RAID（Redundant Arrays of Inexpensive Disks)
It was advocated in thesis "A Case for Redundant Arrays of Inexpensive Disks (RAID)" by David Paterson of the University of California Barkly school, Garth A. Gibson, and Randy H. Katz in 1988. This was announced with SIGMOD Conference 1988: pp 109-116.
This thesis is a proposal how high storage of reliability (auxiliary memory) should be constructed with a large capacity by using hard disk (Inexpensive Disk) of the reliability of low capacity and the price corresponding at a low price. Five kinds from RAID 1 to RAID 5 are defined in the thesis according to the composition of the hard disk. Moreover, striping is written in general in the thesis though it is not proposed as RAID 0.
As for RAID 2 of six kinds of defined in the beginning, it is hardly used, and RAID 3 and RAID 4 are also in the minority in the entire RAID. RAID 0, RAID 1, RAID 5, and these combinations of three methods are used today. RAID 6 that enhances RAID 5 back is defined, and it is used by the scene for which trouble resistance is more necessary than RAID 5.
It thinks whether to value in reliability, the speed, and the budget (The use efficiency of the hard disk is included) for the user who examines the introduction, and the introduction method of putting on the fact together can be selected. There is a realistic method of filling two enough though it is difficult to fill three splendidly.
Though a large-scale composition of combining the disk array device with built-in two or more areas composed of RAID 0 and RAID 5 with the disk array another device in the basic server of the big enterprise and the important business server, etc. that cannot be stopped and making it to double by RAID 1 is not unusual, in a general personal computer user and a small-scale office, it is general to adopt the RAID 1 method with a small load of an initial investment and installation requirements in the beginning, to add RAID 0 if necessary afterwards, and to do the up-grade to RAID 5. It leads to the reduction of the labor for the system administrator on the site because the number of goods that compose RAID beforehand increases in products of a business server, and RAID can be used at once at the time of purchase.
Even if several hard disks break down according to the composition of RAID, a virtual disk might be able to be operated. In that case, it is possible to keep operating it without stopping the device by exchanging it for the hard disk of taking the place detaching the hard disk that breaks down with a virtual disk operated. PC recognizes even if it connects it while PC is starting like this, and the function that can be immediately used is called "Hot-swap (hot plug)". Because it is required to correspond as hardware on the device side to use the hot-swap functionality, the hot-swap is substantially standard equipment in the device from which 24 hour continuous duty like the server usage etc. is requested.
RAID spreads in a large-scale business server, the workstation, and the particular purpose of which the essential high-speed processing of mass data and is the improvement of trouble resistance it is spreading to a small-scale server and the personal computer in recent years though it was used for the manufactured computer equipment etc.
Method of RAID
The method of mounting RAID includes the method of achievement with the method of achievement with hardware (hardware method) and software (software method).
Hardware method
The hardware method has two methods with the one to use the RAID controller card and a disk array unit.
RAID controller card
Serial ATA RA identification card
Serial ATA RA identification card
This method installs the card that is called RAID controller in the device, and leaves the parity operation and the management of the disk. There is a product that mounts the RAID controller beforehand in a part of mother board, and, in that case, the RAID controller card need not be bought.
The load of CPU is decreased so that the RAID controller may do all concrete processing only by sending the instruction with a little CPU. However, because control techniques are different in each controller, it is necessary to install the driver who correspondence to each RAID controller card is necessary on OS side, and attached to the RAID controller card. It doesn't worry about this work it is to be able to install the driver of other peripherals difficultly by the user.
Another hardware is more necessary for the hot-swap though it corresponds to the hot spare (function that switches to the preliminary drive prepared at once beforehand when the drive breaks down) if it is a high-level to some degree RAID controller card.
It is easy for an individual and a small-scale entrepreneur to introduce it because a lot of low-priced products are sold. Moreover, it is the best for a personal personal computer where a processing performance of CPU low, and various work are generated one after another. It can be said the most excellent method for the user who is accustomed to opening the case of PC in the cost performance. However, there is a product of the specification that controls RAID to CPU of the device allotting it for the RAID controller prepared for in the RAID controller card and the mother board that makes the low price a selling point. There is in the existence located in the middle of pure hardware RAID and software RAID this by the performance compared with pure hardware RAID and is a dependency to inferior, specific OS.
