KR100470914B1 - Method and apparatus for controlling a disk cache in a disk array system - Google Patents

Abstract

The present invention relates to a disk cache management apparatus and management method in a disk array system. The disk cache management structure is largely composed of a hash table, a stripe tag, and a cache tag. The hash table consists of an array of hash entries, and each hash entry consists of a double pointer that can link stripe tags into a double linked list. The stripe tack is allocated on a stripe basis, and is composed of a double pointer that connects stripe related information and corresponding cache tacks managing data or parity belonging to the stripe when it is stored in the cache. The cache tag consists of information necessary for data management and a memory address where data is stored. The hash function takes as input the number of stripes to which a particular block belongs and calculates the number of the corresponding hash entry. When the hash entry is determined, a corresponding stripe tag is determined by comparing the stripe numbers of stripe tags sequentially connected from the beginning with the double pointer using a double pointer. Then, the cache tag is determined by sequentially comparing the block numbers of the cache tags constituting the stripe connected to the cache tag lists of the stripe tag with the block number to be searched. Whether a particular block is cached is determined by referring to the state values of blocks constituting the cache line managed by the determined cache tag. That is, in the disk recovery mode, when the cache is searched to determine whether to cache the remaining blocks belonging to the stripe except for the required block, the stripe tag is determined through the cache search of the error data. These searches can be determined directly from the stripe tag. Therefore, the cache search process in the recovery mode is simpler than the structure using the single tag. In addition, since the cache tag that manages parity is managed by the same stripe tag instead of separately, data can be efficiently recovered. In addition, a lock in a stripe unit for interrupting processing of a general input / output request for a stripe during recovery may be easily solved by directly processing the stripe tag in a process of determining a stripe tag without a separate data structure or process. .

Description

Disk cache management device and management method in disk array system {METHOD AND APPARATUS FOR CONTROLLING A DISK CACHE IN A DISK ARRAY SYSTEM}

The present invention relates to a disk cache management technology in a disk array system, in particular, Redundant Array of Inexpensive Disks (RAID) Level 5 systems, and more particularly, in consideration of cache retrieval and operating characteristics of a RAID level 5 system. Disc cache management apparatus and method using a tag.

Generally, a RAID level 5 system is a disk array system that configures and manages multiple physical disks to behave like a single large disk. The RAID level 5 system distributes and stores data in multiple disks, and efficiently uses parity to store data. It refers to the system to protect.

In a RAID level 5 system, data is divided into a certain size and the data are distributed and stored in a 'round-robin' manner to each of the physical disks constituting one logical disk. I / O requests in an application environment affect system performance depending on the size of data being processed. In other words, I / O requests that process large amounts of data can be transferred from multiple disks at the same time, resulting in a shorter response time compared to processing on a single disk. Because each disk can handle I / O requests independently, it can handle many requests simultaneously.

In this RAID level 5 system, parity is provided to protect data on a regular basis in order to reduce the possibility of disk failure due to the use of multiple disks. The parity must be updated whenever the data stored on the disk is updated. The update requires a lot of disk accesses, which causes significant performance degradation.

To solve this problem, most RAID level 5 systems use a disk cache to store data and parity. 1 is a configuration block diagram of a RAID level 5 system providing such a disk cache.

FIG. 2 is a diagram illustrating a data and parity storage method in a general RAID level 5 system.

As shown in FIG. 2, in a typical RAID level 5 system, a total of five disks are used, and four data and one parity are distributed and stored on five disks. stripe). Striping is a unit that is processed when a disk constituting the system fails or a new disk is added to the system to increase disk capacity.

In the degraded mode due to a disk failure, when data or parity stored in the failed disk is requested, the remaining data and parity belonging to the stripe are read, and the data stored in the failed disk is recovered and processed. do. In the expanded mode for expanding a disk, an new disk including an added disk is formed, all data belonging to the newly configured stripe is read from the disk, a new parity is calculated, and the stripe is distributed and stored as a whole disk.

3 is a block diagram of a conventional disk cache management apparatus connected with a hash table and cache headers.

As shown in FIG. 3, a disk cache used in a typical RAID level 5 system manages stored data in units of cache lines, and one cache line is defined as a stripe size or a data size distributed and stored for each disk. .

Each cache line is managed by a cache tag 102. Hash table 100 is composed of a double pointer that can connect each cache tag 102, the cache tag 102 is connected to a double linked list.

In this structure, whether to store a specific block is determined through a cache search. First, by inputting a block address required for a hash function, an array index for finding a corresponding header pointer in the hash table 100 is calculated. When the header pointer is determined by the array index, the cache tag 102 connected to the header pointer determines whether to cache by comparing the block address stored in the cache pointer with the first block address of the cache line to which the requested block address belongs.

