KR101064178B1 - Buffer cache management system and method - Google Patents

Abstract

The present invention relates to a buffer cache management system and method, and more particularly, to a buffer cache management system and method in a system in which heterogeneous storage devices having different access speeds are connected locally or through a network. .
Another system according to an embodiment of the present invention is a buffer cache management system connected to heterogeneous storage devices, comprising: a buffer cache having a plurality of partitions; An application unit for requesting disk input / output of a block stored in the heterogeneous storage devices when an application is executed; A pattern detector for tracking a disk input / output request and detecting a reference pattern of the block indicated by the application; A virtual buffer cache that receives a result of the analysis detected by the pattern detector and determines a partition of the buffer cache in which the block is to be stored; And a block allocator for receiving the block from the heterogeneous storage devices and allocating the received block to the determined partition according to an instruction of the virtual buffer cache.

Description

Buffer cache management system and method {SYSTEM AND METHOD FOR MANAGING BUFFER CACHE}

The present invention relates to a buffer cache management system and method, and more particularly, to a buffer cache management system and method in a system in which heterogeneous storage devices having different access speeds are connected locally or through a network. .

The buffer cache generally refers to a data structure to compensate for a slower data access time by disk I / O than a data access time to memory.

The buffer cache operates by reading a portion of data recorded in permanent storage into a memory and using the same, and leaving it in memory for later reuse. And the buffer cache generally takes up most of the available memory.

Because the amount of physical memory is limited for each system, the buffer cache is also limited in size. Thus, the buffer cache frees up and reclaims memory when it is needed for other programs or for other reasons. That is, the buffer cache reuses the memory used as the cache by using a policy such as reusing the block that is expected to be used last among the plurality of blocks existing in the cache.

The more data blocks that remain in the buffer cache are used more recently, and the longer they take to fetch data, the longer they remain in memory is important for improving system responsiveness. Take part.

Block replacement algorithms used in commonly used buffer caches use a single replacement algorithm based on least recently used (LRU) based on temporal locality. As a result, the sequential reference pattern or the loop reference pattern has a very poor performance.

In addition, the single replacement policy used in the conventional buffer cache is that although the storage devices connected to the buffer cache are different from each other such as SSD, HDD, and NFS, the time required to fetch data is very different. It is assumed that the fetch costs of the storage devices are all the same.

In practice, however, if the buffer cache has the same hit ratio, having longer data from slower storage is more effective in reducing average access time. In addition, blocks showing a sequential reference pattern or a loop reference pattern are a measure of actual cache performance, which reduces the accuracy of the hit ratio. Reduce the effectiveness of the cache replacement policy.

On the other hand, the related art of the present invention is a paper related to the detection of the block access pattern of the application (A low-overhead high-performance unified buffer management scheme that exploits sequential and looping references) published in OSDI in 2000. , "Program-counter-based pattern classification in buffer caching" published in OSDI 2004, and "AMP: Program Context Specific Buffer Caching" published in USENIX ATC 2005.

The three papers see that the single replacement algorithm used in the buffer cache does not find and respond to reference regularity in block access. In particular, a sequential reference pattern and a loop reference pattern that detect such a tendency are detected, and blocks belonging to the same are managed as separate cache partitions. In addition, when replacing a block due to a capacity limitation of a cache, a marginal gain that occurs when a block is replaced in each cache partition separated according to each access pattern. (marginal gain) value is compared to take a method of replacing a block having the smallest change in marginal gain (marginal gain). Here, each article shows a difference in the way of detecting the pattern.

Conventional techniques related to the partitioned use of buffer caches by device include “Storage-Aware Caching: Revisiting Caching for heterogeneous Storage Systems” published in USENIX FAST in 2002, IEICE Transactions of Information and Systems, vol. E91-D No. 2008. “DAC: A Device-Aware Cache Management Algorithm for Heterogeneous Mobile Storage Systems” announced in .12.

The two papers allocate one cache partition to each storage device in a case where heterogeneous storage devices are connected so that data read from the storage device is managed in the partition. In this case, each cache partition is managed using a separate single replacement policy, and when reclaiming a page that has already been used due to insufficient capacity of the cache, the page to be reused is determined according to criteria that meet each goal. do.

In particular, in the case of the 2002 paper, memory page frames are allocated from the cache partition of the shortest average access time among the cache partitions allocated to each storage device to the cache partition of the longest access time. In order to reduce the average input / output time (I / O time) by performing a predetermined number of times for each block reference. In the case of the paper in 2008, it minimizes the total cost of the sum of the average fetch time of the data in each partition multiplied by the miss count in each partition. We find the distribution of and adjust the size of all partitions at regular intervals.