The difficult point is a barrier for the user that there are resistance in opening the case, the installation of the RAID controller card, the generation of work to try to connect the signal cable of the hard disk with the RAID controller card, and its opening the case. Moreover, this method cannot be used because there is often no space that increases the hard disk in compact desktop system PC. The notebook-sized personal computer cannot use it for the same reason.
Recently, the chipset for the server that accumulates the RAID0/1/0+1 controller function in the chipset is widely widespread. This hardware doesn't do the parity operation etc.It is composed of MUX (multiplexer)/DEMUX (demultiplexer), and a basic structure distributes disk I/O operation that CPU issued to disk I/O to two or more hard disks and multiplexes reading from the disk. The synthesis and the accelarator elements that are the speed-ups of the speed of disk I/O are strong to say nothing of reliability.
Disk array unit
There is no load to CPU at all unnecessary because it sees it from the computer and OS side as a drive of mere SCSI and the fiber channel of a special driver. In addition, because it has the auto recovery function from the beginning the hot-swap and the hot spare, the operation cost can be decreased. When the hard disk in the disk array unit breaks down, the function to notify the watch software the state through the serial port and Ethernet is equipped. Moreover, there is the one that one disk array unit can be shared with two servers by using Defarenshal SCSI and the fiber channel in the disk array unit, too. There is the one that two or more disk array units are combined and the server for the control of the exclusive use made a set, too.
If there is already an external, connected interface necessary for the device that wants to connect the disk array unit, it comes to be able to use it only because the case of the device is not opened, and the cable ties.
Software method
As for the software method, OS manages two or more disks through a usual drive controller (IDE, SCSI, and FC, etc.). This method has the advantage that special hardware need not be bought though the load to CPU is higher than that of the hardware method.
Beforehand..hard disk..build into..device..have..device..case..open..controller card..install..existing..hard disk..signal..cable..connection..change..time..at all..unnecessary..become.Therefore, this method (If two hard disks or more are built into beforehand) is recommended to the user who feels resistance in opening the case.
CPU takes charge of the operation of parity in case of RAID 3, 4, 5, and 6 that uses parity. There are a lot of abilities of hardware RAID controller card and cases with the parity operation ability with CPU of the device where it is not so inferior, and can demonstrate the performance that is the hardware method and not inferior even by the software method. However, because a part of the ability of CPU is spared for the parity generation when there are a lot of processing demands for CPU in the device, it becomes the cause of the performance decrease.
The software method is a best choice in the usage in which CPU may concentrate on disk I/O as a file server and a network storage. In the method with hardware, when weak point "Controller" of maximum RAID breaks down and operates abnormally, nothing is mortgaged (MTBF as the whole decreases though it improves if the controller is tediously composed how much it is). There is a technique of making the disk array that is the hardware method double by RAID 1 with software to attempt the improvement of reliability or more, too. Strong reliability and stability can be obtained by making the controller and the power-supply unit another system.
It is an excellent method in the cost performance to adopt the operation method in which shifting to the hardware method when RAID is first united by the software method, it uses actually, and dissatisfaction on the speed side comes out. The advantage that RAID is concealed can freely unite the plan of actually using it is to be able to select various methods according to each one's environment and budget like this and shifting to another method.
Achievement method
In software RAID achievement, two or more methods such as OS and a filesystem, single purpose software chipsets, and increase RAID boards exist.
Windows supports the RAID function in the NTFS filesystem. Linux is supported by the kernel. FreeBSD is supported with software named gmirror. In ZFS of the filesystem, function (RAID-Z) of the RAID5 corresponding exists. In the chipset of Intel, the RAID function is supported by the matrix storage manager function. In a low-priced RAID board of Adaptec, software RAID is done by HOST RAID.
RAID level
It is RAID 0, RAID 1, RAID 5, RAID 6 to be actually used from RAID 0 to RAID 6 of seven kinds of, and has been often limited the level that can be used with the RAID controller and software.
RAID 1 or RAID 5 is chiefly used as a server usage to value Data integrity. There is a case to build RAID 0 into these and to aim at speed-up, too, when the load that rests upon one server is high in the environment from which the number of the server is limited. (It is necessary to examine the of course increase of the server, and the reduction of the load for one, and it is necessary to examine convenience and the budget etc. of the degree and installation features of the load multilaterally. Extravagant expectations should not be drawn only to covering with RAID 0 only and speed-up. ) It doesn't depend on the method of RAID, and it is important to catch the situation to be enforceable of a prompt hard disk exchange when the trouble occurs, and it is preferable for the server usage to use the product of the hot spare and the hot swappable.
