JPS59220853A - Disc cache system - Google Patents

Abstract

PURPOSE:To make the data base file access high-speed by adopting setting of cache buffer area security reference and a cache buffer use mode in the disc cache system. CONSTITUTION:A user task 11 generates information by the access of a data base file 3 and issues a processing request to an operating system OS16. In actual disc input and output, access information of the user and information of control block FCB peculiar to the file are transmitted to a file server 2. The file server 2 recognizes a block corresponding to the transmitted basic input/output request and retrieves a cache management table 18; and if this retrieval results in hitting, physical input/output is not performed, and data is transferred between a cache area 19 and a user buffer 20 indicated by a data base/file managing routine 17. If the retrieval does not result in hitting, the area is secured in the cache area 19, and physical input/output is performed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a disk cache 7 stem that includes a cache buffer and performs input/output to/from a disk device via the cache buffer.

[Technical background of the invention and its problems]

The capacity of magnetic disk devices is steadily increasing. The areal recording density of magnetic disks has improved at a rate of about 10 times over the past 10 years. In contrast, the access speed to magnetic disks has hardly been improved. It has long been known that the bottleneck is the access speed to the magnetic disk, which plays a central role in storing files, and that the processing performance of the entire system cannot be improved.

The access speeds between main memory and magnetic disks tend to differ by orders of magnitude between 10 and 510, which is known as the access gap.

Therefore, attempts have been made to provide a buffer made of semiconductor memory between the main memory and the magnetic disk to improve the actual access time to the magnetic disk. This is a mechanism called a disc cage.

By the way, conventionally, in basic disk access supported by an operating system (hereinafter simply referred to as OS), buffering within a user task and the above-mentioned disk cache mechanism exist as means for speeding up the access.

The former is done by software, and the latter is done by software and hardware (on the disk side).

However, in the former, there are restrictions on the handling of shared files, etc. in a multitasking environment, and in the latter, control is performed regardless of the disk access mode of the user task, and the characteristics of file access are not fully understood. It was not suitable for shared files and large-capacity database files, as it sometimes performed excessive control. For example, data management in the cache buffer is performed regardless of the file block input/output size, which reduces the efficiency of cache memory usage.

In addition, blocks that are frequently accessed, such as the index part of an index file, can be processed using the LRU algorithm (Least), which is physically and temporally simple in a multi-user environment.
Recently Used: Key? One of the ways to create space in the bank. The most recently referenced block is selected to be evicted from the cache to free up space.

This is based on the assumption that this is because it is the least likely to be used. LRU is said to be easy to implement and extremely efficient. ), the probability of staying in the Cassiniva couch was low, which ultimately led to a decline in disk cache usage efficiency.

[Purpose of the invention]

The present invention has been made in view of the above-mentioned drawbacks, and is an improved disk cache that takes into account file access characteristics so that it can be sufficiently accelerated even in multitasking and file sharing environments that were not possible with conventional systems. This system realizes a control method and its device, and takes into account the nonuniformity of each file block length and the access characteristics of the index part of the index file, and aims to speed up database file access in a multitasking environment. The purpose is to provide a file system oriented disk cache system.

[Summary of the invention]

It is well known that the emphasis of cache buffer control in a disk cache system is on the criteria for securing a cache area and improving the hit rate. The present invention aims to realize this by employing (1) and (2) shown below in this control method.

(1) Setting criteria for securing cache buffer area for variable-length 7-aisle blocks due to the presence of various types of files.

<2) Setting of cache-buffer usage mode trusted in 7-r file access mode.

Also, by converting the entire cache buffer area into a logical space,
By making the cache buffer space dynamic and dynamically expanding and contracting the cache block buffer management table, there is no need to determine the table size a priori, and the cache buffer area can be used effectively.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a disc carriage according to the present invention.

FIG. 1 is a block diagram showing the configuration of main parts of the system.

In the figure, 11 is a user task, 12 is a task control block (TCB) that manages the operation of the task 1, 13 is a file access control block (FAB) that shows the file access status etc. for each user, and J4 is a file access control block (FAB). attribute,
A file-specific control block (FOR) indicating the area on the disk, etc. 15 is a work area (user record) where the user processes data records with a program
It is.