Since the above-described prior arts consider only the characteristics of the reference pattern of the application for the blocks read into the buffer cache or only the difference in access time between the storage devices, the single replacement policy used in the buffer cache provides an efficient buffer cache. Can't manage

The present invention provides a buffer cache management system and method coupled to heterogeneous storage devices.

In addition, the present invention provides a method of allocating and reusing a block to a buffer cache.

In addition, the present invention provides a buffer cache management system and method that can reduce the average access time of a storage device by using a cache replacement policy in a direction in which the patch cost is reduced in consideration of different fetch costs of heterogeneous storage devices. To provide.

Another system according to an embodiment of the present invention is a buffer cache management system connected to heterogeneous storage devices, the system comprising: a buffer cache having a plurality of partitions; An application unit for requesting disk input / output of a block stored in the heterogeneous storage devices when an application is executed; A pattern detector for tracking a disk input / output request and detecting a reference pattern of the block indicated by the application; A virtual buffer cache that receives a result of the analysis detected by the pattern detector and determines a partition of the buffer cache in which the block is to be stored; And a block allocator for receiving the block from the heterogeneous storage devices and allocating the received block to the determined partition according to an instruction of the virtual buffer cache.

Here, the pattern detector preferably detects a reference pattern of the block according to an application / file level pattern detection rule.

The management apparatus may further include a management tool for storing a user-defined rule for determining a reference pattern of the block. The pattern detector may access the management tool and, when the user-defined rule is stored in the management tool, select the user-defined rule. It is desirable to apply to detect the reference pattern of the block.

Here, the user-specified rule is preferably stored in the management tool in the form of a table.

Here, the management tool preferably prohibits the extraction of blocks stored in a partition of the buffer cache or limits the amount of memory in the buffer cache.

The storage monitor may further include a storage monitor that records which of the heterogeneous storage devices is read by the block allocator.

Here, the storage monitor preferably measures an average transfer time of a block read by the block allocator in the heterogeneous storage devices.

Here, the storage monitor preferably records which of the plurality of partitions of the buffer cache is allocated a block read from the heterogeneous storage devices.

Here, the plurality of partitions of the buffer cache are divided according to the reference pattern of the block, and each of the plurality of partitions preferably has a replacement policy that can increase the hit ratio of blocks belonging to it individually.

Here, the virtual buffer cache preferably determines which of the blocks stored in the buffer cache to reuse.

In this case, the virtual buffer cache is expressed as a product of a reference intensity monitoring the number of reference of each of the plurality of partitions of the buffer cache and a weighted marginal gain when determining the reuse of the block. It is preferable to use weighted cost.

According to another aspect of the present disclosure, a method of managing a buffer cache in a buffer cache management system connected to heterogeneous storage devices may include: requesting, by an application unit, a disk input / output request to a VFS when an application is executed; A pattern detector tracking the disk input / output request, detecting a reference pattern of a block indicated by the application, and delivering a detection result to a virtual buffer cache; The virtual buffer cache resizing the plurality of partitions of the buffer cache using the detection result and determining a partition in which the block is to be stored; And assigning, by the block allocator, the block read from the heterogeneous storage device to the determined partition according to the instruction of the virtual buffer cache.

Here, when the pattern detector detects a reference pattern of the block, the pattern detector further includes first checking whether there is a user-specified rule, and if there is the user-specified rule, the pattern detector according to the user-specified rule. It is desirable to detect the reference pattern of the block.

Here, if there is no user-specified rule, detecting a reference pattern of the block by applying an application / file level detection rule; If the reference pattern of the block is detected through the application / file level rule, the method may include transmitting a detection result to the virtual buffer cache.

Here, if the reference pattern of the block is not detected through the application / file level rule, the method may include detecting a reference pattern of the block by applying a block access pattern detection rule.

The virtual buffer cache may further comprise determining reuse of a block stored in a plurality of partitions of the buffer cache.

The determining of the reuse of the block may include: selecting, by the virtual buffer cache, N blocks having the lowest weight in each of the plurality of partitions; Calculating weights of all the blocks selected in each of the plurality of partitions; Comparing the weights of all the blocks to select N blocks having the lowest weights; And reusing the selected N blocks.

Using the system and method according to the present invention, there is an advantage that the input and output time can be reduced while maintaining a high hit ratio by increasing the efficiency of the buffer cache.

