KR20120094782A - Storage system with solid state disk - Google Patents

Abstract

PURPOSE: A SSD(Solid State Drive) based storage apparatus system is provided to improve the input and output performance of a storage apparatus system by applying a source code to a file system level or an apparatus driver level. CONSTITUTION: A disk arrangement(120) is constituted by connecting disks from each other. A central control unit(110) distributes and stores data according to data input/output request by encoding the data based on an erasure code. The disk is realized a semi-conductor storage apparatus like an SSD. The central control unit measures the writing and deleting calculation number of the disk. When the number of the calculation is over a threshold value, the central control unit outputs a disk exchange alarm message.

Description

SD-based storage system {STORAGE SYSTEM WITH SOLID STATE DISK}

The present invention relates to an SSD-based storage system applying erasure code for improving reliability and input / output performance.

In general, a storage system such as a redundant array of independent disks (RAID) system in which a plurality of disks are connected in an array structure is used in a large server-class environment. RAID system is a technology that connects several disks in an array structure and makes them appear to the user as one disk while increasing parallelism of access.

Such a RAID system can do three things: First, striping techniques on multiple disks can improve performance in such a way that multiple disks can operate simultaneously to supply or accept a single data flow. Second, there is a mirroring technique for copying data existing on multiple disks, thereby reducing the risk associated with the destruction of a single disk. Third, there is a parity technique for recovering data of a single damaged disk. When the parity technique is performed, the performance of the storage device is changed according to the erase code applied. Therefore, the effect of the erase code on the storage device should be considered.

Although hard disks are mainly used in RAID systems, hard disks have additional mechanical operations, which increases power consumption as the number of disks increases, and data loss is more likely due to external shocks. In addition, the hard disk has a disadvantage that a lot of noise and heat generation easily.

Recently, unlike a hard disk, a solid state disk (SSD), which has no mechanical operation, low power consumption, and strong external shock, has been used. SSD is composed of a number of NAND Flash memory, and has the following two characteristics. First, SSDs consider a limit on the number of block updates called wear leveling. It has a lifetime of about 100,000 times in a single level cell and about 10,000 times in a multi level cell. SSDs have a lower lifetime than single-level cells because they consist of multi-level cells. Second, unlike conventional hard disks, SSD can be divided into three operations: read / write / delete, read / write is performed in units of pages (512B ~ 4KB), and delete operations are blocks (16KB ~ 256KB). Performed in units.

As described above, the (590) SSD has the advantages of high performance and low power, and thus has been in the spotlight as a storage device to replace a hard disk, ranging from a portable device to a high performance system such as an enterprise server.

A parity-based hard disk RAID system used in the existing mass storage system and a Diff-RAID (Differential RAID) system studied for increasing the lifespan of an SSD will be described.

As a representative parity RAID system based on a hard disk, as shown in FIG. 1, there are a RAID-5 system using a method of distributing and storing parity information among all disks in a group and a RAID-6 system using a two-dimensional parity method.

A RAID-5 system can only be configured with at least three disks, usually five or more disks. It is also called a parity RAID system, because when a disk fails, data can be recovered using parity.

Storage systems, such as RAID-5 systems, operate independently of each other, so they can handle multiple small application requests simultaneously, and when a large input / output request needs to be processed, all disks are activated to handle one operation. can do.

However, a RAID system based on a hard disk as described above requires a lot of power consumed due to the mechanical characteristics of the hard disk, and has a problem in that performance is degraded when space of the disk is cooperative.

The RAID-6 system uses two parities so that space efficiency is lower than RAID-5. However, even if two disks fail at the same time, the data is good. Therefore, the RAID-6 system is composed of at least four disks as shown in FIG.

Therefore, the advantage of RAID-6 system is that the data reliability is higher than the RAID-5 system. The disadvantage of RAID-6 systems, however, is that RAID-5 systems only need to generate one parity, while RAID-6 systems only need to generate two parities. Complicated computations degrade performance over RAID-5.