When accessing the database file 3, the user task 11 uses the above TCB J 2 and FAB 13°F.
Generates information such as CB 14 and work area 15, and 081
Issue a processing request to 6. The database management routine 17 (system program) that receives the request from 0816 interprets the request and breaks it down into basic file block accesses (sequences). At this time, when performing actual disk input/output, information such as whether the user's file access is permanent, read-only or update, etc.
CB information is transmitted to the file server 2.

Although the cache buffer area 19 is assumed to be secured on the main storage device J in the embodiment of the present invention, it may be located in the RAM area built into the file server 2. Here, in order to use the main memory effectively and dynamically, the table 18 for managing the cache buffer area and the entity 19 of the cache buffer area are separated. The file server 2 recognizes which block of which file the transmitted basic I/O request corresponds to, searches the file block management table (FBET to be described later) in the cache management table 18, and searches the file block management table (FBET described later) in the cache management table 18. When a hit shift occurs, data is transferred between the cache buffer area and the user buffer 20 indicated by the database/file management routine 17 when physical input/output is performed. When there is no hit, the area is secured in the cache buffer area 19 (in some cases, the area is directly connected to the user buffer 20 without using the cache buffer area 19),
Performs physical input/output.

As mentioned above, the emphasis of cache buffer control is on the criteria for securing this cache buffer area and on improving the hit rate. The present invention aims to achieve this through ``a new standard for securing a cache buffer area for variable-length file blocks due to the presence of various files'' and ``setting up a cache usage mode that depends on the file access type.'' In addition, by converting the entire cache buffer area into a logical space, the cache buffer space becomes dynamic, and the cache block management table dynamically expands and contracts, eliminating the need for a priori table size determination and making effective use of the cache area. be.

First, in order to convert the cache area 19 into a logical space, the second
Cache buffer area table (CRT) shown in the figure
Prepare. Each entry constituting the table (CRT) consists of α, β, and γ, where α is the segment number in which the cache buffer exists, β is the start address of the cache buffer, and γ is the final address. NOT USE is an unused area. As a result, the memory segment, start address, and end address are defined in the continuous memory space of mj. Thereafter, when accessing the cache server 19, the file server 2 uses the set (m, n) of the cache area number m and the cache block number n secured in the buffer area, which will be described later.

Each cache and sea buffer area 1 defined in this CB-T
9 contains the block buffer control table (BBET) from the first entry (IJ), and the actual block buffer data area in the reverse direction from the last entry (IJ), each corresponding to the basic output (block input/output) of the file. is secured. FIG. 3 shows this pattern (data structure within the cache buffer area 19).

Each BBET entry is a correspondence table between file block numbers and cache buffer blocks.
It is designed to link each file individually. Inside this there is a special entry indicating the cache blockage area, which is also linked. In the figure, a is the number of the block, b is the cache buffer data start address of the block, C is the cache block link field for each file (NRU pointer L, d is the flags indicating the usage status of the cache block, and e is the cache block link field for each file. It is assumed that the number of tasks referencing a cache block, f indicates the disk number where the block exists, and g indicates the disk address where the block exists.

On the other hand, in order to reflect the access characteristics of files, an LRU block reuse algorithm for each file is used. The structure of the file block management table (FBET) used for this purpose is shown in FIG. The table (F
BET) includes the file identification number, block length of the file (assigned in units of 256 bytes), usage status (0
PEN, etc.) and the cache block buffer LRU link information (m
, H sets as link information) is placed.

The LRU block is the first block of this LRU link. In the figure, the elements that make up each entry are the number of cache blocks reserved for that file, B is the file identification ID (a file that can be uniquely identified by the system), C is the block sector length of the file, D indicates 7 lags indicating the usage status of the file, and E indicates a block link field (LRU chain) in which cache blocks secured in the file are arranged in order of oldest usage time. The entry for the first file 0 is used to manage unused areas in the cache area. (Entering free space block link chain).

The operation of the present invention will be explained in detail below. File server 2
When receiving a basic input/output request, it recognizes which block of which file it is for, and first searches for FBET (FIG. 4) based on the file identification name. If the file to be used is not found as a result of this table search, please use FB.
Generate one file block control record on the ET.

During generation, the cache area of a file that is not used at all (not opened by any task) in the record area is released, and the file block control record area that is no longer used is used. If there is no such unused area or unnecessary record area, the basic output is normally completed. However, if there is a cache area bypass specification, even in such a case, the waiting status) (WAI
Perform direct input/output to the specified buffer size area without entering T).