1 is a block diagram of a buffer cache management system 100 according to an embodiment of the present invention;
2 is a flowchart illustrating a block allocation process of the buffer cache management system illustrated in FIG. 1;
3 is a block diagram for explaining the flow of the block allocation process of FIG.
FIG. 4 is a flowchart illustrating a block reuse process of the buffer cache management system shown in FIG. 1;
5 is a block diagram illustrating the flow of a block reuse process of FIG. 4;
6 is a view for supplementing the block reuse process of FIG. 4.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a buffer cache management system 100 according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a buffer cache management system 100 according to an exemplary embodiment of the present invention may include an application unit 110, a VFS 120, a virtual buffer cache 130. , Buffer cache (140), block allocator (150), pattern detector (Pattern Monitor / Marker, 160), administration tool (Admin Tool, 170), and storage monitor (180). It may include.

The application unit 110 refers to a set of various applications that are operated in a general computer operating system including the buffer cache 140. Here, the application unit 110 includes various types of applications for performing disk I / O. For example, an application may be a media player or a web browser that plays music and video files. In addition, scientific calculation applications, database applications, UNIX system applications, and the like may be applicable.

The VFS 120 provides a standard interface to the application unit 110, which is a higher layer, to execute disk I / O.

The virtual buffer cache 130 adjusts the size of each of the plurality of partitions 141, 142, 143, and 144 in the buffer cache 140. Here, the virtual buffer cache 130 does not allocate the size of each of the plurality of partitions 141, 142, 143, and 144 at a fixed ratio, and sets the size according to a reference pattern for the corresponding block of the application. I can adjust it.

For example, when a reference pattern of a block read by the block allocator 150 is detected by the pattern detector 160, the virtual buffer cache 130 puts the block to the block allocator 150 in a predetermined partition. In addition to this, the virtual buffer cache 130 may increase the size of the designated partition. On the other hand, when a block is reused in a partition selected by a predetermined criterion, the size of the selected partition can be reduced.

In addition, when the pattern detector 160 detects that a reference pattern of a block managed by the buffer cache 140 has changed, the virtual buffer cache 130 receives the detected information and thus belongs to the corresponding block. Do the work of changing it.

In addition, the virtual buffer cache 130 may determine which block to reclaim. For this purpose, the reference intensity is calculated by monitoring how much each partition is referenced.

The buffer cache 140 divides the buffer cache memory constituting the cache to access sequential reference patterns, loop reference patterns, and other reference patterns. The partitions are divided into a plurality of partitions 141, 142, 143, and 144.

Each of partitions 141, 142, 143, and 144 of the buffer cache 140 manages the blocks classified by the pattern detector 160.

The number of partitions in the buffer cache 140 is determined by which reference patterns the buffer cache 140 classifies and manages. Here, the number of partitions is preferably at least two or more. For example, like the sequential reference pattern, the repeated reference pattern, and the temporally-clustered reference pattern, the number of partitions may be increased by subdividing according to an access pattern.

Each of partitions 141, 142, 143, and 144 has a replacement policy that can increase the hit ratio of blocks belonging to it. As one example, the partition 143 managing the sequential reference pattern and the repeating reference pattern may have an MRU replacement policy, and the partition 142 managing other reference patterns may have an LRU replacement policy or an ARC replacement policy.

The reuse of a block cached in partitions 141, 142, 143, and 144 is performed by the virtual buffer cache 130, where candidate blocks selected by the replacement algorithm in each partition are actually When eviction is made in the partition, the one with the smallest change in the estimated hit ratio is selected and performed.

The block allocator 150 stores a predetermined block newly read from the storage device 200 among the plurality of partitions 141, 142, 143, and 144 in the buffer cache 140 according to the instruction of the virtual buffer cache 130. Assign to the selected partition.

The pattern detector 160 traces information (read / write) from the application unit 110 to the VFS 120 so that a block indicated by an application to be executed is any reference pattern such as a sequential reference pattern or a repeating reference pattern. Analyze if Then, the analyzed result is reported to the virtual buffer cache 130. Here, the result analyzed by the pattern detector 160 indicates that the block indicated by the application, that is, the block read by the block allocator 150 has a reference pattern, and to which partition in the buffer cache 140 the block should be located. Say the results for.

Here, two examples of the reference pattern detection method of the pattern detector 160 may be mentioned. However, the present invention is not limited only to the reference pattern detection method described below.

The reference pattern detection method can be divided into a user-defined rule-based detection method and an application / file level pattern detection method.