This single cache managed disk cache structure has the following problems in terms of cache retrieval and data management in a RAID level 5 system.

First, in terms of cache retrieval, assuming that the size of the hash table and the cache line size are fixed, as the size of the disk cache increases, the number of cache tags that must be managed by one header pointer increases, so that cache tags that are not frequently referenced In this case, the cache search time may be long. However, if you increase the size of the cache line in consideration of the number of cache tags, the cache will fill up quickly, and as a result, cache data replacement will frequently occur, which removes existing data from the cache for data that is newly stored in the cache. Rather, cache performance may be degraded.

In terms of data management, most RAID level 5 systems offer the ability to expand disk capacity by adding new disks or restoring stored data when a disk fails on-line or off-line. This is because the data is processed on a stripe basis, so if the size of the cache line is the same as the stripe size, the cache belonging to the same stripe is smaller. Since the tags are also linked to different header pointers, you may have to search for multiple header pointers to determine whether to cache them. In addition, in the case of caching not only data but also parity, there is a problem that a separate cache search process must be accompanied to determine whether to cache parity.

The present invention has been made to solve the above-described problems of the prior art, by applying a double tag of a separate stripe tag to independently manage only the cache tag belonging to the same stripe and the cache tag for managing data and parity, It is an object of the present invention to provide a disk cache management apparatus and management method in a disk array system that maintains or shortens search time uniformly and efficiently handles disk recovery or expansion.

According to an embodiment of the present invention for achieving the above object, a disk array system having a cache tack is allocated in units of cache lines when the number of disk blocks distributed to each disk in one stripe is defined as a cache line size. In the disk cache management apparatus of the present invention, a hash table having a unique entry array composed of double pointers and cache entries belonging to the same stripe as the cache tag are managed independently for each stripe. In addition, the present invention provides a disk cache management apparatus in a disk array system having a stripe tack that connects the cache tacks to a one defined for the stripe in a dual linked list.

According to another embodiment for achieving the object of the present invention, a disk cache management method in a disk array system having a stripe tack linking the cache tacks to a single striped tack defined for the stripe in a dual link list, When a request for a specific block is requested, the first number of determining a stripe number to which the block belongs by obtaining a quotient of dividing the number of the specific block by the cache line size and applying a determined stripe number to the hash function determines the hash entry number. A second step of determining a hash entry corresponding to the hash table using the hash entry number calculated in the first step, and searching for a stripe tag having the same stripe number starting from the first stripe tag among the stripe tags. Swap with the same stripe number When the life tag is found, the third step of determining whether the operation mode of the current system is the recovery mode or the extension mode, and the third step determines whether the current stripe is locked if the recovery mode or the extension mode is used. And if the current stripe is locked, wait until the lock is released. If not, determine that the specific block is cached or not, and if it is determined that the normal mode is a normal mode, a cache tag exists. Determining whether the block exists in the cache tag, and if the block does not exist in the cache tag, searching for the next cache tag, and From the first cache tag to the last cache tag, the required specific block among the blocks managed by the cache tag is included until there are cache tags connected to the tag. In the sixth step of searching whether it is included or the result of the fifth step, if there is a matching cache tag, the process proceeds to the cache success mode. It provides a disk cache management method in a disk array system comprising a seventh step.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

Prior to the description, the disk array system according to the present embodiment may be applied to a RAID level 5 system, and the cache tag applied to the present invention defines the number of disk blocks distributed and stored to each disk in one stripe as a cache line size. It is characterized in that the allocation to the cache line unit. In addition, the stripe tack is characterized in that it manages the cache tacks that manage the data and parity constituting the stripe, and is allocated to each stripe constituting the RAID system.

The core technical gist of the present invention having these characteristics is to configure the disk cache management structure by double tag to determine whether to cache a specific block, and from this technical concept, it is easy to achieve the purpose of the present invention. .

4A is a block diagram illustrating a disk cache management apparatus in which a hash table, a stripe tag, and a cache tag are connected according to the present invention.

As shown in Fig. 4A, the hash table 100 is an array structure of entries, and an array index of each entry is used as a unique entry number.

Each hash entry consists of a double pointer that connects the first and last stripe tacks, and the management of the stripe tack 104 to be linked to each hash entry consists of hashing the unique stripe number of each stripe tack 104. Implemented by comparing / determining with hash entry number.

Here, the total number of stripe tacks is defined to be equal to the number of total stripes constituting one RAID level 5 system. The stripe tack manages the cache tack 102 as many as the number of disks constituting the stripe by connecting them in the form of a dual linked list.

FIG. 4B is a detailed view of the structure of one stripe tack 104 of FIG. 4A.