SSDs are limited to 10,000 erase operations. As a result, data loss and disk failure due to the deterioration of SSD's lifespan are fatal. Recently, the research of SSD-based RAID system has been studied the Diff-RAID (Differential RAID) method considering the life of the SSD.

3 illustrates a parity distribution method of a Diff-RAID system. RAID system consists of four SSDs (SSD0 ~ SSD3), and each SSD is stored with parity distribution at a rate of 70%, 10%, 10%, and 10%. Since SSD0 has the highest parity dispersion rate of 70%, write / erase operations are concentrated on SSD0. Therefore, SSD0 has a lower life rate than the remaining SSDs SSD1 to SSD3.

If the disk fails due to the reduced life of SSD0, one SSD (SSD4) is added to maintain the existing RAID system. SSD1 will show a system where the parity dispersion rate will increase from 10% to 70% instead of SSD0, and the write / erase operations of data will be concentrated on SSD1.

The above-described Diff-RAID system considers the lifespan of a disk. The Diff-RAID system manages the lifespan of an SSD-based RAID system efficiently, but does not consider the effect of the erase code operation on the SSD.

The present invention has been made to solve the above problems, the present invention is to provide an SSD-based storage system that applies the erase code to the file system level or device driver level.

In addition, the present invention is to provide an SSD-based storage system that minimizes the deterioration of the life of the disk by applying a data allocation technique to centrally store the parity block with a frequent update to a single disk.

In the SSD-based storage system according to an embodiment of the present invention for realizing the above object is a disk array consisting of a plurality of disks connected in an array, and using an erase code according to the data input and output request to the plurality of disks And a central control unit for encoding and distributing data on the plurality of disks.

In addition, the disk is implemented as a semiconductor storage device such as a solid state device (SSD).

The central controller may further include a file system for dividing input data into a data file and a parity file using an erase code, a virtual file system connected to the file system and receiving the divided file, and a virtual file system. A device driver for receiving the received divided file, and a block allocation management unit for classifying into a data area and a parity area based on the metadata information of the divided files, and storing them in the disk array.

The central control unit may further include: a virtual file system for receiving data on request; a device driver for encoding data transmitted from the virtual file system through an erase code; and dividing the encoded data into a data area and a parity area. And a block allocation management unit to sequentially store the disk array.

The present invention relates to at least one embodiment of the present invention, which is configured as described above, replaces a hard disk with a solid state disk (SSD) and applies an erase code to a file system level or a device driver level, thereby providing an input / output performance of a storage system. Can improve.

In addition, since the present invention performs data allocation so that the parity block is centrally stored in one disk, reliability of the disk can be improved by reducing the lifespan of the disk.

1 is a configuration of a typical RAID-5 system.
2 is a configuration of a conventional hard disk-based RAID-6 system.
3 is a chart showing the life rate according to the parity distribution of the conventional differential RAID.
4 is an SSD-based RAID-6 system to which the Reed-Solomon code according to the present invention is applied.
5 is an SSD-based RAID-6 system to which the EVENODD code according to the present invention is applied.
6 is a SSD-based RAID-6 system to which the Liberation code according to the present invention is applied.
7 is a block diagram illustrating a structure of an SSD-based storage system according to an embodiment of the present invention.
8 is a block diagram illustrating a structure of an SSD-based storage system according to another embodiment of the present invention.

Hereinafter, exemplary embodiments will be described in more detail with reference to the accompanying drawings.

The present invention proposes a RAID system in which a hard disk is replaced with a solid state device (SSD) in a conventional hard disk based RAID (Redundant Array of Independent Disks) system used as a mass storage system. In addition, the RAID system used in the present invention is a RAID-6 system using two parity, and an erasure code is applied to the RAID-6 system.

Since erasure codes vary depending on the encoding method, performance and power consumption may vary depending on the impact on the SSD. Therefore, an erase code suitable for the SSD environment should be used.

The erase codes applicable to the SSD environment are as follows.

