KR100974514B1 - A sequential prefetching method in computer system - Google Patents

Abstract

The present invention relates to a sequential prefetching method in a disk array composed of a plurality of disks by a striping technique such as a RAID. The present invention relates to a first process of receiving a read command from a process, and an initial step indicated by the received read command. A second process of determining a prefetching size and successive blocks to be prefetched, a third process of checking whether the successive blocks are included in different strips, and if the successive blocks are included in different strips, A fourth process of reducing the initial prefetching size by the number of blocks included in the strip to which the start block belongs among consecutive blocks; and a fifth process of generating one disk command according to the reduced initial prefetching size. Characterized in that it comprises a. The present invention described above has the effect of preventing performance degradation due to loss of independence in a disk storage device using a striping technique.

Storage Management, Prefetching, Disk Arrays

Description

Sequential prefetching method in computer system {A SEQUENTIAL PREFETCHING METHOD IN COMPUTER SYSTEM}

The present invention relates to a prefetching method in a computer system, and more particularly, to a sequential prefetching method in a disk array composed of a plurality of disks by a striping technique such as a RAID (Redundant Array of Independent Disks).

Prefetching technology is needed to reduce the latency between the processor and main memory, which dates back to the 1960s when the processor reads a number of words per cache line in advance. In addition, many prefetching techniques have been developed in current processors, and this prefetching technique has been applied to disk prefetching, which is read in advance from disk to memory. As a pre-fetching technique in the processor and other features of disk storage, many prefetching techniques have been studied. Disk prefetching increases disk performance by lowering disk read costs or increases overall system performance by overlapping operations and disk I / O.

In general, the most widely used prefetching technique in disk systems is sequential prefetching. This sequential prefetching method is based on the fact that many disk reads occupy a large portion of a large file read or a sequential read. However, such a sequential prefetching method is designed for a single disk, and there is no changed sequential prefetching method specialized for a disk array using a plurality of disks. In addition, the prior arts only consider the parallelism and the load balance in prefetching the disk array, but do not consider the independence of the disks.

For example, the conventional sequential prefetching method does not consider the position of a physical disk at the file level prefetching. That is, when the positions of blocks to be read by one prefetching are included in two or more strips, the actual load burden and the processing speed are slowed down by generating two instructions when they can be performed by one disk instruction. This happens.

Accordingly, an object of the present invention is to propose a sequential prefetching method that can prevent performance degradation by considering independence in prefetching in a computer storage device using a plurality of disks.

In order to achieve the above, the present invention provides a sequential prefetching method in a computer storage device using a plurality of disks, the method comprising: a first step of generating a read command according to a predetermined process; A second process of determining the start and prefetching sizes of consecutive blocks to fetch, a third process of checking whether the consecutive blocks are included in two or more strips, and if the blocks are included in two or more strips, A fourth step of reducing the prefetching size such that blocks belonging to the prefetching request are included in only one strip; and a fifth step of generating a disk command according to the start of the contiguous blocks to be prefetched and the reduced prefetching size. It is characterized by.

The fourth process is a sixth process of setting a start block number (S) and an end block number (E) of the file level prefetching data, and a seventh process of assigning 'S + 1' to the variable 'P'. And an eighth process of determining a strip number including the start block number, and a strip number in which the current block of the file is stored is the same as the strip number, and 'P' is the end block number (E). If the ninth process and the condition of the ninth process is false, the end of the block to be prefetched is set to 'P-1' and the size of the set prefetching is determined to be 'P-S'. It characterized in that it comprises a tenth process.

If the condition of the ninth process is true, the method further includes a tenth process of increasing 'P' by '1' and repeating the eighth process.

The strip number may be determined as an integer value of a value obtained by dividing a 'physical block number in which corresponding data is stored' by 'number of blocks per strip'.

The present invention has the effect of improving the performance of multiple simultaneous reads of a disk storage device using a striping technique.

In addition, the present invention is a general-purpose cache management method developed without any assumption, and there is an effect that can be easily applied to existing small to large computers or storage devices without any hardware change.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a detailed description of preferred embodiments of the present invention will be given with reference to the accompanying drawings. It should be noted that the same components in the figures represent the same numerals wherever possible. Specific details are set forth in the following description, which is provided to aid a more general understanding of the invention. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

The present invention proposes a prefetching method for preventing performance degradation due to independence in a disk array composed of a plurality of disks by a striping technique such as a RAID (Redundant Array of Independent Disks).

