CN102750111A

CN102750111A - Method for storing redundant array of independent disk (RAID) bitmaps

Info

Publication number: CN102750111A
Application number: CN2012101721268A
Authority: CN
Inventors: 古世磊
Original assignee: Inspur Electronic Information Industry Co Ltd
Current assignee: Inspur Electronic Information Industry Co Ltd
Priority date: 2012-05-30
Filing date: 2012-05-30
Publication date: 2012-10-24
Anticipated expiration: 2032-05-30
Also published as: CN102750111B

Abstract

The invention discloses a method for storing redundant array of independent disk (RAID) bitmaps and belongs to the field of a computer system and storage. Disk data are correspondingly divided into several equal parts according to the number of the bitmaps in a disk array, and each bitmap only corresponds to one part of data region and is unrelated with other data regions. The bitmaps are plated in the back of the shown data regions. The method for storing the RAID bitmaps has the advantages of enabling the read-write performance of disks to be in the best state and improving the disk performance.

Description

A kind of method of RAID bitmap storage

Technical field

The present invention relates to a kind of computer system and field of storage, specifically a kind of method of RAID bitmap storage.

Background technology

At RAID1, RAID5 etc. exist in the magnetic battle array structure of data redundancy, and the integrality of data requires the redundant data and the information of raw data will guarantee consistent.The write operation of magnetic battle array is a process, takes place in this process possibly cause redundant data and raw data inconsistent unusually, and for example disk is changed, and writes the dish failure, system's power down etc.

So, successfully writing data, whole band should be consistent, does not have successfully (or not exclusively) but write, it is inconsistent that strip data is likely.After system re-powered, perhaps the zone maybe be inconsistent just to need a kind of mechanism to come those bands.If there is not recorded information, it can only suppose that all data all maybe be inconsistent, carries out the consistance of all data blocks and recovers.For RAID5, the effect that recomputates all bands exactly with; For RAID1, be exactly again that synchronous all data blocks are to mirror drive.This global data can cause system boot process very very long (being in inconsistent state after perhaps powering on regard to whole laissez-faire magnetic battle array) synchronously.

If NVRAM (Nonvolatile memory) is arranged, at first, in NVRAM, preserve writing position information to a data block write operation, after write operation finishes, this information of deletion in NVRAM.Like this, when system boot, from NARAM, read the writing position information of preservation, these positional informations have write down the data block that does not successfully write, and system only need carry out getting final product synchronously to these data blocks, and magnetic battle array total data synchronously.The magnetic battle array that we use software RAID to make up at present, this has just drawn the bitmap file, is appreciated that the effect of bitmap file is equal to the effect of the NVRAM of above-mentioned professional magnetic battle array (above-mentioned scene is the little aspect that NVRAM uses certainly); Behind the power-down rebooting; Can start synchronizing thread, the bitmap file is arranged, can judge whether set through the bitmap file; If there is not set; Just explain whole zone power down that constantly, do not have the data of writing, it is just passable directly to skip this zone.

At present the method to the bitmap file backup is, when carrying out write operation, system can be set to 1 with pairing bitmap bits, writes in all disks of magnetic battle array reserved area synchronously and goes.Have only after the bitmap file Flushing success in the disk, just begin a normal data I write operation.

Since each band all to wait corresponding bitmap write back earlier could be real behind the disk the processing of writing, and the performance of disk and internal memory etc. are too wide in the gap, and are very big to performance loss, so consider and get around disk, so will change a kind of backup mode.

Summary of the invention

Technical assignment of the present invention provides and lets the readwrite performance of disk reach better state, can improve the method for a kind of RAID bitmap storage of disk performance.

Technical assignment of the present invention realizes several bitmaps being arranged just the corresponding several five equilibriums of data in magnetic disk in disk array by following mode, the just corresponding wherein data area of a five equilibrium of each bitmap is irrelevant with other data area; Bitmap is placed on the back of represented data area.

Disk is made up of the transmission time between seek time, rotational delay time and sector and the disk adapter interface service time; Wherein seek time accounts for about 70% to 80% of T.T., so seek time is crucial for some IO requests, particularly writes request; To constantly revise superblock and bitmap; Because these metadata informations are stored in the disk end at present, magnetic head wants tracking to terminal, and the distance that moves can be excessive; Staying a metadata reserved area in the centre of disk array is bitmap, and the disk end blocks just passable equally.

The position that logic is corresponding be to be in different cylinders on actual physical address, but the data corresponding relationship is consistent.Like this, for the operation of first half, need not move to magnetic head the end of disk, it is just passable to move on to position intermediate, has reduced the displacement of magnetic head, makes seek time reduce.And for the data of latter half, total displacement remains unchanged.But from whole magnetic disk, for the operation of metadata, total seek time can reduce much relatively originally, thereby can improve performance.