In the stripe tack 104 structure shown in FIG. 4B, the 'front pointer' is used to connect the next stripe tack in the list, and the 'rear pointer' is used to connect the previous stripe tack.

The 'stripe number' is a unique number assigned to each stripe allocated in a RAID system and used to determine a stripe tag.

The 'stripe lock state' indicates whether or not the entire stripe is locked. The stripe lock state determines whether access to data belonging to the stripe is possible.

'Cache tack count' represents the total number of cache tacks currently managed by the stripe, and the maximum value represents the maximum number of cache tacks that can be managed by the stripe.

The first cache tag pointer indicates the first cache tag of the dual linked list, and is a pointer to newly connected cache tags linked to the stripe tag.

'Last cache tag pointer' means the cache tag attached to the end of the list.

4C is a detailed view of the structure of one cache tack 102 of FIG. 4A,

In the cache tag 102 structure shown in FIG. 4C, the 'forward pointer' refers to a cache tag connected to the next of the cache tag list, and the 'rear pointer' refers to a cache tag previously connected.

'Stripe tag pointer' means a stripe tag that manages the corresponding cache tag.

The 'LRU list forward pointer' refers to a cache tag connected next to the LRU list, and the 'LRU list back pointer' refers to a cache tag previously connected. The LRU list is one of methods for selecting an optimal cache block when cache block replacement is required, and is a circular list that manages cache blocks that preferentially replace blocks that are not referenced for a long time.

The first cache block address corresponds to the address of the first block among the blocks constituting the cache line managed by the cache tag.

'Cache Line State' is the state of the entire cache line, determined by the state of each block.

'Cache block state' is defined as an array structure to indicate the state of each block constituting the cache line.

'Cache data pointer' refers to the address of the memory where the actual cache data is stored.

Hereinafter, with the configuration described above, a disk cache management process in a RAID level 5 system according to a preferred embodiment of the present invention will be described in detail with reference to the flowchart of FIG. 5.

First, when a request for a specific block is requested, the stripe number to which the block belongs belongs is determined, and the flow proceeds to step S502. The stripe number is equal to the quotient of the block number divided by the cache line size.

In step S502, the hash entry number is determined by applying the stripe number determined in step S500 to the HASH 1 function. When the total number of stripe tacks is greater than the total number of hash entries, the HASH 1 function determines the remaining values of dividing the stripe number by the total number of hash entries as the hash entry number, and vice versa. The remaining value divided by the number is determined by the hash entry number.

In step S504, the hash entry corresponding to the hash table is determined using the calculated hash entry number.

Once the hash entry is determined, the stripe tag having the same stripe number is searched starting from the first stripe tag using the 'front pointer' as shown in FIG. 4B (S506) (S508).

If the stripe tack having the same stripe number is found after the step S506 and the step S508, the flow proceeds to step S510 to determine whether the operation mode of the current system is the recovery mode or the extension mode or the normal mode.

As a result of the determination in step S510, in the recovery mode or the extension mode, the process proceeds to step S512 to determine whether the stripe is in the locked state.

That is, since the lock is not allocated to the entire stripe in the normal input / output processing mode, once the stripe tag is determined, as described below, the connected cache tags are searched to determine whether to cache a specific block. On the other hand, in the recovery mode or the extended mode, the stripe lock is checked once because the stripe lock is required due to data recovery or stripe reconstruction.

As a result of the determination in step S512, if the stripe is in the locked state, the process proceeds to step S514 and waits for the lock to be released. Otherwise, the process proceeds to determining whether to cache a specific block as in the normal mode. .

On the other hand, if the determination result in the step S510, if it is determined that the normal mode, go to step S516 to determine whether there is a cache tag, if there is a cache tag proceeds to step (S520) and the block to the cache tag Determine if it exists.

As a result of the determination in step S520, if the corresponding block does not exist in the cache tag, the next cache tag is searched (S524), and if the number of cache tags connected to the corresponding tag is 0 (S526), that is, there are no connected cache tags. If not, the process proceeds to the cache failure step (S528).

If, as a result of the determination in step S526, if the number of cache tags connected to the corresponding tag is not 0, the required cache among the blocks managed by the cache tag from the first cache tag to the last using the 'front pointer' of the cache tag is requested. Search whether the specific block is included (S516) (S520) (S526). This is determined by the state value of each block stored in the cache tag.

If there is a matching cache tag, it is determined that the specific block is stored in the cache and proceeds to the cache success step (S522). However, if a matching tag is not found until the end, the specific block does not exist in the cache. S528).