First, a general sequential prefetching method will be described in order to help understanding of the present invention, and then a sequential prefetching method and apparatus according to the present invention will be described.

First, the term will be described before the description.

Prefetching refers to a method that reads data on the disk and stores it in the system's memory, called cache, in advance before any program requesting the data on the disk. It is implemented as a computer program to operate.

In this prefetching technique, when a program reads data in an arbitrary file sequentially, if the program requires block b, the general sequential prefetching reads block b + p from block b into storage and caches it. If the block to which the program approaches is sequential, the size p of the prefetching increases linearly or exponentially from the preset minimum value to the preset maximum value every prefetching.

In addition, the prefetching block means data in units of physical blocks required for prefetching.

Next, a structure of a disk array using a plurality of disks will be described.

1 is a diagram illustrating an example of a RAID (Redundant Array of Independent Disks) structure having a plurality of disks. The raid has a structure of dividing and storing data into a plurality of disks as one virtual disk.

Referring to Figure 1, the rade is an example consisting of four disks (1, 2, 3, 4). The stripe 30 in the raid consists of strips 20 of disks 1, 2, 3, and 4, and the strips 20 are continuous blocks 10 in the disk. )

One read request generated by a typical sequential prefetching involves several consecutive blocks. However, if a storage device to which sequential prefetching is applied is a disk array using a striping technique, a plurality of blocks belonging to one read request may belong to one strip and may be spread over two or more strips or scattered. . A read request that spans more than one strip generates two disk commands compared to a read request that corresponds to one strip, resulting in a drastic performance degradation. This degradation problem is referred to as loss of independence below. Loss of independence especially occurs when multiple read requests are entered at the same time.

FIG. 2 is a diagram illustrating an example of a general sequential prefetching method, and more easily described the loss of independence.

Referring to FIG. 2, there are three independent processes 40, 41, and 42, and each process independently reads four consecutive blocks. For example, the first process 40 generates read commands 50, 51, 52, 53 for four consecutive blocks. By sequential prefetching, reads to the blocks may be replaced by a single disk prefetching request 60 by the prefetcher 55. Here, if four blocks belonging to the four read commands 50, 51, 52, 53 belong to two strips, the disk prefetching read request 60 is a block on two disks 80, 81. Disk commands 70A and 70B are generated for the respective devices. Because two consecutive strips must be divided into two disks.

An example in which one disk prefetch request is generated with two disk commands is explained using one embodiment shown in FIG. 1, where one disk prefetch request 60 is divided into four consecutive blocks 6 of FIG. If blocks 7, 8, and 9 are read, blocks 6 and 7 belong to disk 1, and blocks 8 and 9 belong to disk 2. Therefore, the disk prefetching request 60 is generated by a disk command 70A reading blocks 6 through 7 and a disk command 70B reading blocks 8 through 9.

In order to send a disk command to disks 80 and 81 for the disk prefetch request A 60, a disk array administrator (DAA) 90 sends the disk prefetch request A 60 to it. It is divided into two Disk Commands (DC) 70A and 70B for the disks 80 and 81. That is, since one prefetching read request 60 generated by the first process 40 relates to blocks spanning the two disks 80 and 81, the two disks 80 and 81 are eventually run. The other two processes 41 and 42 also generate second and third read requests 61 and 62 at the same time as the first process 40, and the read request 61 of the second process 41 If two disks 81 and 82 are to be divided, another two disk commands 71A and 71B are generated, and the third disk request 62 similarly issues another two disk commands 72A and 72B. When generated, three data read requests 60, 61, 62 eventually generate six disk commands 70A, 70B, 71A, 71B, 72A, 72B.

In this case, each disk must handle two separate commands. However, if each disk prefetch read request (60, 61, 62) is limited to one strip, each read request will each generate one disk command. In this case, three disk commands are generated. In the example of FIG. 2 which causes loss of independence, a small amount of disk commands are generated as compared with six disk commands generated, thereby further improving performance. On the other hand, if consecutive blocks to be read by one read request span two or more strips, many disk commands are generated, and thus disk I / O performance is reduced due to loss of independence.

In a conventional sequential prefetching method based on file-level sequential block access, disk requests are spread across multiple disks, resulting in a loss of independence for concurrent reads with multiple processes.

Next, the sequential prefetching at the file level without considering the physical location of the prefetching block will be described with reference to FIG. 3.

3 is a diagram illustrating an example of a file level sequential prefetching operation in a disk array structure.