A certain kind of application preserves the control information in the filesystem of RAID 1 or RAID 5, and moreover, the multimedia data is preserved in RAID 0 and backed up to a tape and optical media.
It is possible to make reliability combine each RAID levels and united to the speed. However, the power supply can endure the load for the number of the hard disk when the RAID controller card is especially used so that the necessary number of the hard disk may increase and the hard disk is installed in the device or you should note it.
RAID 0: Disk array (striping) in which breakdown-proof of breakdown
in not finding array is not found RAID 0 RAID 0
RAID 0 was distributed data to two or more hard disks, read and written, and sped up. This is called striping. It is assumed that it is not included in RAID because there is no tedium and is not trouble resistance either, and is called RAID 0. Should two drives or more to RAID 0.
Because all data in the array is lost only because one drive broke down, failure rates are higher than single purpose drives. For instance, it becomes about 2%(1-0.99*0.99=0.0199) and the failure rate rises to about twice when RAID 0 is composed of two though the failure rate is 1% if it is one when assuming that the failure rate of the drive when using it for a certain period on a certain condition was 1%.
It is often used from the decrease of reliability (breakdown) in shape to make RAID 1 described later and RAID 5, etc. striping to attempt the improvement of the speed in single RAID 0 though the speed improves.
Merit
* All the capacity of the composition drive can be used.
* The more the number of drives increases, the more the access speed
goes up.
Weak point
* There is no tedium at all.
* The more the number of drives increases, the more the failure rate
goes up.
* In drawing out the performance by 100%, the bus architecture with the band that corresponds to the improvement of the I/O performance is needed.
RAID 1: Doubling (mirroring) RAID 1 RAID 1
RAID 1 writes the same at the same time content on two or more hard disks. This is called a mirroring. RAID 1 is simplest RAID, and RAID that trouble resistance is high. Moreover, it is easy to correspond to the breakdown in the controller who is the maximum weak point of RAID. Should two drives or more to RAID 1.
The possibility that the other breaks down at the same time, too can keep be low when one
breaks down, and the system being operated.
For instance, about 0.01%(0.01*0.01=0.0001) and the failure rate become 1/100, and trouble resistance goes up when RAID 1 is composed of two though the failure rate is 1% if it is one when assuming that the failure rate of the drive when using it for a certain period on a certain condition was 1%.
However, because it has the same data as the plural, it becomes below the half of the capacity of the hard disk that composes the array as a data capacity that can be treated. The more the number of the hard disk increases in RAID 1, the more the use efficiency of the hard disk worsens. RAID 5 is better if attaching importance to the use efficiency of the hard disk.
Reading RAID 1 is originally accessed only to one drive, and the other is a complete backup drive. It is possible to read it by enhancing the algorithm to read this from both at the speed twice theoretical and a single purpose disk.
However, it reads in software RAID 1 that is not done mounting that always values the performance, and being mounted by Windows NT, and reading is always both done only from the other disk by one as for writing. Windows Server 2003 does the load-balancing from one at a high load at a low load.
Merit
* Restoration is fast.
* It is possible to use it without RAID hardware/software.
* The redundant configuration corresponding to the controller's
breakdown is possible.
* The more the number of drives increases, the more trouble resistance
goes up.
Weak point
* Because a capable amount of the data logging to the capacity of the composition disk is always the reciprocal twice the number of the composition disk, it is inefficient.
* The capacity of RAID 1 is decided to the smallest capacity in the drive that composes, and a remaining part cannot be used.
* It has only aimed at without making to a large capacity nor the speed-up of
the size of the disk in RAID.
* When the drive is added to an existing system, and RAID 1 is introduced, close attention is necessary. Because there is danger of reproducing the drive of the addition that contents are empty, and all critical data might be lost. Therefore, it is necessary to confirm the direction of the reproduction of data enough. Especially, because the confirmation on the operation screen becomes difficult when the drive of the same model is used, further attention is requested. (However, this is a problem that only has to be borne in mind to take the backup beforehand. RAID and the backup are good examples of the thing not substituted though complementary. )
RAID 0+1
RAID 0+1
RAID 1+0
RAID 1+0
RAID 1+0 and RAID 0+1
The composition in which RAID 0 is combined with RAID 1 is specially called RAID 1+0(RAID 10)/RAID 0+1(RAID 01). The improvement of the speed, capacity, and trouble resistance can be attempted by combining RAID 0 that aims at speeding up and making to a large capacity and RAID 1 by which high reliability was requested. It is necessary by four drives or more.