On the other hand, if a file to be used exists on FBET, 1 the LRU link is searched by block number. LRU
The link ends when (X'FF', X'FF') is found. When the target block number is found in this LRU link, the BBET
The address of the cache buffer block is found from the entry number (FIG. 3), and input/output is performed between the cache buffer and the buffer specified by O8. This can reduce physical input/output. This operation differs slightly depending on input (READ) and output (WRITE). This pattern is shown as a flowchart in FIG. 5 and FIG. 6, respectively. The hit block is placed in the MRU (Most Recently Used) at the leading edge of the LRU link.
d). If the target block does not exist in the LRU link, a cache buffer is secured based on the following criteria.

(1) If the number of cache block buffers exceeds the number of cache block buffers for the same file (the maximum number of cache block buffers that can be secured at the same time), the LRU buffer of that file is used.

(2) If (1) is not possible, secure a buffer area for the target file block Nagai from the cache free area.

(3) If (2) is not possible, the LRU buffer is extracted from the file with the smallest block length among other files having a block length of N-size or more and used. (
(The rest becomes free space) (4) If (3) is not possible, reduce the LRU buffer of files less than the same size by 1.
They are released one by one, merged, and cut out for use.

(5) If (4) is not possible, use a general buffer specified by the program (O8 or user program).

Similarly, in relation to this standard, for frequently accessed items such as the root index block of an in-disk file, a non-resident mode flag is set in the cache buffer block by requesting access to the file server 2 in cache non-resident mode. It operates in such a way that it is subject to the LRU extraction mentioned above and is treated as ■■.

In order to further speed up the operation, MOVE mode (
transfer) and LOCATE (location) modes have been established,
In the OCATE mode, only the address of the corresponding cache buffer block is returned and no data transfer is performed. The output behavior is the same, but for data integrity,
It only works in % rights mode. Furthermore, there are some cases, such as sequential read of a sequential file, which are read once and are hardly ever reused, and there are cases where it is meaningless to use the cache buffer area. this,
Because of various types of access, a mode (cache bypass mode) is set in which a cache buffer block is not newly secured when there is no hit, thereby restricting the use of unnecessary cache areas. On the other hand, during output operations, writing is performed only to intermediate files such as temporary files and cache buffers for high-speed output, and physical writing to actual disk files is performed when they are extracted as LRUs. A mode is set to perform the checkpoint asynchronously, such as at an arbitrary point in time when the file is closed.

The above operation is shown as block-n-row in FIGS. 5 and 6.

〔Effect of the invention〕

As explained above, according to the present invention, the following effects can be obtained.

(1) The hit rate can be greatly improved by specifying the cache buffer usage mode based on file access characteristics.

(2) The cache area is effectively used by using the LRU algorithm for each file and by setting the non-resident mode.

(3) The cache-buffer block can accommodate variable block lengths, making better use of the cache area and improving the hit rate.

(4) Even under multitask access, the cache area usage efficiency does not decrease, and the hit rate improves, especially for index files, shared files, etc., and a significant increase in speed is achieved.

(5) Cache areas can be dynamically defined and management within them is performed automatically, achieving faster speeds than conventional cache control.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the disk key 51-touch system in the present invention, and FIG. 2 is a cache/cache area definition table used in the present invention.
3 is a diagram showing the internal data structure of the cache buffer used in the present invention, and FIG. 4 is a diagram showing the table format of the file and cache before use table (FBET). The diagrams shown in FIGS. 5 and 6 are schematic blocks 70-- showing the operation of the present invention during input and output, respectively. 1...Main storage device, 2...File nanos, 3
...Large capacity file device (magnetic disk drive (Office)).Applicant's agent Patent attorney Takehiko Suzue Figure 2 Figure 3 Figure 4 Figure 5

Claims (1)

[Claims] (1) In a disk cache system that includes a cache buffer and performs input/output control with the disk via the cache buffer, a first table in which a cache buffer area is defined, a file block, and the first table a second table that manages the usage status of defined cache blocks; a third table that shows the correspondence between the file blocks and the cache blocks; The first, second,
means for accessing a cache buffer according to the contents of a third table; and means for cutting out a designated block Nagai area from the cache buffer area and caching it based on the contents of the first, second, and third tables and a predetermined standard. A means for allocating buffers and an LRU (Least Recently) for maintaining the reference order of the cache blocks.
1. A disk cache system comprising a control means.