The detection method based on a user-defined rule blocks a predetermined partition in the user-specified buffer cache 140 without analyzing a separate reference pattern when a user sets a predetermined rule through the management tool 170. Is to be moved. For example, if a user opens a file with an extension of "mpg", and the rule specifies that all blocks associated with that file should be regarded as a sequential reference pattern, then blocks belonging to a file with an extension of "mpg" They all judge the sequential reference pattern, and cause these blocks to be moved to the partition managing the sequential reference pattern.

On the other hand, blocks that are not defined by a user may be detected as a sequential reference pattern or a repeating reference pattern through pattern detection at an application level or file level. Here, the pattern detection at the application level is a reference pattern by tracking the logical block number of the block to be requested when the application unit 110 transmits information (read / write) from the application unit 110 to the VFS 120. Say how to detect. Meanwhile, the pattern detection at the file level refers to a method of detecting a reference pattern with respect to blocks belonging to a specific file based on an offset value of the specific file.

The management tool 170 is a user level application that creates a rule to allow a user to specify a specific user specified rule and separate the corresponding block reference pattern. If the user can clearly know the block reference pattern through the management tool 170, for example, the file specified by the user for the file extension "mpg", the buffer associated with the file with the file extension "mpg" are all buffer cached. The control may be controlled to move to a specific partition of 140. In addition, the method may further include a restriction such as prohibiting eviction in a specific partition of the buffer cache 140 or allocating an amount of memory corresponding to the specified size until the process is terminated. Custom rules may be created and managed in the form of a table, and the virtual buffer cache 130 accesses the generated table, and then the virtual buffer cache 130 is a custom rule. The block allocator 150 is referred to as a reference.

The storage monitor 180 records from which memory device 200 the blocks read by the block allocator 150 are read.

In addition, the storage monitor 180 measures an average transfer time of blocks transmitted from the storage device 200 to the block allocator 150. When the information of the average transmission time is collected, the information is transmitted to the virtual buffer cache 130. The virtual buffer cache 130 then adjusts the size of each of the partitions 141, 142, 143, and 144 in the buffer cache 140 using the transmitted information.

In addition, the storage monitor 180 records which blocks are allocated to which partitions, and transmits the recorded information to the pattern detector 160. The information transmitted in this way is utilized to detect the reference pattern of the pattern detector 160.

The memory device 200 is a permanent storage device in which data indicated by an application of the application unit 110 of the buffer cache management system 100 is recorded, and is a block allocator of the buffer cache management system 100. It is connected to the 150 through a local or network. Here, the memory device 200 may include a plurality of memory devices 201, 202, and 203. Preferably, the plurality of memory devices 201, 202, and 203 are all different memory devices. For example, the plurality of storage devices may be the HDD 201, the SSD / Flash 202, and the external storage device 203 connected through a network such as NFS or SAN. Here, the HDD 201 and the SSD / Flash 202 may be locally connected to the buffer cache management system 100, and the external storage device 203 may be connected to the buffer cache management system 100 through a network.

The access speeds of the HDD 201, the SSD 202, the external storage device 203, and the buffer cache management system 100 are different. This access speed is monitored by the storage monitor 180.

The operation of the buffer cache management system 100 according to the preferred embodiment of the present invention shown in FIG. 1 is divided into two parts. One is the block allocation process and the other is the block reuse process. Reference is made to the accompanying drawings to describe these two operations.

FIG. 2 is a flowchart illustrating a block allocation process of the buffer cache management system of FIG. 1, and FIG. 3 is a block diagram illustrating the flow of the block allocation process of FIG. 2. A block allocation process of the buffer cache management system will be described in detail with reference to FIGS. 1 to 3.

First, referring to FIGS. 1 to 3, when a predetermined application among a plurality of applications of the upper application unit 110 is executed by the user (s201), the application unit 110 may input / output a disk input / output (VFS 120) to the VFS 120. disk I / O) request (read / write) (s202).

When the step S202 is executed, the pattern detector 160 detects the reference pattern of the block (s) indicated by the application by tracing the disk I / O request transmitted to the VFS 120. In order to detect the reference pattern, the pattern detecting unit 160 first checks whether the management tool 170 has a rule predetermined by the user (S203).

If there is a rule predefined by the user (hereinafter referred to as a 'user-specified rule'), the reference pattern of the block (s) is analyzed using the user-specified rule (S204). When the analysis is completed, the pattern detector 160 transmits the analyzed result to the virtual buffer cache 130.