That is, this process is implemented to perform a recovery or extension mode after performing a stripe lock when there is no request signal processed for the stripe when processing in stripe units in the recovery or extension mode. In such a recovery or extended mode, a lock on a stripe tack is potentially locked to all cache tags belonging to the stripe, so a separate cache tag is not needed.

delete

According to the present invention, in a RAID level 5 system having a characteristic of managing data on a stripe basis, each cache tag belonging to the same stripe as the cache tag used to divide and manage cache data in a distributed size for each disk is managed independently for each stripe. Therefore, there is almost no change in cache search time due to the change in the size of the disk cache. In addition, the present invention does not require unnecessary cache retrieval because only the stripe tack is determined immediately in the disk recovery mode or the extended mode, thereby eliminating the need for unnecessary cache retrieval. There is an effect that the recovery or expansion can be more efficiently handled. [0075] Although the present invention has been described in detail based on the embodiments, the present invention is not limited to these embodiments, but the scope of the present invention is not limited to the above. Of course, many variations are possible.

1 is a structural diagram of a conventional RAID system providing a disk cache,

2 is a view for explaining how data and parity are stored in a disk at RAID level 5;

3 is a block diagram of a conventional disk cache management apparatus connected with a hash table and cache headers;

4A is a block diagram of a disk cache management apparatus in which a hash table, a stripe tag, and a cache tag are connected according to the present invention;

4B is a detailed view of one stripe tack structure of FIG. 4A;

4C is a detailed view of one cache tack structure of FIG. 4A;

5 is a flowchart illustrating a disk cache management process in a disk array system, in particular, a RAID level 5 system, in accordance with a preferred embodiment of the present invention.

<Description of the code | symbol about the principal part of drawing>

100: hash table

102: Cash Tag

104: stripe tag

Claims (9)

delete In the disk array system having a cache tag allocated in cache lines when the number of disk blocks distributed to each disk in one stripe is defined as a cache line size. In the disk cache management apparatus, A hash table with a unique array of entries consisting of double pointers, A stripe tag that is connected to each entry in the hash table and manages cache tags belonging to the same stripe as the cache tag independently for each stripe, and connects the cache tags to one defined for the stripe in a double linked list. Disk cache management apparatus in the disk array system having a. The method of claim 2, The stripe tack is, A forward point region connecting the next stripe tack in the double linked list, A rear point area connecting the previous stripe tack, A stripe number area for determining the stripe tag; A stripe lock state area indicating whether or not the entire stripe has been locked; A cache tag number area indicating the total number of cache tags managed by the current stripe; A first cache tag pointer area indicating a first cache tag of the dual linked list, Last cache tag pointer area indicating a cache tag connected at the end of the dual linked list. Disk cache management apparatus in the disk array system comprising a. The method of claim 2, The disk array system is a disk cache management apparatus in a disk array system, characterized in that the RAID (Redundant Array of Inexpensive Disks) level 5 system. delete A disk cache management method in a disk array system having a stripe tag that connects cache tags to a single striped tag defined for the stripe in a dual linked list, When a request for a specific block is requested, the first number of determining a stripe number to which the block belongs by obtaining a quotient of dividing the number of the specific block by the cache line size and applying a determined stripe number to the hash function determines the hash entry number. Steps, A second step of determining a hash entry corresponding to a hash table by using the hash entry number calculated in the first step; When a stripe tag having the same stripe number is searched starting from the first stripe tag among the stripe tags and a stripe tag having the same stripe number is searched, it is determined whether the operation mode of the current system is a recovery or extended mode or a normal mode. With the third step, As a result of the determination of the third step, in the recovery mode or the extended mode, it is determined whether the current stripe is locked, and if the current stripe is locked, it waits until the lock is released; otherwise, it is determined whether to cache a specific block. With the fourth step, A fifth step of judging whether the cache tag exists if the normal mode is determined as the determination result of the third step; and determining whether the corresponding block exists in the cache tag if the cache tag exists; As a result of the determination of the fifth step, if the corresponding block does not exist in the cache tag, the next cache tag is searched and managed by the cache tag from the first cache tag to the last cache tag until there are cache tags connected to the corresponding tag. A sixth step of retrieving whether the requested specific block is included among the blocks; As a result of the search of the fifth step, if there is a matching cache tag, the process proceeds to the cache success mode; Disk cache management method in the disk array system comprising a. The method of claim 6, If the total number of stripe tacks is greater than the total number of hash entries, the hash function determines a remaining value of dividing the stripe number by the total number of hash entries as a hash entry number, and the total number of stripe tacks is greater than the total number of hash entries. And if less, the remaining value obtained by dividing the stripe number by the total number of stripe tacks is determined as the hash entry number. The method of claim 6, The search of the stripe tack or the cache tack is performed using the front pointer area in the stripe tack or the cache tack. The method according to any one of claims 6 to 8, And the disk array system is a RAID level 5 system.