As mentioned above, even if a file is not aligned and fragmented in strips, typical file level sequential prefetching has a loss of independence. In one embodiment of the file level sequential prefetching shown in Figure 3, any file is not fragmented and consists of 20 contiguous blocks. The file is perfectly aligned on five strips (strips 4 to 8) and sequentially stored in the physical blocks B16 to B35. In this configuration, an example of a conventional sequential prefetching operation that reads sequentially from the first block B16 to the last block B35 of the file is shown.

The structure of the disk array used in FIG. 3 is the same as that of FIG. In typical sequential prefetching, the size of prefetching increases linearly or exponentially from any minimum value to any maximum value. In FIG. 3, the size of prefetching is exponential from 2 ⁰ (= 1) to 2 ² (= 4). Increase by enemy.

The operation of sequential prefetching will be described with reference to FIG. 3. First, since the size of prefetching increases exponentially, the size of the first prefetching is 2 ⁰ = 1. Therefore, B16, the first block of the file, is required for disk prefetching, and then B17 and B18 are required for the second disk prefetching, since 2 ¹ = 2. Since the size of the third prefetching is 2 ² = 4, disc prefetching is required to include four prefetching blocks B19 to B22. However, the four prefetching blocks B19 to B22 span two strips (strip 4 and strip 5), so the third prefetching is the disk commands for disk 0 (1) and disk 1 (2). Are generated separately, resulting in loss of independence. Subsequent prefetching blocks, that is, prefetching for B23 to B26, prefetching for B27 to B30, and prefetching for B31 to B34 are also not aligned in the strip. As a result, two disk commands are generated when one prefetching request is performed. Thus, performance is degraded.

The present invention is a strip-aligned sequential prefetching that solves the loss of independence of existing sequential prefetching. This provides an additional function of sequential prefetching, allowing each prefetching to belong to only one strip. The sequential prefetching method according to the present invention is hereinafter referred to as strip-aligned sequential prefetching.

4 is a view showing an example of a strip-aligned sequential prefetching method according to the present invention, which is an example applied in the same state as in FIG.

Referring to FIG. 4, first, the size of the first prefetching is 2 ⁰ = 1, so that the first block of the file, B16, is required for disk prefetching (unified with FIG. 2), and then 2 ¹ = 2, so that B17 and B18 is required for the second disk prefetching. Since the size of the third prefetching is 2 ² = 4, disc prefetching is required to include four prefetching blocks B19 to B22. However, the four prefetching blocks B19 to B22 span two strips (strip 4 and strip 5), so the third prefetching is the disk commands for disk 0 (1) and disk 1 (2). Are generated separately, resulting in loss of independence.

In this case, the present invention reduces the prefetching size to satisfy the condition that all the blocks belonging to the disk prefetching request are included in one strip in consideration of the positions of the physical blocks. The largest value among the reduced prefetching sizes while satisfying this condition becomes the final prefetching size. Therefore, upon receiving the third disk prefetching request, the size of the prefetching must be reduced to satisfy the condition that the prefetching blocks belong to the strip 4 to which its start block B19 belongs. In this case, the reduced maximum prefetching size that satisfies this condition is 1. Thus, the disk command reduced to one block B19 is generated. This is to prevent two disk commands because the strip (strip 5) to which physical block B20 belongs is different from the strip (strip 4) to which block B19 belongs.

After the third prefetching, by setting the size of the prefetching block to 2 ² = 4, the fourth disk prefetching request is aligned to the strip, thus eliminating the need to generate two disk commands.

In other words, the conventional sequential prefetching shown in FIG. 3 requires 11 disk requests in order to read the prefetching file sequentially, whereas the strip-aligned sequential prefetching shown in FIG. 4 requires 7 disk requests. Able to know. Therefore, it can be seen that the strip-aligned sequential prefetching of the present invention does not generate unnecessary disk commands, thereby improving performance over conventional sequential prefetching.

As described above, the strip-aligned sequential prefetching method of the present invention is to adjust the prefetching size so that the disc prefetching request belongs to only one strip, which is the first to identify and solve the problem of loss of independence in sequential prefetching. It is a way. Next, the process of determining the start and end of the block to be prefetched by the prefetcher 55 will be described.

FIG. 5 is a flowchart illustrating a strip-aligned sequential prefetching process according to an exemplary embodiment of the present invention, and illustrates an operation of the prefetcher 55.