The virtual buffer cache 130 selects a partition in which the block (s) are to be stored among the plurality of partitions 141, 142, 143, and 144 in the buffer cache 140 using the analysis result received from the pattern detector 160. Determine (s205). Here, the virtual buffer cache 130 may adjust the size of each of the plurality of partitions 141, 142, 143, and 144.

After determining the partition, the virtual buffer cache 130 instructs the block allocator 150 to assign the block (s) to the determined partition. Then, the block allocator 150 allocates the block (s) to the determined partition according to the command of the virtual buffer cache 130 (S206).

When the block (s) are allocated to the partition by the block allocator 150, the partition to which the block (s) are allocated updates the location of the block (s) according to its management policy (s207).

On the other hand, in step s203, when there is no separate user-specified rule by the management tool 170, the pattern detector 160 detects the reference pattern of the block (s) by applying an application / file level detection rule (s208). ).

Here, it is determined whether the reference pattern of the block (s) is analyzed by the application / file level detection rule (s209). When the reference pattern of the block (s) is analyzed by an application / file level detection rule, the pattern detector 160 transmits the analyzed result to the virtual buffer cache 130 according to step s205, and the virtual buffer cache ( 130 uses the analyzed result to determine the partition of the block (s). On the other hand, when the reference pattern of the block (s) is not analyzed by an application / file level detection rule, a block access pattern rule is applied (s210) to apply the reference pattern of the block (s). Analyze (s204).

On the other hand, when the pattern detector 160 does not detect the reference pattern of the block (s) even by the above-described rules, that is, when the block (s) cannot belong to any partition, the virtual buffer cache 130 may be configured. Allow block (s) to be allocated to a specific predetermined partition. At this time, the location of the block (s) in the partition in which the block (s) will be stored follows the location update policy when the replacement algorithm being used in the partition is referenced.

4 is a flowchart illustrating a block reuse process of the buffer cache management system of FIG. 1, FIG. 5 is a block diagram illustrating the flow of the block reuse process of FIG. 4, and FIG. 6 is a block diagram of FIG. 4. A diagram for further explaining the reuse process. A block reuse process of the buffer cache management system will be described in detail with reference to FIGS. 1 and 4 to 6.

If the reuse of the block is required (s401), the virtual buffer cache 130 selects N blocks from each of all partitions 141, 142, 143, and 144 in the buffer cache 140 (s402). Here, the N blocks selected in each partition are N blocks having the lowest weighted cost sequentially in the partition. As shown in FIG. 6, when N is 4, the virtual buffer cache selects four blocks in the order of least weight in all partitions 140-a, 140-b, and 140-c, respectively. Thus, the total number of selected blocks is twelve.

After selecting N blocks in each partition, the virtual buffer cache 130 calculates a weight of each of the selected blocks (s403). Here, the equation for calculating the weight of the selected blocks may use Equation 1 below.

In Equation 1 above, (reference intensity) is a normalized value of the reference number of each partition, as shown in Equation 2 below.

In Equation 1, Weighted Marginal Gain (Weighted MG) may be calculated as in Equation 3 below.

In Equation 3, the weighted margin gain (Weighted MG) is the access time according to the storage location of the block at the marginal gain shown in the general reference pattern. This value can be obtained by multiplying the normalized value of) by the weight based on the ratio of the request (read / write).

In the present invention, the marginal gain is calculated as the rate of change of the hit ratio. In other words, the marginal gain is calculated by subtracting the hit rate when the block does not exist from the hit rate when the block is present.

Here, the general marginal gain is estimated differently according to each reference pattern. As an example, the limiting gain is 0 for sequential reference patterns and the limiting gain is 1 / pattern length for repeating reference patterns. Here, the pattern length is the number of blocks constituting the repeating reference pattern.

In addition, when the LRU strategy is not applied to the above two examples, the marginal gain may be calculated using a Belady's lifetime function as shown in Equation 4 below. In Equation 4 below, n is the size of a partition.

In Equation 4, c and k are values defined in Belady's lifetime function (Communications of the ACM, Vol. 12, Issue 5, 1969), where c is a constant given to a general task, and k is a constant. Means a constant determined near 2. C and k are determined to approximate a hit ratio curve measured for a predetermined cache size. Here, c and k corresponding to each pattern are determined.

Therefore, the weighted marginal gain is calculated as in Equation 5 below.

