needs to be larger than the number of disk areas forming a parity group, but load balancing does not have that restriction. For example, in a 3D + 1P cluster type disk array device consisting of two boxes, if a parity group is configured by selecting, for example, two physical disks from each box, data cannot be read in the event of a box failure. Load balancing, which is one of the effects, is more effective, and for this purpose, the present invention can be implemented to configure a cluster-type disk array device that achieves targeted performance characteristics.

(Second Embodiment) A preferred second embodiment of the present invention will be described below. FIG. 4 shows an example of the configuration of the cluster type disk array device of this embodiment, and FIG. 6 shows the configuration of the disk array device based on the reference technology.

In the present embodiment, as illustrated in FIG. 4, unlike the first embodiment, the collective disk device 2
Each of the interfaces 21a and 21b, the interface 22a and the interface 22b, the interface 23a and the interface 23b, the interface 24a and the interface 24b of the two systems, which are provided in the units 1, 22, 23 and 24, are provided in their own boxes (disk array controller ) And another box (disk array control device) are connected to improve fault tolerance.

That is, in the configuration of the reference technique of FIG. 6, these two systems of interfaces are connected to the same box, whereas in the present embodiment of FIG. 4, they are connected across different boxes. The difference is.

In this embodiment, like the first embodiment, four disk array control devices (hereinafter referred to as boxes to clarify the distinction by terms) 11, 12, 1
The cluster coupling device 30 connects the devices 3 and 14 to form an equivalent disk array device. Each control device includes a collective disk device 21 including a plurality of disks,
22, 23 and 24 are combined.

As described above, the collective disk device in this embodiment has two interfaces, and in the structure of this embodiment, the collective disk device 21 is the disk array controllers 11 and 12, and the collective disk device 24 is the disk. Coupled in turn to array controllers 13 and 14, respectively. This is a difference from the configuration of the reference technology of FIG. 6, and in the case of the reference technology, the two interfaces of the collective disk device 21 are both connected to the same disk array control device 11. The inside of each aggregate disk device is the same as that of the first embodiment.

In addition, all the boxes 11, 12, 13,
There is a management console 16 coupled to 14, and settings and the like are carried out using this console as in the first embodiment. Individual disk array controller 11, 12, 1
The internal structures of 3 and 14 are the same, and for example, the disk array control device 11 side has a host control device 111, a control storage device 112, a disk control device 113, and a cache storage device 114. Further, in the present embodiment, 3D + 1P as the parity group, that is, the data area 3
On the other hand, a description will be given in the context of a disk array device that employs RAID 5 control using parity 1. In this embodiment, the placement of the parity group on the disk array is performed from the management console 16.

When the configuration of the present embodiment is adopted, the processing is the same as that of the configuration of the reference technology of FIG. 6, and the difference is that from the box in operation which remains at the time of the occurrence of a box failure to the target aggregate disk device. It is necessary to be aware of access in terms of processing, and the load should be appropriately distributed to the two interfaces of the connected aggregate disk device. Regarding the latter load distribution, the host controller of each box accepts the processing from the actual host, and from that point it allocates to the disk controller considering the load distribution, but the distribution processing itself. Is no different from the processing that distributes the load to the two interfaces in the box in the reference technology of Fig. 6, and the different part considers the load distribution evaluation by anticipating the overhead of the processing that crosses the boxes because one side is a different box. It is only to do. However, as long as the throughput of the cluster coupling device 30 is sufficiently large, there is no problem in terms of performance even if the processing across the boxes is not taken into consideration as in the method of the reference technique.

Further, in this embodiment, two boxes are grouped to share the collective disk device with each other. For example, the collective disk device 21 is in the boxes 11 and 12, and the collective disk device 22 is in the collective disk device 22. In the boxes 12 and 13, the collective disk device 23 is
14, the collective disk device 24 includes boxes 14 and 11
It is also possible to shift in sequence and connect in a chained manner as in, and the like, and the same idea is applied in processing.

Although the invention made by the present inventor has been specifically described based on the embodiments, the present invention is not limited to the above-mentioned embodiments, and various modifications can be made without departing from the scope of the invention. Needless to say.

[0057]

According to the present invention, it is possible to realize optimization of performance and reliability in a cluster type disk array device comprising a plurality of disk array devices having a cluster configuration.

According to the present invention, in a cluster type disk array device composed of a plurality of disk array devices having a cluster configuration, it is possible to prevent performance deterioration due to concentration of access to a specific disk array device. Is obtained.

According to the present invention, in a cluster type disk array device consisting of a plurality of disk array devices having a cluster configuration, it is possible to prevent a decrease in availability of the disk device due to a failure of a specific disk array device or the like. The effect is obtained.

[Brief description of drawings]

FIG. 1 is a conceptual diagram illustrating an example of an operation of a method of configuring a cluster type disk array device according to a first exemplary embodiment of the present invention.

FIG. 2 is a conceptual diagram showing an example of a configuration of a cluster type disk array device according to the first exemplary embodiment of the present invention.

FIG. 3 is a conceptual diagram showing an example of a configuration of a management table used in the cluster type disk array device according to the first exemplary embodiment of the present invention.

FIG. 4 is a conceptual diagram showing a configuration example of a cluster type disk array device which is a second exemplary embodiment of the present invention.

FIG. 5 is a conceptual diagram showing a setting state of a parity group in a cluster type disk array device which is a reference technique corresponding to the method of configuring the cluster type disk array device according to the first exemplary embodiment of the present invention.

FIG. 6 is a conceptual diagram showing a configuration of a cluster-type disk array device of a reference technique corresponding to a second embodiment of the present invention.

[Explanation of symbols]

11, 12, 13, 14 ... Disk array control device (disk control means), 16 ... Management console, 21 ... Collective disk device, 21a, 21b ... Interface, 22
... aggregate disk device, 22a, 22b ... interface, 23 ... aggregate disk device, 23a, 23b ... interface, 24 ... aggregate disk device, 24a, 24b ...
Interface, 30 ... Cluster coupling device, 300 ... Management table (configuration control means), 301 ... Logical volume number, 302 ... Parity group number, 303-306
… Physical disk entry.

Claims (3)

[Claims] 1. A cluster type disk array having a cluster configuration in which a plurality of disk array devices each having a plurality of disk devices and a disk control means for controlling the disk devices are connected by an information transmitting means. A method for configuring a cluster-type disk array device, wherein a plurality of the disk devices forming a parity group are distributed and arranged in the plurality of disk array devices forming a cluster. 2. The method for configuring a cluster-type disk array device according to claim 1, wherein the arrangement positions of the disk devices forming the parity group in one or more disk array devices forming a cluster are variably controlled. A cluster-type disk array device configuration method comprising a configuration control means capable of performing the configuration control. 3. A cluster type disk array device having a cluster configuration in which a plurality of disk array devices each having a plurality of disk devices and a disk control means for controlling the disk devices are connected by an information transmission means. In each of the disk array devices, the disk device includes first and second connection interfaces, and the disk control means of the own disk array device is connected to the first connection interface, A method of configuring a cluster type disk array device, characterized in that each of the disk devices is shared by a plurality of the disk array devices by connecting a connection interface to the disk control means of another disk access device.