RAID 0 and RAID 1 are congenial, and trouble resistance rises more than RAID 6 depending on the combination. When trouble resistance that capacity is necessary and strong is requested, it is often adopted because it is possible to correspond also to the controller's making to double because RAID 1 is used.
The name is changed according to RAID 0, RAID 1, and which is done early. It is different in respect of trouble resistance though it seemingly both looks similar.
* RAID 0+1: It is a mirror as for the area where the stripe was done.
* RAID 1+0: It is a stripe as for the mirror set.
As for the tolerance to the drive breakdown, RAID 1+0 is more excellent. Data is one not destroyed as long as neither of two composition drives of the RAID single-unit that composes RAID 0 breaks down in RAID 1+0, too though the data when each drive and one breaks down in the RAID 0 area where RAID 1 is composed in RAID 0+1 is destroyed. RAID 1+0 is two according to increasing if the number increases where trouble resistance goes up. You may basically think that RAID 1+0 is more excellent about the tolerance to the controller breakdown three the phase that RAID 0+1 exceeds though it can exist.
RAID 2: Special error correcting code drive in each bit(ECC)
The mechanism of making to tedium of RAID 2 is a humming sign, and the unit of the stripe is one bit. RAID 2 doesn't have the practicality because the data restoration by the humming code is not lower than that of the necessity the reliability of HDD, and the product is not marketed.
Because a number of drives necessary for mounting RAID 2 is different depending on the number of bits that can be corrected, it is not possible to describe it uniquely. Five drives are needed lowest.
Merit
* It has the highest trouble resistance at all RAID level.
Weak point
* The humming code calculation cost is enormous.
* The use efficiency of the disk is extremely bad (Correspond RAID 1 for 1
bitECC in six composition).
RAID 3: Every bit/byte special parity drive RAID 3 RAID 3
RAID 3 changes the error correcting code of RAID 2 to parity by the exclusive-OR, and is a decrease
of the operation cost.
It is composed of three drives or more, and one is allocated in the error correcting
code, and data is recorded in the plural of the remainder.
RAID 3 took the place of RAID 5. You may think that it is impossible to be going to obtain the equipment corresponding to RAID 3 in the future. There is no reason to use RAID 3 with the personal computer and the server any longer though RAID 3 might be still used from the sequential access almost of the access in the video edit equipment.
Merit
* Because parity is used as a correction sign, the calculation cost is
lower than RAID 2.
* The uselessness of the capacity of the disk is suppressed by the minimum because it
can secure the capacity of -1 composition drive number.
Weak point
* It is inefficient because it accesses it by each bit/byte of I/O.
* It becomes a bottleneck when the parity drive processes writing.
RAID 4: Special parity drive RAID 4 RAID 4 with block unit
RAID 4 expands the unit of I/O of RAID 3 to the block, and is a measurement of the improvement of the I/O efficiency. RAID 4 inferior to RAID 5 on the performance side is going out of use.
Merit
* I/O that is more high-speed than RAID 3 can be hoped for because the
unit of the access is a block.
Weak point
* The parity drive can become a bottleneck when writing is processed. (The solution
to this is RAID5. )
RAID 5: Parity decentralization record RAID 5 RAID 5 with block unit
RAID 5 distributes to two or more hard disks with the error correcting code data and has evaded the bottleneck of RAID 3 and RAID 4 by recording. Four for which three drives or more are necessary. It is excellent method in the balance of the speed, trouble resistance, and the disk use efficiency that can be said now, "Leading part" of various RAID.
Merit
* It becomes a bottleneck, and a special parity drive such as RAID 3 and RAID 4
doesn't exist.
* Speed-up can be expected by increasing of the number of the drive.
Weak point
* Writing is low-speed though reading is high-speed (Especially, when the
number of the drive is few).
* The restoration work at the failure is slow.
* When two drives or more break down at the same time, it is not possible to
recover. (The solution to this is RAID 6. )
Combination that uses RAID 5
The weak point can be covered by combining with other RAID as well as RAID 0+1 and 1+0 when there is dissatisfaction in the speed side and trouble resistance, etc. in RAID 5.