In Equation 5 above, (normalized fetch time from storage) measures the average time taken from the storage location of the block to the storage device 200 and normalizes it based on the speed of the shortest storage device 200. normalized). (Defined read write ratio) is advantageous to slow it down because it must flush the data in order to reuse. Therefore, since the cost of writing is higher than that of read, the ratio is specified in advance to reflect the flush cost of the dirty block in the weighted marginal gain.

When the weighted cost of each block is calculated by the above equations, the weights of the target blocks selected in each partition are compared (S404). Subsequently, N blocks having the smallest weight value are sequentially selected (s405) and reused (s406).

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

100: buffer cache management system,
110: the application unit,
120: VFS,
130: virtual buffer cache,
140: buffer cache,
150: block allocator,
160: pattern detector,
170: management tools,
180: storage monitor,
200: memory device.

Claims (17)

In a buffer cache management system connected to heterogeneous storage devices,
A buffer cache having a plurality of partitions;
An application unit for requesting disk input / output of a block stored in the heterogeneous storage devices when an application is executed;
A pattern detector for tracking a disk input / output request and detecting a reference pattern of the block indicated by the application;
A virtual buffer cache that receives a result of the analysis detected by the pattern detector and determines a partition of the buffer cache in which the block is to be stored; And
A block allocator for receiving the block from the heterogeneous storage devices and allocating the received block to the determined partition according to an instruction of the virtual buffer cache;
Buffer cache management system comprising a.
The method of claim 1,
And the pattern detector detects a reference pattern of the block according to an application / file level pattern detection rule.
The method according to claim 1 or 2,
A management tool for storing a custom rule for determining a reference pattern of the block,
And the pattern detector accesses the management tool and detects a reference pattern of the block by applying the user-defined rule when a user-defined rule is stored in the management tool.
The method of claim 3, wherein
The user-specified rule is stored in the management tool in the form of a table, the buffer cache management system.
The method of claim 3, wherein
Wherein said management tool prohibits the expulsion of blocks stored in partitions of said buffer cache or limits the amount of memory in said buffer cache.
The method of claim 1,
And a storage monitor for recording which of said heterogeneous storage devices is a block read by said block allocator.
The method according to claim 6,
The storage monitor measures the average transfer time of a block read into the block allocator in the heterogeneous storage devices.
The method according to claim 6,
And the storage monitor records which of the plurality of partitions of the buffer cache is allocated a block read from the heterogeneous storage devices.
The method of claim 1,
And a plurality of partitions of the buffer cache are divided according to a reference pattern of the block, and each of the plurality of partitions individually has a replacement policy for increasing a hit ratio of blocks belonging to the plurality of partitions.
The method of claim 1,
And the virtual buffer cache determines which of the blocks stored in the buffer cache to reuse.
The method of claim 10,
When determining the reuse of a block, the virtual buffer cache may have a weight expressed as a product of a reference intensity monitoring a reference number of each of the partitions of the buffer cache and a weighted marginal gain. Buffer management system using Weighted Cost.
In the buffer cache management method in a buffer cache management system connected to heterogeneous storage devices,
Requesting a disk input / output request to the VFS by the application unit when the application is executed;
A pattern detector tracking the disk input / output request, detecting a reference pattern of a block indicated by the application, and delivering a detection result to a virtual buffer cache;
The virtual buffer cache resizing the plurality of partitions of the buffer cache using the detection result and determining a partition in which the block is to be stored; And
Allocating a block read from the heterogeneous storage device to the determined partition according to an instruction of the virtual buffer cache;
Buffer cache management method comprising a.
The method of claim 12,
If the pattern detector detects a reference pattern of the block, the pattern detector further includes first checking whether there is a user-defined rule;
If there is the user-specified rule, detecting the reference pattern of the block according to the user-specified rule.
The method of claim 13,
If there is no user-specified rule, detecting a reference pattern of the block by applying an application / file level detection rule;
And forwarding a detection result to the virtual buffer cache when a reference pattern of the block is detected through the application / file level rule.
The method of claim 14,
If the reference pattern of the block is not detected through the application / file level rule, applying a block access pattern detection rule to detect the reference pattern of the block.
The method of claim 12,
Determining, by the virtual buffer cache, reuse of a block stored in a plurality of partitions of the buffer cache.
17. The method of claim 16,
Determining the reuse of the block,
Selecting, by the virtual buffer cache, N blocks having the lowest weight in each of the plurality of partitions;
Calculating weights of all the blocks selected in each of the plurality of partitions;
Comparing the weights of all the blocks to select N blocks having the lowest weights; And
Reusing the selected N blocks;
Buffer cache management method comprising a.

Images