First, Reed-Solomon code is the most powerful and universal coding technique. Reed-Solomon coding techniques can recover multiple errors into complex matrix products. As shown in FIG. 4, the dispersion matrix B is composed of a matrix I and one matrix, and may generate data and coding information by a multiplication operation with the data D.

However, Reed-Solomon code operations based on distributed matrices are expensive to compute. Therefore, in the SSD-based environment, as the number of bit updates increases, the impact on SSD wear may increase, thereby reducing the overall performance of the system.

Second, the EVENODD code scheme has two independent parity columns as shown in FIG. 5, and the two parity columns can be calculated as checksums of data, respectively.

모든 열의 XOR 계산 된다.
The first parity (C _0,0 ? C _0,3 ) is a value obtained by XORing each row of the disk matrix and can be expressed as the following equation. Looking at Equation 1, j is a column

The XOR of all columns is calculated.

The second parity is the result of XOR operation of the special diagonal pattern of the EVENODD code and Syndrome together, which can be expressed as Equation 2.

With the above calculation method, the effect of EVENODD code on SSD is simple XOR operation and the operation is simpler than Reed-Solomon code.

Third, the Liberation code technique is calculated very similarly to the Reed-Solomon code. Unlike the dispersion matrix used in the Reed-Solomon code, the Liberation code technique uses a coding technique that divides the binary bit matrix into data.

In general, the Liberation code is composed of k data devices, m coding devices, and bits fixed to each w as shown in FIG. Also, because all elements are composed of binary, it is called BDM (Binary Distribution Matrix), and it can operate faster than Reed-Solomon coding.

Liberation code is very similar to Reed-Solomon code, but computed using binary matrices, resulting in low computational complexity and fast operation. Therefore, when applied to SSD-based environment, it has less effect on SSD wear and SSD lifespan due to lower bit update times than Reed-Solomon.

7 is a block diagram illustrating a structure of an SSD-based storage system according to an embodiment of the present invention. In the present embodiment, a write operation is performed when an erasure code is applied to a file system.

The SSD-based storage system 100 includes a central controller 110 and a disk array 120.

The central controller 110 performs a process corresponding to the request when a system call is requested by an external input or an interrupt is requested from hardware in the storage system. The central controller 110 encodes or decodes data using an erase code when inputting / outputting data to a plurality of disks constituting the disk array 120.

The central control unit 110 includes a file system 111, a virtual file system (VFS) 112, a device driver 113, and a block allocation management unit 114.

The file system 111 services file creation and access control, directory management, and the like. When the write request is received, the file system 111 divides one data into data files (files 0 to 3) and parity files (files p and file q) by using an erase code. In other words, the file system 111 encodes data through an erase code to generate the file unit.

The virtual file system 112 serves to provide an interface to the file system when a specific event comes from the user. The virtual file system 112 transmits the divided files to the device driver 113.

The device driver 113 is composed of drivers for driving peripheral devices such as disks, terminals, and network cards. The device driver 113 controls data input / output with the disk array 120, and transmits the divided files received through the virtual file system 112 to the block allocation manager 114.

The block allocation manager 114 includes a data area and a parity area therein. The block allocation manager 114 classifies the data area and the parity area in sector units 512B to 4KB based on metadata information of the divided files and transmits the data area and the parity area to the disk array 120. In other words, the block allocation management unit 114 classifies the data with metadata information of the data encoded by the file system 111. Sectors transferred to the disk array 120 are sequentially stored in each page of the disk array 120 in page units (512B to 4KB).

8 is a block diagram illustrating a structure of an SSD-based storage system according to another exemplary embodiment of the present invention. In the present embodiment, a write operation is performed by applying an erase code to a device drive.

Referring to FIG. 8, if there is a write request, the virtual file system 112 transmits the data to the device driver 113 when it receives data.