Referring to FIG. 5, when a disk prefetching request is input for an arbitrary file, the prefetcher 55 determines a start block and an end block of file level blocks to be prefetched as in step 201. The starting block number is assigned to the variable 'S' and the ending block number is assigned to the variable 'E'. The size of the first prefetching is then set to 'E-S + 1'. Thereafter, in step 203, 'S + 1' is substituted for an arbitrary variable 'P'.

Then, the number of the strip where the S-th block of the file is stored is substituted into the variable 'stripstart'. The strip number may be determined by taking only an integer value of 'physical block numbers in which corresponding data is stored' divided by 'number of blocks per strip'. However, the method of calculating the strip number may vary depending on the arrangement of data in the disk array.

Thereafter, as in step 207, the process of checking whether the number of the strip where the P-th block of the file is stored is equal to the variable 'stripstart' and 'P' is less than or equal to 'E' is performed.

If the condition in step 207 is correct, as in step 209, 'P' is increased by 1, and step 207 is performed. On the other hand, if the conditions in step 207 are not met, the process proceeds to step 211, where the end of the block of the file level to be prefetched is set to 'P-1', and the size of the set prefetching is set to 'P-S'. . By performing this process, the size of the prefetching is determined to belong to only one strip for one disc prefetching request.

Referring to FIG. 4, the prefetching process of the present invention is described using the third disk prefetching request as an example. Are assigned to '3' and '6' respectively. In step 203, 'P' is substituted for 'S + 1', that is, '4'. In step 205, since the 'physical block number in which the 3 (= S) th block of the file is stored' is '19' and the 'number of blocks per strip' is '4', the 'physical block in which the corresponding data is stored' is stored. The number 'divided by' number of blocks per strip 'is' 4', so '4' is substituted for 'stripstart'.

Thereafter, in step 207, 'stripstart' is '4', but the number of strips in which the P (= 4) th block is stored is not the same as '5', so the conditional expression is false. Accordingly, as in step 211, the end of the file level blocks to be prefetched is determined as 'P-1' or '3'. Therefore, the P-1 (= 3) th blocks in S (= 3) are adjusted to be the target of the final prefetching.

In the subsequent disk prefetching, in step 201, '4' and '7', respectively, are assigned the block number of the file at the beginning 'S' and the end 'E' of the file level blocks. In step 211, 'P' is '8', so that the end of the file level blocks to be prefetched is 7. This results in file-level blocks to be prefetched from 4 (= S) to 7 (= P-1).

In this case, since the disk command is generated such that the number of blocks requested is included in one strip even after the fifth disk prefetching request, the load during prefetching can be reduced.

 Meanwhile, in the detailed description of the present invention, specific embodiments have been described. However, various modifications may be made without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited by the described embodiments, but should be determined by the scope of the appended claims and equivalents thereof.

1 shows an example of a RAID structure having a plurality of disks;

2 illustrates an example of a general sequential prefetching method.

3 illustrates an example of a file-level sequential prefetching operation common in a disk array structure.

4 illustrates an example of a strip-aligned sequential prefetching method of the present invention.

5 is a flowchart illustrating a strip-aligned sequential prefetching process according to a preferred embodiment of the present invention.

Claims (4)

In the sequential prefetching method in a computer storage device using a plurality of disks, A first process of receiving a read command from the process, A second process of determining an initial prefetching size indicated by the received read command and successive blocks to be prefetched; A third process of checking whether the consecutive blocks are included in different strips; If the consecutive blocks are included in different strips, a fourth process of reducing the initial prefetching size by the number of blocks included in the strip to which a start block belongs among the consecutive blocks; A fifth process of generating one disk command according to the reduced initial prefetching size Sequential prefetching method characterized in that it comprises a. The method of claim 1, wherein the fourth process, A file block number setting process of setting a file block number (S) of the start block and a file block number (E) of an end block in the consecutive blocks; An assignment process of setting a variable block number (P) and adding 1 to S and assigning to P; A strip number search process of searching for strip numbers of strips to which the start block belongs; A checking process of checking whether the found strip number is the same as the strip number to which P belongs, and if P is less than or equal to E; If the condition of the verification process is false, the determination process of setting E to 'P-1' and determining the initial prefetching size to 'P-S' Sequential prefetching method characterized in that it comprises a. The method of claim 2, And if the condition of the verification process is true, increasing the 'P' by '1' and repeating the verification process. The method of claim 3, wherein the retrieval of the strip number is The sequential prefetching method of claim 1, wherein the physical block number corresponding to S is searched by an integer value obtained by dividing the number of blocks constituting one strip by the number of blocks constituting one strip.

Images