RAID 5+0 and RAID 0+5
Hard disks of the same number as the number used are added to improve the speed of RAID 5, it combines with RAID 0 or it is effective to increase the server and to distribute the load. When RAID 5+0 and RAID 0+5 are composed, should six drives or more.
The name is changed which RAID 5 or RAID 0 to be done early as well as RAID 1+0 and 0+1. It can be said that RAID 5+0 that does striping by the set of RAID 5 is more excellent in the following reason.
* Excellent five in tolerance to drive breakdown.
* The effect of the improvement at the writing speed can be stronger and it expect it by adjusting RAID 0 that assumes the high speed of reading and writing to be an advantage to the outside.
RAID 5+1 and RAID 1+5
When strong trouble resistance more than RAID 6 is demanded, this combination becomes choices. Six drives or more are needed as well as RAID 5+0 and RAID 0+5.
Six with trouble resistance that RAID 1+5 is stronger though the breakdown simultaneously until RAID 5+1, RAID 1+5, and driving three can be endured.
It excels also in maintenance. The inside is seven necessary for exchanging all drives by some reasons replacing the one of the mirror when causing it disk at a time, exchanging the remainder after Ribild, and without stopping the device according to a handy procedure named Ribild ..again.. where the trouble resistance of RAID 5 remains. can the completion of the exchange and this work
RAID 5+5
RAID 5+5 that unites RAID 5 by RAID 5 is thought. Nine drives or more are required for
this composition.
The breakdown until driving three can be endured at the same time as well as RAID 5+1 and 1+5 and they are exceeded by the disk use efficiency. Eight that becomes the middle level of RAID 5+1 and 1+5 in trouble resistance.
Similarly, RAID 5+5+5 being made a dimension three and RAID 5+5+5+5, etc. made 4th Dimension are thought, too. RAID 5+5+5 is seven RAID 5+5+5+5 is nine from respect of needed number of drives and disk efficiency not practicable though can endure breakdowns simultaneously up to 15.
RAID 6: Block unit and two or more parity decentralization record
RAID 6 RAID 6
RAID 6 records two paritys. Even if two drives break down at the same time being possible to correspond to a double trouble by doubly making the parity data, it is restorable. Four drives or more are needed.
Merit
* There is trouble resistance that has the merit equal with RAID 5, and
is higher.
Weak point
* An initial investment is large. (However, a long-term operation cost is
neither RAID 5 nor a great difference. )
* The writing speed decreases more than RAID 5 to generate parity doubly.
* When three drives or more break down at the same time, it is not
possible to recover.
The calculation of parity and the recovery of data are done with software by using the EVENODD algorithm in RAID6 of EMC CLARiX/CLARiiON in XOR instruction the X86 processor.
In a large-scale system, multiple RAID that uses RAID 6 is thought as well as RAID 5.
RAID 6+0 and RAID 0+6
After the trouble resistance of RAID 6 is made the best use of, speed-up is aimed at. Eight drives or more can be required, and the breakdown until driving two be endured.
RAID 6+1 and RAID 1+6
Trouble resistance is very strong. Eight drives or more can be required, and
the breakdown until driving five be endured.
RAID 6+5 and RAID 5+6
Strong trouble resistance and the speed are had both. 12 drives or more can be required,
and the breakdown until driving five be endured.
RAID 6+6
Extremely strong trouble resistance. 16 drives or more can be required, and
the breakdown until driving eight be endured.
RAID News
- RAID原理:RAID Level・RAID5・RAIDデータ復旧 ...
- サーバ復旧、RAID復旧は信頼出来るデータ復旧テクニカルセンターにお任せください。 ... RAID 1には最低2ドライブが必要です。 ... RAID 1ではハードディスク台数が増えれば増えるほどハードディスクの利用効率が悪くなります。 ...
- RAID用語集 - TYAN
- RAID機器からのエラーメッセージの詳細はコチラ. RAIDが壊れたときに、絶対やってはいけない３つのこと ... RAID用語集の編集は、管理者権限を持つユーザーのみが可能です。 RAID復旧|データ復旧|ハードディスク ...
- RAIDデータ復旧・RAID5データ復旧:サーバー復元・復活・修復の ...
- サーバ復旧、RAID復旧は信頼出来るデータ復旧テクニカルセンターにお任せください。 ... 対応可能なRAIDシステムの障害内容 ... 変形されたRAID情報保によったデータにアクセス不可 - RAID再構成リビルド時システム ダウン、 ...