The device driver 113 encodes the data transmitted from the virtual file system 112 in sector units using the erase code. Here, the erase code is any one of a Reed-Solomon code, an EVENODD code, and a Liberation code. The device driver 113 transmits the encoded data to the block allocation manager 114.

The block allocation manager 114 divides the encoded data transmitted from the device driver 113 into a data area and a parity area, and sequentially stores the encoded data transmitted to the disks of the disk array 120.

The block allocation manager 114 may classify the data by metadata information of the encoded data. For example, if the metadata information of the encoded data is a data block, the block allocation management unit 114 classifies the encoded data into a data area in the data block allocation management unit 114 and stores the data disks SSD0 to the data array of the disk array 120. SSDn). If the metadata information of the encoded data is parity block_p or parity block_q, the block allocation management unit 114 classifies the encoded data into a parity area in the block allocation management unit 114, and then the disk array 120. To parity disks (SSDp and SSDq).

In addition, the block allocation management unit 114 concentrates and stores the parity blocks with frequent updates to one disk. For example, when the disk array 120 is configured with four data disks and two parity disks, the block allocation manager 114 distributes and stores the data into four data disks, and parity p and parity q are two. Stored centrally on one of the parity disks. Therefore, it is possible to minimize the deterioration of the life of the other one of the two parity disks.

The disk array 120 has at least two data disks and two parity disks connected in an array. Here, the disk is implemented as a semiconductor storage device such as a solid state disk (SSD). Each disk measures the number of write / delete operations, and notifies the central controller 110 that the disk needs to be replaced if the measured number of write / delete operations is greater than or equal to a threshold. The central control unit 110 outputs a disk replacement notification message. For example, since the SSD lifespan is limited to 10,000 times, when the number of SSD write / delete operations reaches 9,000 times, the SSD detects this and the central controller 110 outputs a disk replacement notification for the user to recognize. The disk replacement notification is output as a message and / or a warning sound.

The disk array 130 may further include a disk controller for controlling at least one or more disk operations among the disks constituting the disk array 120 according to a control command. The disk controller detects a state of each of the disks to determine a connection state, and transfers the identified disk state information to the central controller 110. The disk controller measures the number of write / delete operations of each disk, and transmits information indicating whether the disk needs to be replaced to the central controller 110 when the measured number of write / delete operations is equal to or greater than a threshold value.

100: storage system 110: central control unit
111: File System 112: Virtual File System
113: device driver 114: block allocation management unit
120: disk array

Claims (8)

A disk array in which a plurality of disks are connected and arranged in an array structure;
And a central controller configured to encode data by using an erase code according to data input / output requests to the plurality of disks, and to distribute the data to the plurality of disks. The method of claim 1, wherein the disk,
SSD-based storage system, characterized in that implemented as a semiconductor storage device, such as a solid state drive (SSD). The method of claim 1, wherein the central control unit,
And measuring a number of write / erase operations of the disk and outputting a disk replacement notification message if the measured number of operations is greater than or equal to a threshold value. The method of claim 1, wherein the central control unit,
A file system for dividing input data into a data file and a parity file using an erase code;
A virtual file system connected to the file system and receiving the divided file;
A device driver for receiving the divided file received through the virtual file system;
And a block allocation manager configured to classify the data file into a data area and a parity area based on the metadata information of the divided files and to store them in the disk array. The method of claim 4, wherein the block allocation management unit,
SSD-based storage system, characterized in that divided into data area and parity area by sector based on the metadata information of the divided files. The method of claim 1, wherein the central control unit,
A virtual file system that receives data on request,
A device driver for encoding data transmitted from the virtual file system through an erase code;
And a block allocation management unit for dividing the encoded data into a data area and a parity area and sequentially storing the encoded data in the disk array. The method according to claim 4 or 6, wherein the block allocation management unit,
SSD-based storage system, characterized in that to store the parity block with a high number of updates to one disk. The method of claim 1, wherein the erase code,
SSD-based storage system, characterized in that any one of the Reed-Solmon code, EVENODD code, and Liberation code.

Images