A position in the bitmap is the state in a zone of expression, if this zone is more little, just explains that this bitmap is accurate more, and synchronizing process just can be skipped more zone so, has reduced synchronous process.So a bitmap is the division of comparison coarseness, if think to show more accurately whether synchronous state, just need thinner division, also just need more a plurality of bitmap pages.

More bitmap is exactly that thinner division has several bitmaps just several five equilibriums of data in magnetic disk, and each bitmap is the data area of a corresponding part just, and is irrelevant with other data area.For example, two pages are arranged, modes of emplacement originally is all to be placed on the bottom to two bitmap pages, and new mode is to be divided into two parts to the data area of disk array earlier, and a bitmap is placed on the back of represented data area.In like manner, the modes of emplacement of four bitmaps is similar.Reason pushing away successively, and in the time of a plurality of again page, modes of emplacement is identical.

In addition, for RAID5, allow a dish and lost efficacy, so, can leave metadata simultaneously on two dishes according to this principle.If because a dish in these two dishes lost efficacy, metadata can also find from another dish, if these two dishes all lost efficacy, whole RAID 5 had just lost efficacy, and metadata is nonsensical.Can certainly select to put several dishes more.If be placed on two dishes, when rewriting metadata, just needn't all participate in by each disk, reduce the IO request of total system, thereby improved performance., so just can be stored in metadata on three dishes because can allow the inefficacy of two dishes for RAID6.For RAID1, at least also can be stored on two dishes.

When dish lost efficacy,, when adding HotSpare disk, just need write this HotSpare disk to metadata if this dish is a disk of depositing metadata; If not the disk of depositing metadata, can write metadata toward HotSpare disk.

Be based on the consideration of performance, let readwrite performance reach better state.Its core is to leave on the rational position with reasonable manner.The reasonable manner of mentioning is meant on the disk of only writing a part.The rational position of mentioning is meant the position that can influence readwrite performance.

RAID is the abbreviation of English Redundant Arrays of Inexpensive Disks, is translated into disk array.

The method of a kind of RAID bitmap storage of the present invention has the following advantages: let the readwrite performance of disk reach better state, improve disk performance; Thereby, have good value for applications.

Description of drawings

Below in conjunction with accompanying drawing the present invention is further specified.

Accompanying drawing 1, Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6 are the data layout figure of disk array of the method for a kind of RAID bitmap storage.

Embodiment

Explanation at length below with reference to Figure of description and specific embodiment the method for a kind of RAID bitmap storage of the present invention being done.

Embodiment:

The method of a kind of RAID bitmap storage of the present invention has several bitmaps just the corresponding several five equilibriums of data in magnetic disk in disk array, the just corresponding wherein data area of a five equilibrium of each bitmap is irrelevant with other data area; Bitmap is placed on the back of represented data area.

Disk is made up of the transmission time between seek time, rotational delay time and sector and the disk adapter interface service time; Wherein seek time accounts for about 70% to 80% of T.T., so seek time is crucial for some IO requests, particularly writes request; Will constantly revise superblock and bitmap, because these metadata informations are stored in the disk end at present, magnetic head wants tracking to terminal; The distance that moves can be excessive, and like Fig. 1, staying a metadata reserved area in the centre of disk array is bitmap; The disk end blocks just passable, as shown in Figure 2 equally;

What showed among Fig. 2 is the corresponding position of logic, be to be in different cylinders on actual physical address, but the data corresponding relationship is consistent.Like this, for the operation of first half, need not move to magnetic head the end of disk, it is just passable to move on to position intermediate, has reduced the displacement of magnetic head, makes seek time reduce.And for the data of latter half, total displacement remains unchanged.But from whole magnetic disk, for the operation of metadata, the original relatively Fig. 1 of total seek time can reduce a lot, thereby can improve performance.

More bitmap is exactly that thinner division has several bitmaps just several five equilibriums of data in magnetic disk, and each bitmap is the data area of a corresponding part just, and is irrelevant with other data area.For example, two pages are arranged, modes of emplacement originally is shown in Figure 3 to be all to be placed on the bottom to two bitmap pages, and new mode Fig. 4 is divided into two parts to the data area of disk array earlier, and a bitmap is placed on the back of represented data area.In like manner, like Fig. 5 and shown in Figure 6, the modes of emplacement of four bitmaps is similar.Reason pushing away successively, and in the time of a plurality of again page, modes of emplacement is identical.

The method of a kind of RAID bitmap storage of the present invention except that the described technical characterictic of instructions, is the known technology of those skilled in the art.

Claims

1. the method for a RAID bitmap storage is characterized in that in disk array, having several bitmaps just the corresponding several five equilibriums of data in magnetic disk, and the just corresponding wherein data area of a five equilibrium of each bitmap is irrelevant with other data area; Bitmap is placed on the back of represented data area.